Upload
dovescry
View
18
Download
2
Embed Size (px)
Citation preview
THE SKILL CONTENT OF RECENT TECHNOLOGICALCHANGE AN EMPIRICAL EXPLORATION
DAVID H AUTOR
FRANK LEVY
RICHARD J MURNANE
We apply an understanding of what computers do to study how computerizationalters job skill demands We argue that computer capital (1) substitutes for workersin performing cognitive and manual tasks that can be accomplished by followingexplicit rules and (2) complements workers in performing nonroutine problem-solving and complex communications tasks Provided that these tasks are imperfectsubstitutes our model implies measurable changes in the composition of job taskswhich we explore using representative data on task input for 1960 to 1998 We ndthat within industries occupations and education groups computerization is asso-ciated with reduced labor input of routine manual and routine cognitive tasks andincreased labor input of nonroutine cognitive tasks Translating task shifts intoeducation demand the model can explain 60 percent of the estimated relative de-mand shift favoring college labor during 1970 to 1998 Task changes within nomi-nally identical occupations account for almost half of this impact
INTRODUCTION
A wealth of quantitative and case-study evidence documentsa striking correlation between the adoption of computer-basedtechnologies and the increased use of college-educated laborwithin detailed industries within rms and across plants withinindustries1 This robust correlation is frequently interpreted asevidence of skill-biased technical change Yet as critics point outthis interpretation merely labels the correlation without explain-
We thank the Alfred P Sloan Foundation the Russell Sage Foundation andthe MIT-Ford Research Collaboration for nancial support and Kokkeong PuahDetelina Vasileva and especially Melissa S Kearny for research assistance Weare indebted to Daron Acemoglu Joshua Angrist Lex Borghans Nicole FortinEdward Glaeser Lawrence Katz Kevin Lang Thomas Lemieux Sendhil Mullain-athan Richard Nelson Kathryn Shaw Marika Tatsutani Bas ter Weel threeanonymous referees and numerous seminar participants for excellent sugges-tions We thank Randy Davis of the Massachusetts Institute of Technology Arti-cial Intelligence Laboratory and Peter Szolovits of the Massachusetts Institute ofTechnology Laboratory for Computer Science for clarifying issues in articialintelligence and Michael Handel for providing key data sources and expert adviceon use of the Dictionary of Occupational Titles
1 Berman Bound and Griliches [1994] Autor Katz and Krueger [1998]Machin and Van Reenen[1998] Berman Bound and Machin [1998 2000] and GeraGu and Lin [2001] present evidence on industry level demand shifts from the UnitedStates OECD Canada and other developed and developing countries Levy andMurnane [1996] Doms Dunne and Troske [1997] and Bresnahan BrynjolfssonandHitt [2002] provide evidence on rm and plant level shifts Katz and Autor [1999]summarize this literature Card and DiNardo [2002] offer a critique
copy 2003 by the President and Fellows of Harvard College and the Massachusetts Institute ofTechnologyThe Quarterly Journal of Economics November 2003
1279
ing its cause It fails to answer the question of what it is thatcomputers domdash or what it is that people do with computersmdashthatcauses educated workers to be relatively more in demand
This paper proposes an answer to this question We formalizeand test a simple theory of how the rapid adoption of computertechnologymdashspurred by precipitous real price declinesmdash changesthe tasks performed by workers at their jobs and ultimately thedemand for human skills Our approach builds on an intuitive setof observations offered by organizational theorists computer sci-entists and most recently economists about what computers domdashthat is the tasks they are best suited to accomplishmdashand howthese capabilities complement or substitute for human skills inworkplace settings2 The simple observations that undergird ouranalysis are (1) that computer capital substitutes for workers incarrying out a limited and well-dened set of cognitive and man-ual activities those that can be accomplished by following explicitrules (what we term ldquoroutine tasksrdquo) and (2) that computercapital complements workers in carrying out problem-solving andcomplex communication activities (ldquononroutinerdquo tasks) (See Ta-ble I for examples) Provided that routine and nonroutine tasksare imperfect substitutes these observations imply measurablechanges in the task composition of jobs which we test below
To answer the core questions of our paper the ideal experi-ment would provide two identical autarkic economies one facinga dramatic decline in the price of computing power and the othernot By contrasting these economies it would be straightforwardto assess how computerization reshapes the task composition ofwork and hence the structure of labor demand Because thisexperiment is not available we develop a simple economic modelto predict how demand for workplace tasks responds to an econ-omywide decline in the price of computer capital The model predictsthat industries and occupations that are initially intensive in laborinput of routine tasks will make relatively larger investments incomputer capital as its price declines These industries and occupa-tions will reduce labor input of routine tasks for which computer
2 Simon [1960] provides the rst treatment of this question with which we arefamiliar and his essay introduces many of the ideas explored here Other early worksinclude Drucker [1954] and Nelson and Winter [1982] Adler [1986] Orr [1996] andZuboff [1988] discuss what computers and related technology do in the workplace butdo not consider economic implications Acemoglu [1998] Goldin and Katz [1998]Bresnahan [1999] Bartel Ichniowski and Shaw [2000] Lindbeck and Snower[2000] Lang [2002] and Bresnahan Brynjolfsson and Hitt [2002] provide economicanalyses of why technology and human capital are complementary
1280 QUARTERLY JOURNAL OF ECONOMICS
capital substitutes and increase demand for nonroutine task inputwhich computer capital complements In net these forces will raiserelative demand for highly educated workers who hold comparativeadvantage in nonroutine versus routine tasks
To test these predictions we pair representative data on jobtask requirements from the Dictionary of Occupational Titles(DOT) with samples of employed workers from the Census andCurrent Population Survey to form a consistent panel of industryand occupational task input over the four-decade period from1960 to 1998 A unique virtue of this database is that it permitsus to analyze changes in task input within industries educationgroups and occupationsmdashphenomena that are normally unob-servable By measuring the tasks performed in jobs rather thanthe educational credentials of workers performing those jobs webelieve our study supplies a missing conceptual and empiricallink in the economic literature on technical change and skilldemand
Our analysis provides four main pieces of evidence support-ing our model
1) Commencing in the 1970s labor input of routine cognitiveand manual tasks in the U S economy declined and laborinput of nonroutine analytic and interactive tasks rose
2) Shifts in labor input favoring nonroutine and against rou-tine tasks were concentrated in rapidly computerizingindustries These shifts were small and insignicant inthe precomputer decade of the 1960s and accelerated ineach subsequent decade
3) The substitution away from routine and toward nonrou-tine labor input was not primarily accounted for by edu-cational upgrading rather task shifts are pervasive at alleducational levels
4) Paralleling the within-industry task shifts occupationsundergoing rapid computerization reduced input of rou-tine cognitive tasks and increased input of nonroutinecognitive tasks
We consider a number of economic and purely mechanicalalternative explanations for our results Two supply side factorsthat we study in particular are the rising educational attainmentof the workforce and the rising human capital and labor forceattachment of womenmdashboth of which could potentially generateshifts in job task composition independent of demand shifts Aswe show below the task shifts that we documentmdashand their
1281SKILL CONTENT OF TECHNICAL CHANGE
associations with the adoption of computer technologymdashare aspervasive within gender education and occupation groups asbetween indicating that these supply side forces are not theprimary explanation for our results
We begin by presenting our informal ldquotask modelrdquo describinghow computerization affects the tasks that workers and machinesperform We next formalize this model in a production frameworkto develop empirical implications for task demand at the industryand occupation level Subsequent sections describe our datasources and test the modelrsquos main implications Drawing togetherthe empirical strands we nally assess the extent to whichchanges in the task composition can account for recent demandshifts favoring more educated workers This exercise shows thatestimated task shifts are economically large underscoring thepotential of the conceptual model to reconcile key facts
I THE TASK MODEL
We begin by conceptualizing work from a ldquomachinersquos-eyerdquoview as a series of tasks to be performed such as moving anobject performing a calculation communicating a piece of infor-mation or resolving a discrepancy Our model asks which ofthese tasks can be performed by a computer A general answer isfound by examining what is arguably the rst digital computerthe Jacquard Loom of 1801 Jacquardrsquos invention was a machinefor weaving fabrics with inlaid patterns specied by a programpunched onto cards and fed into the loom Some programs werequite sophisticated one surviving example uses more than 10000cards to weave a black and white silk portrait of Jacquard him-self3 Two centuries later the electronic descendants of Jacquardrsquosloom share with it two intrinsic traits First they rapidly and accu-rately perform repetitive tasks that are deterministically speciedby stored instructions (programs) that designate unambiguouslywhat actions the machine will perform at each contingency toachieve the desired result Second computers are ldquosymbolic proces-sorsrdquo acting upon abstract representations of information such asbinary numbers or in the loomrsquos case punched cards
Spurred by a more than trillionfold decline in the real price of
3 The Jacquard loom was also the inspiration for Charles Babbagersquos analyti-cal engine and Herman Hollerithrsquos punch card reader used to process the 1910United States Census
1282 QUARTERLY JOURNAL OF ECONOMICS
computing power [Nordhaus 2001] engineers have become vastlymore procient at applying the loomrsquos basic capabilitiesmdashrapidexecution of stored instructionsmdashto a panoply of tasks How doesthis advance affect the task composition of human work Theanswer depends both upon how computers substitute for or com-plement workers in carrying out specic tasks and how thesetasks substitute for one another We illustrate these cases byconsidering the application of computers to routine and nonrou-tine cognitive and manual tasks
In our usage a task is ldquoroutinerdquo if it can be accomplished bymachines following explicit programmed rules Many manualtasks that workers used to perform such as monitoring the tem-perature of a steel nishing line or moving a windshield into placeon an assembly line t this description Because these tasksrequire methodical repetition of an unwavering procedure theycan be exhaustively specied with programmed instructions andperformed by machines
A problem that arises with many commonplace manual andcognitive tasks however is that the procedures for accomplishingthem are not well understood As Polanyi [1966] observed ldquoWecan know more than we can tell [p 4] The skill of a drivercannot be replaced by a thorough schooling in the theory of themotorcar the knowledge I have of my own body differs altogetherfrom the knowledge of its physiology and the rules of rhymingand prosody do not tell me what a poem told me without anyknowledge of its rules [p 20]rdquo We refer to tasks tting Polanyirsquosdescription as ldquononroutinerdquo that is tasks for which the rules arenot sufciently well understood to be specied in computer codeand executed by machines Navigating a car through city trafcor deciphering the scrawled handwriting on a personal checkmdashminor undertakings for most adultsmdashare not routine tasks by ourdenition (see Beamish Levy and Murnane [1999] and AutorLevy and Murnane [2002] for examples) The reason is that thesetasks require visual and motor processing capabilities that can-not at present be described in terms of a set of programmablerules [Pinker 1997]4
Our conceptual model suggests that because of its decliningcost computer-controlled machinery should have substantially
4 If a manual task is performed in a well-controlled environment howeverit can often be automated despite the seeming need for adaptive visual or manualprocessing As Simon [1960] observed environmental control is a substitute forexibility
1283SKILL CONTENT OF TECHNICAL CHANGE
substituted for workers in performing routine manual tasks Thisphenomenon is not novel Substitution of machinery for repetitivehuman labor has been a thrust of technological change through-out the Industrial Revolution [Hounshell 1985 Mokyr 1990 Gol-din and Katz 1998] By increasing the feasibility of machinesubstitution for repetitive human tasks computerization fur-thersmdashand perhaps acceleratesmdashthis long-prevailing trend
The advent of computerization also marks a qualitative en-largement in the set of tasks that machines can perform Becausecomputers can perform symbolic processingmdashstoring retrievingand acting upon informationmdashthey augment or supplant humancognition in a large set of information-processing tasks that his-torically were not amenable to mechanization Over the last threedecades computers have substituted for the calculating coordi-nating and communicating functions of bookkeepers cashierstelephone operators and other handlers of repetitive informa-tion-processing tasks [Bresnahan 1999]
This substitution marks an important reversal Previousgenerations of high technology capital sharply increased demandfor human input of routine information-processing tasks as seenin the rapid rise of the clerking occupation in the nineteenthcentury [Chandler 1977 Goldin and Katz 1995] Like these tech-nologies computerization augments demand for clerical and in-formation-processing tasks But in contrast to its nineteenth cen-tury predecessors it permits these tasks to be automated
The capability of computers to substitute for workers in carry-ing out cognitive tasks is limited however Tasks demanding exi-bility creativity generalized problem-solving and complex commu-nicationsmdashwhat we call nonroutine cognitive tasksmdashdo not (yet)lend themselves to computerization [Bresnahan 1999] At presentthe need for explicit programmed instructions appears a bindingconstraint There are very few computer-based technologies that candraw inferences from models solve novel problems or form persua-sive arguments5 In the words of computer scientist Patrick Winston
5 It is however a fallacy to assume that a computer must reproduce all of thefunctions of a human to perform a task traditionally done by workers For exampleAutomatic Teller Machines have supplanted many bank teller functions althoughthey cannot verify signatures or make polite conversation while tallying changeClosely related computer capital may substitute for the routine components ofpredominantly nonroutine tasks eg on-board computers that direct taxi cabs Whatis required for our conceptual model is that the routine and nonroutine tasks embod-ied in a job are imperfect substitutes Consequently a decline in the price of accom-plishing routine tasks does not eliminate demand for nonroutine tasks
1284 QUARTERLY JOURNAL OF ECONOMICS
[1999] ldquoThe goal of understanding intelligence from a computa-tional point of view remains elusive Reasoning programs still ex-hibit little or no common sense Todayrsquos language programs trans-late simple sentences into database queries but those languageprograms are derailed by idioms metaphors convoluted syntax orungrammatical expressionsrdquo6
The implication of our discussion is that because presentcomputer technology is more substitutable for workers in carry-ing out routine tasks than nonroutine tasks it is a relativecomplement to workers in carrying out nonroutine tasks From aproduction function standpoint outward shifts in the supply ofroutine informational inputs both in quantity and quality in-crease the marginal productivity of workers performing nonrou-tine tasks that demand these inputs For example comprehen-sive bibliographic searches increase the quality of legal researchand timely market information improves the efciency of mana-gerial decision-making More tangibly because repetitive pre-dictable tasks are readily automated computerization of theworkplace raises demand for problem-solving and communica-tions tasks such as responding to discrepancies improving pro-duction processes and coordinating and managing the activitiesof others This changing allocation of tasks was anticipated byDrucker [1954] in the 1950s ldquoThe technological changes nowoccurring will carry [the Industrial Revolution] a big step furtherThey will not make human labor superuous On the contrarythey will require tremendous numbers of highly skilled andhighly trained menmdashmanagers to think through and plan highlytrained technicians and workers to design the new tools to pro-duce them to maintain them to direct themrdquo [p 22 bracketsadded]
Table I provides examples of tasks in each cell of our two-by-two matrix of workplace tasks (routine versus nonroutine man-ual versus information processing) and states our hypothesisabout the impact of computerization for each cell The next sec-tion formalizes these ideas and derives empirical implications7
6 Software that recognizes patterns (eg neural networks) or solves prob-lems based upon inductive reasoning from well-specied models is under devel-opment But these technologies have had little role in the ldquocomputer revolutionrdquo ofthe last several decades As one example current speech recognition softwarebased on pattern recognition can recognize words and short phrases but can onlyprocess rudimentary conversational speech [Zue and Glass 2000]
7 Our focus on task shifts in the process of production within given jobsoverlooks two other potentially complementary avenues by which technical
1285SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
ing its cause It fails to answer the question of what it is thatcomputers domdash or what it is that people do with computersmdashthatcauses educated workers to be relatively more in demand
This paper proposes an answer to this question We formalizeand test a simple theory of how the rapid adoption of computertechnologymdashspurred by precipitous real price declinesmdash changesthe tasks performed by workers at their jobs and ultimately thedemand for human skills Our approach builds on an intuitive setof observations offered by organizational theorists computer sci-entists and most recently economists about what computers domdashthat is the tasks they are best suited to accomplishmdashand howthese capabilities complement or substitute for human skills inworkplace settings2 The simple observations that undergird ouranalysis are (1) that computer capital substitutes for workers incarrying out a limited and well-dened set of cognitive and man-ual activities those that can be accomplished by following explicitrules (what we term ldquoroutine tasksrdquo) and (2) that computercapital complements workers in carrying out problem-solving andcomplex communication activities (ldquononroutinerdquo tasks) (See Ta-ble I for examples) Provided that routine and nonroutine tasksare imperfect substitutes these observations imply measurablechanges in the task composition of jobs which we test below
To answer the core questions of our paper the ideal experi-ment would provide two identical autarkic economies one facinga dramatic decline in the price of computing power and the othernot By contrasting these economies it would be straightforwardto assess how computerization reshapes the task composition ofwork and hence the structure of labor demand Because thisexperiment is not available we develop a simple economic modelto predict how demand for workplace tasks responds to an econ-omywide decline in the price of computer capital The model predictsthat industries and occupations that are initially intensive in laborinput of routine tasks will make relatively larger investments incomputer capital as its price declines These industries and occupa-tions will reduce labor input of routine tasks for which computer
2 Simon [1960] provides the rst treatment of this question with which we arefamiliar and his essay introduces many of the ideas explored here Other early worksinclude Drucker [1954] and Nelson and Winter [1982] Adler [1986] Orr [1996] andZuboff [1988] discuss what computers and related technology do in the workplace butdo not consider economic implications Acemoglu [1998] Goldin and Katz [1998]Bresnahan [1999] Bartel Ichniowski and Shaw [2000] Lindbeck and Snower[2000] Lang [2002] and Bresnahan Brynjolfsson and Hitt [2002] provide economicanalyses of why technology and human capital are complementary
1280 QUARTERLY JOURNAL OF ECONOMICS
capital substitutes and increase demand for nonroutine task inputwhich computer capital complements In net these forces will raiserelative demand for highly educated workers who hold comparativeadvantage in nonroutine versus routine tasks
To test these predictions we pair representative data on jobtask requirements from the Dictionary of Occupational Titles(DOT) with samples of employed workers from the Census andCurrent Population Survey to form a consistent panel of industryand occupational task input over the four-decade period from1960 to 1998 A unique virtue of this database is that it permitsus to analyze changes in task input within industries educationgroups and occupationsmdashphenomena that are normally unob-servable By measuring the tasks performed in jobs rather thanthe educational credentials of workers performing those jobs webelieve our study supplies a missing conceptual and empiricallink in the economic literature on technical change and skilldemand
Our analysis provides four main pieces of evidence support-ing our model
1) Commencing in the 1970s labor input of routine cognitiveand manual tasks in the U S economy declined and laborinput of nonroutine analytic and interactive tasks rose
2) Shifts in labor input favoring nonroutine and against rou-tine tasks were concentrated in rapidly computerizingindustries These shifts were small and insignicant inthe precomputer decade of the 1960s and accelerated ineach subsequent decade
3) The substitution away from routine and toward nonrou-tine labor input was not primarily accounted for by edu-cational upgrading rather task shifts are pervasive at alleducational levels
4) Paralleling the within-industry task shifts occupationsundergoing rapid computerization reduced input of rou-tine cognitive tasks and increased input of nonroutinecognitive tasks
We consider a number of economic and purely mechanicalalternative explanations for our results Two supply side factorsthat we study in particular are the rising educational attainmentof the workforce and the rising human capital and labor forceattachment of womenmdashboth of which could potentially generateshifts in job task composition independent of demand shifts Aswe show below the task shifts that we documentmdashand their
1281SKILL CONTENT OF TECHNICAL CHANGE
associations with the adoption of computer technologymdashare aspervasive within gender education and occupation groups asbetween indicating that these supply side forces are not theprimary explanation for our results
We begin by presenting our informal ldquotask modelrdquo describinghow computerization affects the tasks that workers and machinesperform We next formalize this model in a production frameworkto develop empirical implications for task demand at the industryand occupation level Subsequent sections describe our datasources and test the modelrsquos main implications Drawing togetherthe empirical strands we nally assess the extent to whichchanges in the task composition can account for recent demandshifts favoring more educated workers This exercise shows thatestimated task shifts are economically large underscoring thepotential of the conceptual model to reconcile key facts
I THE TASK MODEL
We begin by conceptualizing work from a ldquomachinersquos-eyerdquoview as a series of tasks to be performed such as moving anobject performing a calculation communicating a piece of infor-mation or resolving a discrepancy Our model asks which ofthese tasks can be performed by a computer A general answer isfound by examining what is arguably the rst digital computerthe Jacquard Loom of 1801 Jacquardrsquos invention was a machinefor weaving fabrics with inlaid patterns specied by a programpunched onto cards and fed into the loom Some programs werequite sophisticated one surviving example uses more than 10000cards to weave a black and white silk portrait of Jacquard him-self3 Two centuries later the electronic descendants of Jacquardrsquosloom share with it two intrinsic traits First they rapidly and accu-rately perform repetitive tasks that are deterministically speciedby stored instructions (programs) that designate unambiguouslywhat actions the machine will perform at each contingency toachieve the desired result Second computers are ldquosymbolic proces-sorsrdquo acting upon abstract representations of information such asbinary numbers or in the loomrsquos case punched cards
Spurred by a more than trillionfold decline in the real price of
3 The Jacquard loom was also the inspiration for Charles Babbagersquos analyti-cal engine and Herman Hollerithrsquos punch card reader used to process the 1910United States Census
1282 QUARTERLY JOURNAL OF ECONOMICS
computing power [Nordhaus 2001] engineers have become vastlymore procient at applying the loomrsquos basic capabilitiesmdashrapidexecution of stored instructionsmdashto a panoply of tasks How doesthis advance affect the task composition of human work Theanswer depends both upon how computers substitute for or com-plement workers in carrying out specic tasks and how thesetasks substitute for one another We illustrate these cases byconsidering the application of computers to routine and nonrou-tine cognitive and manual tasks
In our usage a task is ldquoroutinerdquo if it can be accomplished bymachines following explicit programmed rules Many manualtasks that workers used to perform such as monitoring the tem-perature of a steel nishing line or moving a windshield into placeon an assembly line t this description Because these tasksrequire methodical repetition of an unwavering procedure theycan be exhaustively specied with programmed instructions andperformed by machines
A problem that arises with many commonplace manual andcognitive tasks however is that the procedures for accomplishingthem are not well understood As Polanyi [1966] observed ldquoWecan know more than we can tell [p 4] The skill of a drivercannot be replaced by a thorough schooling in the theory of themotorcar the knowledge I have of my own body differs altogetherfrom the knowledge of its physiology and the rules of rhymingand prosody do not tell me what a poem told me without anyknowledge of its rules [p 20]rdquo We refer to tasks tting Polanyirsquosdescription as ldquononroutinerdquo that is tasks for which the rules arenot sufciently well understood to be specied in computer codeand executed by machines Navigating a car through city trafcor deciphering the scrawled handwriting on a personal checkmdashminor undertakings for most adultsmdashare not routine tasks by ourdenition (see Beamish Levy and Murnane [1999] and AutorLevy and Murnane [2002] for examples) The reason is that thesetasks require visual and motor processing capabilities that can-not at present be described in terms of a set of programmablerules [Pinker 1997]4
Our conceptual model suggests that because of its decliningcost computer-controlled machinery should have substantially
4 If a manual task is performed in a well-controlled environment howeverit can often be automated despite the seeming need for adaptive visual or manualprocessing As Simon [1960] observed environmental control is a substitute forexibility
1283SKILL CONTENT OF TECHNICAL CHANGE
substituted for workers in performing routine manual tasks Thisphenomenon is not novel Substitution of machinery for repetitivehuman labor has been a thrust of technological change through-out the Industrial Revolution [Hounshell 1985 Mokyr 1990 Gol-din and Katz 1998] By increasing the feasibility of machinesubstitution for repetitive human tasks computerization fur-thersmdashand perhaps acceleratesmdashthis long-prevailing trend
The advent of computerization also marks a qualitative en-largement in the set of tasks that machines can perform Becausecomputers can perform symbolic processingmdashstoring retrievingand acting upon informationmdashthey augment or supplant humancognition in a large set of information-processing tasks that his-torically were not amenable to mechanization Over the last threedecades computers have substituted for the calculating coordi-nating and communicating functions of bookkeepers cashierstelephone operators and other handlers of repetitive informa-tion-processing tasks [Bresnahan 1999]
This substitution marks an important reversal Previousgenerations of high technology capital sharply increased demandfor human input of routine information-processing tasks as seenin the rapid rise of the clerking occupation in the nineteenthcentury [Chandler 1977 Goldin and Katz 1995] Like these tech-nologies computerization augments demand for clerical and in-formation-processing tasks But in contrast to its nineteenth cen-tury predecessors it permits these tasks to be automated
The capability of computers to substitute for workers in carry-ing out cognitive tasks is limited however Tasks demanding exi-bility creativity generalized problem-solving and complex commu-nicationsmdashwhat we call nonroutine cognitive tasksmdashdo not (yet)lend themselves to computerization [Bresnahan 1999] At presentthe need for explicit programmed instructions appears a bindingconstraint There are very few computer-based technologies that candraw inferences from models solve novel problems or form persua-sive arguments5 In the words of computer scientist Patrick Winston
5 It is however a fallacy to assume that a computer must reproduce all of thefunctions of a human to perform a task traditionally done by workers For exampleAutomatic Teller Machines have supplanted many bank teller functions althoughthey cannot verify signatures or make polite conversation while tallying changeClosely related computer capital may substitute for the routine components ofpredominantly nonroutine tasks eg on-board computers that direct taxi cabs Whatis required for our conceptual model is that the routine and nonroutine tasks embod-ied in a job are imperfect substitutes Consequently a decline in the price of accom-plishing routine tasks does not eliminate demand for nonroutine tasks
1284 QUARTERLY JOURNAL OF ECONOMICS
[1999] ldquoThe goal of understanding intelligence from a computa-tional point of view remains elusive Reasoning programs still ex-hibit little or no common sense Todayrsquos language programs trans-late simple sentences into database queries but those languageprograms are derailed by idioms metaphors convoluted syntax orungrammatical expressionsrdquo6
The implication of our discussion is that because presentcomputer technology is more substitutable for workers in carry-ing out routine tasks than nonroutine tasks it is a relativecomplement to workers in carrying out nonroutine tasks From aproduction function standpoint outward shifts in the supply ofroutine informational inputs both in quantity and quality in-crease the marginal productivity of workers performing nonrou-tine tasks that demand these inputs For example comprehen-sive bibliographic searches increase the quality of legal researchand timely market information improves the efciency of mana-gerial decision-making More tangibly because repetitive pre-dictable tasks are readily automated computerization of theworkplace raises demand for problem-solving and communica-tions tasks such as responding to discrepancies improving pro-duction processes and coordinating and managing the activitiesof others This changing allocation of tasks was anticipated byDrucker [1954] in the 1950s ldquoThe technological changes nowoccurring will carry [the Industrial Revolution] a big step furtherThey will not make human labor superuous On the contrarythey will require tremendous numbers of highly skilled andhighly trained menmdashmanagers to think through and plan highlytrained technicians and workers to design the new tools to pro-duce them to maintain them to direct themrdquo [p 22 bracketsadded]
Table I provides examples of tasks in each cell of our two-by-two matrix of workplace tasks (routine versus nonroutine man-ual versus information processing) and states our hypothesisabout the impact of computerization for each cell The next sec-tion formalizes these ideas and derives empirical implications7
6 Software that recognizes patterns (eg neural networks) or solves prob-lems based upon inductive reasoning from well-specied models is under devel-opment But these technologies have had little role in the ldquocomputer revolutionrdquo ofthe last several decades As one example current speech recognition softwarebased on pattern recognition can recognize words and short phrases but can onlyprocess rudimentary conversational speech [Zue and Glass 2000]
7 Our focus on task shifts in the process of production within given jobsoverlooks two other potentially complementary avenues by which technical
1285SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
capital substitutes and increase demand for nonroutine task inputwhich computer capital complements In net these forces will raiserelative demand for highly educated workers who hold comparativeadvantage in nonroutine versus routine tasks
To test these predictions we pair representative data on jobtask requirements from the Dictionary of Occupational Titles(DOT) with samples of employed workers from the Census andCurrent Population Survey to form a consistent panel of industryand occupational task input over the four-decade period from1960 to 1998 A unique virtue of this database is that it permitsus to analyze changes in task input within industries educationgroups and occupationsmdashphenomena that are normally unob-servable By measuring the tasks performed in jobs rather thanthe educational credentials of workers performing those jobs webelieve our study supplies a missing conceptual and empiricallink in the economic literature on technical change and skilldemand
Our analysis provides four main pieces of evidence support-ing our model
1) Commencing in the 1970s labor input of routine cognitiveand manual tasks in the U S economy declined and laborinput of nonroutine analytic and interactive tasks rose
2) Shifts in labor input favoring nonroutine and against rou-tine tasks were concentrated in rapidly computerizingindustries These shifts were small and insignicant inthe precomputer decade of the 1960s and accelerated ineach subsequent decade
3) The substitution away from routine and toward nonrou-tine labor input was not primarily accounted for by edu-cational upgrading rather task shifts are pervasive at alleducational levels
4) Paralleling the within-industry task shifts occupationsundergoing rapid computerization reduced input of rou-tine cognitive tasks and increased input of nonroutinecognitive tasks
We consider a number of economic and purely mechanicalalternative explanations for our results Two supply side factorsthat we study in particular are the rising educational attainmentof the workforce and the rising human capital and labor forceattachment of womenmdashboth of which could potentially generateshifts in job task composition independent of demand shifts Aswe show below the task shifts that we documentmdashand their
1281SKILL CONTENT OF TECHNICAL CHANGE
associations with the adoption of computer technologymdashare aspervasive within gender education and occupation groups asbetween indicating that these supply side forces are not theprimary explanation for our results
We begin by presenting our informal ldquotask modelrdquo describinghow computerization affects the tasks that workers and machinesperform We next formalize this model in a production frameworkto develop empirical implications for task demand at the industryand occupation level Subsequent sections describe our datasources and test the modelrsquos main implications Drawing togetherthe empirical strands we nally assess the extent to whichchanges in the task composition can account for recent demandshifts favoring more educated workers This exercise shows thatestimated task shifts are economically large underscoring thepotential of the conceptual model to reconcile key facts
I THE TASK MODEL
We begin by conceptualizing work from a ldquomachinersquos-eyerdquoview as a series of tasks to be performed such as moving anobject performing a calculation communicating a piece of infor-mation or resolving a discrepancy Our model asks which ofthese tasks can be performed by a computer A general answer isfound by examining what is arguably the rst digital computerthe Jacquard Loom of 1801 Jacquardrsquos invention was a machinefor weaving fabrics with inlaid patterns specied by a programpunched onto cards and fed into the loom Some programs werequite sophisticated one surviving example uses more than 10000cards to weave a black and white silk portrait of Jacquard him-self3 Two centuries later the electronic descendants of Jacquardrsquosloom share with it two intrinsic traits First they rapidly and accu-rately perform repetitive tasks that are deterministically speciedby stored instructions (programs) that designate unambiguouslywhat actions the machine will perform at each contingency toachieve the desired result Second computers are ldquosymbolic proces-sorsrdquo acting upon abstract representations of information such asbinary numbers or in the loomrsquos case punched cards
Spurred by a more than trillionfold decline in the real price of
3 The Jacquard loom was also the inspiration for Charles Babbagersquos analyti-cal engine and Herman Hollerithrsquos punch card reader used to process the 1910United States Census
1282 QUARTERLY JOURNAL OF ECONOMICS
computing power [Nordhaus 2001] engineers have become vastlymore procient at applying the loomrsquos basic capabilitiesmdashrapidexecution of stored instructionsmdashto a panoply of tasks How doesthis advance affect the task composition of human work Theanswer depends both upon how computers substitute for or com-plement workers in carrying out specic tasks and how thesetasks substitute for one another We illustrate these cases byconsidering the application of computers to routine and nonrou-tine cognitive and manual tasks
In our usage a task is ldquoroutinerdquo if it can be accomplished bymachines following explicit programmed rules Many manualtasks that workers used to perform such as monitoring the tem-perature of a steel nishing line or moving a windshield into placeon an assembly line t this description Because these tasksrequire methodical repetition of an unwavering procedure theycan be exhaustively specied with programmed instructions andperformed by machines
A problem that arises with many commonplace manual andcognitive tasks however is that the procedures for accomplishingthem are not well understood As Polanyi [1966] observed ldquoWecan know more than we can tell [p 4] The skill of a drivercannot be replaced by a thorough schooling in the theory of themotorcar the knowledge I have of my own body differs altogetherfrom the knowledge of its physiology and the rules of rhymingand prosody do not tell me what a poem told me without anyknowledge of its rules [p 20]rdquo We refer to tasks tting Polanyirsquosdescription as ldquononroutinerdquo that is tasks for which the rules arenot sufciently well understood to be specied in computer codeand executed by machines Navigating a car through city trafcor deciphering the scrawled handwriting on a personal checkmdashminor undertakings for most adultsmdashare not routine tasks by ourdenition (see Beamish Levy and Murnane [1999] and AutorLevy and Murnane [2002] for examples) The reason is that thesetasks require visual and motor processing capabilities that can-not at present be described in terms of a set of programmablerules [Pinker 1997]4
Our conceptual model suggests that because of its decliningcost computer-controlled machinery should have substantially
4 If a manual task is performed in a well-controlled environment howeverit can often be automated despite the seeming need for adaptive visual or manualprocessing As Simon [1960] observed environmental control is a substitute forexibility
1283SKILL CONTENT OF TECHNICAL CHANGE
substituted for workers in performing routine manual tasks Thisphenomenon is not novel Substitution of machinery for repetitivehuman labor has been a thrust of technological change through-out the Industrial Revolution [Hounshell 1985 Mokyr 1990 Gol-din and Katz 1998] By increasing the feasibility of machinesubstitution for repetitive human tasks computerization fur-thersmdashand perhaps acceleratesmdashthis long-prevailing trend
The advent of computerization also marks a qualitative en-largement in the set of tasks that machines can perform Becausecomputers can perform symbolic processingmdashstoring retrievingand acting upon informationmdashthey augment or supplant humancognition in a large set of information-processing tasks that his-torically were not amenable to mechanization Over the last threedecades computers have substituted for the calculating coordi-nating and communicating functions of bookkeepers cashierstelephone operators and other handlers of repetitive informa-tion-processing tasks [Bresnahan 1999]
This substitution marks an important reversal Previousgenerations of high technology capital sharply increased demandfor human input of routine information-processing tasks as seenin the rapid rise of the clerking occupation in the nineteenthcentury [Chandler 1977 Goldin and Katz 1995] Like these tech-nologies computerization augments demand for clerical and in-formation-processing tasks But in contrast to its nineteenth cen-tury predecessors it permits these tasks to be automated
The capability of computers to substitute for workers in carry-ing out cognitive tasks is limited however Tasks demanding exi-bility creativity generalized problem-solving and complex commu-nicationsmdashwhat we call nonroutine cognitive tasksmdashdo not (yet)lend themselves to computerization [Bresnahan 1999] At presentthe need for explicit programmed instructions appears a bindingconstraint There are very few computer-based technologies that candraw inferences from models solve novel problems or form persua-sive arguments5 In the words of computer scientist Patrick Winston
5 It is however a fallacy to assume that a computer must reproduce all of thefunctions of a human to perform a task traditionally done by workers For exampleAutomatic Teller Machines have supplanted many bank teller functions althoughthey cannot verify signatures or make polite conversation while tallying changeClosely related computer capital may substitute for the routine components ofpredominantly nonroutine tasks eg on-board computers that direct taxi cabs Whatis required for our conceptual model is that the routine and nonroutine tasks embod-ied in a job are imperfect substitutes Consequently a decline in the price of accom-plishing routine tasks does not eliminate demand for nonroutine tasks
1284 QUARTERLY JOURNAL OF ECONOMICS
[1999] ldquoThe goal of understanding intelligence from a computa-tional point of view remains elusive Reasoning programs still ex-hibit little or no common sense Todayrsquos language programs trans-late simple sentences into database queries but those languageprograms are derailed by idioms metaphors convoluted syntax orungrammatical expressionsrdquo6
The implication of our discussion is that because presentcomputer technology is more substitutable for workers in carry-ing out routine tasks than nonroutine tasks it is a relativecomplement to workers in carrying out nonroutine tasks From aproduction function standpoint outward shifts in the supply ofroutine informational inputs both in quantity and quality in-crease the marginal productivity of workers performing nonrou-tine tasks that demand these inputs For example comprehen-sive bibliographic searches increase the quality of legal researchand timely market information improves the efciency of mana-gerial decision-making More tangibly because repetitive pre-dictable tasks are readily automated computerization of theworkplace raises demand for problem-solving and communica-tions tasks such as responding to discrepancies improving pro-duction processes and coordinating and managing the activitiesof others This changing allocation of tasks was anticipated byDrucker [1954] in the 1950s ldquoThe technological changes nowoccurring will carry [the Industrial Revolution] a big step furtherThey will not make human labor superuous On the contrarythey will require tremendous numbers of highly skilled andhighly trained menmdashmanagers to think through and plan highlytrained technicians and workers to design the new tools to pro-duce them to maintain them to direct themrdquo [p 22 bracketsadded]
Table I provides examples of tasks in each cell of our two-by-two matrix of workplace tasks (routine versus nonroutine man-ual versus information processing) and states our hypothesisabout the impact of computerization for each cell The next sec-tion formalizes these ideas and derives empirical implications7
6 Software that recognizes patterns (eg neural networks) or solves prob-lems based upon inductive reasoning from well-specied models is under devel-opment But these technologies have had little role in the ldquocomputer revolutionrdquo ofthe last several decades As one example current speech recognition softwarebased on pattern recognition can recognize words and short phrases but can onlyprocess rudimentary conversational speech [Zue and Glass 2000]
7 Our focus on task shifts in the process of production within given jobsoverlooks two other potentially complementary avenues by which technical
1285SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
associations with the adoption of computer technologymdashare aspervasive within gender education and occupation groups asbetween indicating that these supply side forces are not theprimary explanation for our results
We begin by presenting our informal ldquotask modelrdquo describinghow computerization affects the tasks that workers and machinesperform We next formalize this model in a production frameworkto develop empirical implications for task demand at the industryand occupation level Subsequent sections describe our datasources and test the modelrsquos main implications Drawing togetherthe empirical strands we nally assess the extent to whichchanges in the task composition can account for recent demandshifts favoring more educated workers This exercise shows thatestimated task shifts are economically large underscoring thepotential of the conceptual model to reconcile key facts
I THE TASK MODEL
We begin by conceptualizing work from a ldquomachinersquos-eyerdquoview as a series of tasks to be performed such as moving anobject performing a calculation communicating a piece of infor-mation or resolving a discrepancy Our model asks which ofthese tasks can be performed by a computer A general answer isfound by examining what is arguably the rst digital computerthe Jacquard Loom of 1801 Jacquardrsquos invention was a machinefor weaving fabrics with inlaid patterns specied by a programpunched onto cards and fed into the loom Some programs werequite sophisticated one surviving example uses more than 10000cards to weave a black and white silk portrait of Jacquard him-self3 Two centuries later the electronic descendants of Jacquardrsquosloom share with it two intrinsic traits First they rapidly and accu-rately perform repetitive tasks that are deterministically speciedby stored instructions (programs) that designate unambiguouslywhat actions the machine will perform at each contingency toachieve the desired result Second computers are ldquosymbolic proces-sorsrdquo acting upon abstract representations of information such asbinary numbers or in the loomrsquos case punched cards
Spurred by a more than trillionfold decline in the real price of
3 The Jacquard loom was also the inspiration for Charles Babbagersquos analyti-cal engine and Herman Hollerithrsquos punch card reader used to process the 1910United States Census
1282 QUARTERLY JOURNAL OF ECONOMICS
computing power [Nordhaus 2001] engineers have become vastlymore procient at applying the loomrsquos basic capabilitiesmdashrapidexecution of stored instructionsmdashto a panoply of tasks How doesthis advance affect the task composition of human work Theanswer depends both upon how computers substitute for or com-plement workers in carrying out specic tasks and how thesetasks substitute for one another We illustrate these cases byconsidering the application of computers to routine and nonrou-tine cognitive and manual tasks
In our usage a task is ldquoroutinerdquo if it can be accomplished bymachines following explicit programmed rules Many manualtasks that workers used to perform such as monitoring the tem-perature of a steel nishing line or moving a windshield into placeon an assembly line t this description Because these tasksrequire methodical repetition of an unwavering procedure theycan be exhaustively specied with programmed instructions andperformed by machines
A problem that arises with many commonplace manual andcognitive tasks however is that the procedures for accomplishingthem are not well understood As Polanyi [1966] observed ldquoWecan know more than we can tell [p 4] The skill of a drivercannot be replaced by a thorough schooling in the theory of themotorcar the knowledge I have of my own body differs altogetherfrom the knowledge of its physiology and the rules of rhymingand prosody do not tell me what a poem told me without anyknowledge of its rules [p 20]rdquo We refer to tasks tting Polanyirsquosdescription as ldquononroutinerdquo that is tasks for which the rules arenot sufciently well understood to be specied in computer codeand executed by machines Navigating a car through city trafcor deciphering the scrawled handwriting on a personal checkmdashminor undertakings for most adultsmdashare not routine tasks by ourdenition (see Beamish Levy and Murnane [1999] and AutorLevy and Murnane [2002] for examples) The reason is that thesetasks require visual and motor processing capabilities that can-not at present be described in terms of a set of programmablerules [Pinker 1997]4
Our conceptual model suggests that because of its decliningcost computer-controlled machinery should have substantially
4 If a manual task is performed in a well-controlled environment howeverit can often be automated despite the seeming need for adaptive visual or manualprocessing As Simon [1960] observed environmental control is a substitute forexibility
1283SKILL CONTENT OF TECHNICAL CHANGE
substituted for workers in performing routine manual tasks Thisphenomenon is not novel Substitution of machinery for repetitivehuman labor has been a thrust of technological change through-out the Industrial Revolution [Hounshell 1985 Mokyr 1990 Gol-din and Katz 1998] By increasing the feasibility of machinesubstitution for repetitive human tasks computerization fur-thersmdashand perhaps acceleratesmdashthis long-prevailing trend
The advent of computerization also marks a qualitative en-largement in the set of tasks that machines can perform Becausecomputers can perform symbolic processingmdashstoring retrievingand acting upon informationmdashthey augment or supplant humancognition in a large set of information-processing tasks that his-torically were not amenable to mechanization Over the last threedecades computers have substituted for the calculating coordi-nating and communicating functions of bookkeepers cashierstelephone operators and other handlers of repetitive informa-tion-processing tasks [Bresnahan 1999]
This substitution marks an important reversal Previousgenerations of high technology capital sharply increased demandfor human input of routine information-processing tasks as seenin the rapid rise of the clerking occupation in the nineteenthcentury [Chandler 1977 Goldin and Katz 1995] Like these tech-nologies computerization augments demand for clerical and in-formation-processing tasks But in contrast to its nineteenth cen-tury predecessors it permits these tasks to be automated
The capability of computers to substitute for workers in carry-ing out cognitive tasks is limited however Tasks demanding exi-bility creativity generalized problem-solving and complex commu-nicationsmdashwhat we call nonroutine cognitive tasksmdashdo not (yet)lend themselves to computerization [Bresnahan 1999] At presentthe need for explicit programmed instructions appears a bindingconstraint There are very few computer-based technologies that candraw inferences from models solve novel problems or form persua-sive arguments5 In the words of computer scientist Patrick Winston
5 It is however a fallacy to assume that a computer must reproduce all of thefunctions of a human to perform a task traditionally done by workers For exampleAutomatic Teller Machines have supplanted many bank teller functions althoughthey cannot verify signatures or make polite conversation while tallying changeClosely related computer capital may substitute for the routine components ofpredominantly nonroutine tasks eg on-board computers that direct taxi cabs Whatis required for our conceptual model is that the routine and nonroutine tasks embod-ied in a job are imperfect substitutes Consequently a decline in the price of accom-plishing routine tasks does not eliminate demand for nonroutine tasks
1284 QUARTERLY JOURNAL OF ECONOMICS
[1999] ldquoThe goal of understanding intelligence from a computa-tional point of view remains elusive Reasoning programs still ex-hibit little or no common sense Todayrsquos language programs trans-late simple sentences into database queries but those languageprograms are derailed by idioms metaphors convoluted syntax orungrammatical expressionsrdquo6
The implication of our discussion is that because presentcomputer technology is more substitutable for workers in carry-ing out routine tasks than nonroutine tasks it is a relativecomplement to workers in carrying out nonroutine tasks From aproduction function standpoint outward shifts in the supply ofroutine informational inputs both in quantity and quality in-crease the marginal productivity of workers performing nonrou-tine tasks that demand these inputs For example comprehen-sive bibliographic searches increase the quality of legal researchand timely market information improves the efciency of mana-gerial decision-making More tangibly because repetitive pre-dictable tasks are readily automated computerization of theworkplace raises demand for problem-solving and communica-tions tasks such as responding to discrepancies improving pro-duction processes and coordinating and managing the activitiesof others This changing allocation of tasks was anticipated byDrucker [1954] in the 1950s ldquoThe technological changes nowoccurring will carry [the Industrial Revolution] a big step furtherThey will not make human labor superuous On the contrarythey will require tremendous numbers of highly skilled andhighly trained menmdashmanagers to think through and plan highlytrained technicians and workers to design the new tools to pro-duce them to maintain them to direct themrdquo [p 22 bracketsadded]
Table I provides examples of tasks in each cell of our two-by-two matrix of workplace tasks (routine versus nonroutine man-ual versus information processing) and states our hypothesisabout the impact of computerization for each cell The next sec-tion formalizes these ideas and derives empirical implications7
6 Software that recognizes patterns (eg neural networks) or solves prob-lems based upon inductive reasoning from well-specied models is under devel-opment But these technologies have had little role in the ldquocomputer revolutionrdquo ofthe last several decades As one example current speech recognition softwarebased on pattern recognition can recognize words and short phrases but can onlyprocess rudimentary conversational speech [Zue and Glass 2000]
7 Our focus on task shifts in the process of production within given jobsoverlooks two other potentially complementary avenues by which technical
1285SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
computing power [Nordhaus 2001] engineers have become vastlymore procient at applying the loomrsquos basic capabilitiesmdashrapidexecution of stored instructionsmdashto a panoply of tasks How doesthis advance affect the task composition of human work Theanswer depends both upon how computers substitute for or com-plement workers in carrying out specic tasks and how thesetasks substitute for one another We illustrate these cases byconsidering the application of computers to routine and nonrou-tine cognitive and manual tasks
In our usage a task is ldquoroutinerdquo if it can be accomplished bymachines following explicit programmed rules Many manualtasks that workers used to perform such as monitoring the tem-perature of a steel nishing line or moving a windshield into placeon an assembly line t this description Because these tasksrequire methodical repetition of an unwavering procedure theycan be exhaustively specied with programmed instructions andperformed by machines
A problem that arises with many commonplace manual andcognitive tasks however is that the procedures for accomplishingthem are not well understood As Polanyi [1966] observed ldquoWecan know more than we can tell [p 4] The skill of a drivercannot be replaced by a thorough schooling in the theory of themotorcar the knowledge I have of my own body differs altogetherfrom the knowledge of its physiology and the rules of rhymingand prosody do not tell me what a poem told me without anyknowledge of its rules [p 20]rdquo We refer to tasks tting Polanyirsquosdescription as ldquononroutinerdquo that is tasks for which the rules arenot sufciently well understood to be specied in computer codeand executed by machines Navigating a car through city trafcor deciphering the scrawled handwriting on a personal checkmdashminor undertakings for most adultsmdashare not routine tasks by ourdenition (see Beamish Levy and Murnane [1999] and AutorLevy and Murnane [2002] for examples) The reason is that thesetasks require visual and motor processing capabilities that can-not at present be described in terms of a set of programmablerules [Pinker 1997]4
Our conceptual model suggests that because of its decliningcost computer-controlled machinery should have substantially
4 If a manual task is performed in a well-controlled environment howeverit can often be automated despite the seeming need for adaptive visual or manualprocessing As Simon [1960] observed environmental control is a substitute forexibility
1283SKILL CONTENT OF TECHNICAL CHANGE
substituted for workers in performing routine manual tasks Thisphenomenon is not novel Substitution of machinery for repetitivehuman labor has been a thrust of technological change through-out the Industrial Revolution [Hounshell 1985 Mokyr 1990 Gol-din and Katz 1998] By increasing the feasibility of machinesubstitution for repetitive human tasks computerization fur-thersmdashand perhaps acceleratesmdashthis long-prevailing trend
The advent of computerization also marks a qualitative en-largement in the set of tasks that machines can perform Becausecomputers can perform symbolic processingmdashstoring retrievingand acting upon informationmdashthey augment or supplant humancognition in a large set of information-processing tasks that his-torically were not amenable to mechanization Over the last threedecades computers have substituted for the calculating coordi-nating and communicating functions of bookkeepers cashierstelephone operators and other handlers of repetitive informa-tion-processing tasks [Bresnahan 1999]
This substitution marks an important reversal Previousgenerations of high technology capital sharply increased demandfor human input of routine information-processing tasks as seenin the rapid rise of the clerking occupation in the nineteenthcentury [Chandler 1977 Goldin and Katz 1995] Like these tech-nologies computerization augments demand for clerical and in-formation-processing tasks But in contrast to its nineteenth cen-tury predecessors it permits these tasks to be automated
The capability of computers to substitute for workers in carry-ing out cognitive tasks is limited however Tasks demanding exi-bility creativity generalized problem-solving and complex commu-nicationsmdashwhat we call nonroutine cognitive tasksmdashdo not (yet)lend themselves to computerization [Bresnahan 1999] At presentthe need for explicit programmed instructions appears a bindingconstraint There are very few computer-based technologies that candraw inferences from models solve novel problems or form persua-sive arguments5 In the words of computer scientist Patrick Winston
5 It is however a fallacy to assume that a computer must reproduce all of thefunctions of a human to perform a task traditionally done by workers For exampleAutomatic Teller Machines have supplanted many bank teller functions althoughthey cannot verify signatures or make polite conversation while tallying changeClosely related computer capital may substitute for the routine components ofpredominantly nonroutine tasks eg on-board computers that direct taxi cabs Whatis required for our conceptual model is that the routine and nonroutine tasks embod-ied in a job are imperfect substitutes Consequently a decline in the price of accom-plishing routine tasks does not eliminate demand for nonroutine tasks
1284 QUARTERLY JOURNAL OF ECONOMICS
[1999] ldquoThe goal of understanding intelligence from a computa-tional point of view remains elusive Reasoning programs still ex-hibit little or no common sense Todayrsquos language programs trans-late simple sentences into database queries but those languageprograms are derailed by idioms metaphors convoluted syntax orungrammatical expressionsrdquo6
The implication of our discussion is that because presentcomputer technology is more substitutable for workers in carry-ing out routine tasks than nonroutine tasks it is a relativecomplement to workers in carrying out nonroutine tasks From aproduction function standpoint outward shifts in the supply ofroutine informational inputs both in quantity and quality in-crease the marginal productivity of workers performing nonrou-tine tasks that demand these inputs For example comprehen-sive bibliographic searches increase the quality of legal researchand timely market information improves the efciency of mana-gerial decision-making More tangibly because repetitive pre-dictable tasks are readily automated computerization of theworkplace raises demand for problem-solving and communica-tions tasks such as responding to discrepancies improving pro-duction processes and coordinating and managing the activitiesof others This changing allocation of tasks was anticipated byDrucker [1954] in the 1950s ldquoThe technological changes nowoccurring will carry [the Industrial Revolution] a big step furtherThey will not make human labor superuous On the contrarythey will require tremendous numbers of highly skilled andhighly trained menmdashmanagers to think through and plan highlytrained technicians and workers to design the new tools to pro-duce them to maintain them to direct themrdquo [p 22 bracketsadded]
Table I provides examples of tasks in each cell of our two-by-two matrix of workplace tasks (routine versus nonroutine man-ual versus information processing) and states our hypothesisabout the impact of computerization for each cell The next sec-tion formalizes these ideas and derives empirical implications7
6 Software that recognizes patterns (eg neural networks) or solves prob-lems based upon inductive reasoning from well-specied models is under devel-opment But these technologies have had little role in the ldquocomputer revolutionrdquo ofthe last several decades As one example current speech recognition softwarebased on pattern recognition can recognize words and short phrases but can onlyprocess rudimentary conversational speech [Zue and Glass 2000]
7 Our focus on task shifts in the process of production within given jobsoverlooks two other potentially complementary avenues by which technical
1285SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
substituted for workers in performing routine manual tasks Thisphenomenon is not novel Substitution of machinery for repetitivehuman labor has been a thrust of technological change through-out the Industrial Revolution [Hounshell 1985 Mokyr 1990 Gol-din and Katz 1998] By increasing the feasibility of machinesubstitution for repetitive human tasks computerization fur-thersmdashand perhaps acceleratesmdashthis long-prevailing trend
The advent of computerization also marks a qualitative en-largement in the set of tasks that machines can perform Becausecomputers can perform symbolic processingmdashstoring retrievingand acting upon informationmdashthey augment or supplant humancognition in a large set of information-processing tasks that his-torically were not amenable to mechanization Over the last threedecades computers have substituted for the calculating coordi-nating and communicating functions of bookkeepers cashierstelephone operators and other handlers of repetitive informa-tion-processing tasks [Bresnahan 1999]
This substitution marks an important reversal Previousgenerations of high technology capital sharply increased demandfor human input of routine information-processing tasks as seenin the rapid rise of the clerking occupation in the nineteenthcentury [Chandler 1977 Goldin and Katz 1995] Like these tech-nologies computerization augments demand for clerical and in-formation-processing tasks But in contrast to its nineteenth cen-tury predecessors it permits these tasks to be automated
The capability of computers to substitute for workers in carry-ing out cognitive tasks is limited however Tasks demanding exi-bility creativity generalized problem-solving and complex commu-nicationsmdashwhat we call nonroutine cognitive tasksmdashdo not (yet)lend themselves to computerization [Bresnahan 1999] At presentthe need for explicit programmed instructions appears a bindingconstraint There are very few computer-based technologies that candraw inferences from models solve novel problems or form persua-sive arguments5 In the words of computer scientist Patrick Winston
5 It is however a fallacy to assume that a computer must reproduce all of thefunctions of a human to perform a task traditionally done by workers For exampleAutomatic Teller Machines have supplanted many bank teller functions althoughthey cannot verify signatures or make polite conversation while tallying changeClosely related computer capital may substitute for the routine components ofpredominantly nonroutine tasks eg on-board computers that direct taxi cabs Whatis required for our conceptual model is that the routine and nonroutine tasks embod-ied in a job are imperfect substitutes Consequently a decline in the price of accom-plishing routine tasks does not eliminate demand for nonroutine tasks
1284 QUARTERLY JOURNAL OF ECONOMICS
[1999] ldquoThe goal of understanding intelligence from a computa-tional point of view remains elusive Reasoning programs still ex-hibit little or no common sense Todayrsquos language programs trans-late simple sentences into database queries but those languageprograms are derailed by idioms metaphors convoluted syntax orungrammatical expressionsrdquo6
The implication of our discussion is that because presentcomputer technology is more substitutable for workers in carry-ing out routine tasks than nonroutine tasks it is a relativecomplement to workers in carrying out nonroutine tasks From aproduction function standpoint outward shifts in the supply ofroutine informational inputs both in quantity and quality in-crease the marginal productivity of workers performing nonrou-tine tasks that demand these inputs For example comprehen-sive bibliographic searches increase the quality of legal researchand timely market information improves the efciency of mana-gerial decision-making More tangibly because repetitive pre-dictable tasks are readily automated computerization of theworkplace raises demand for problem-solving and communica-tions tasks such as responding to discrepancies improving pro-duction processes and coordinating and managing the activitiesof others This changing allocation of tasks was anticipated byDrucker [1954] in the 1950s ldquoThe technological changes nowoccurring will carry [the Industrial Revolution] a big step furtherThey will not make human labor superuous On the contrarythey will require tremendous numbers of highly skilled andhighly trained menmdashmanagers to think through and plan highlytrained technicians and workers to design the new tools to pro-duce them to maintain them to direct themrdquo [p 22 bracketsadded]
Table I provides examples of tasks in each cell of our two-by-two matrix of workplace tasks (routine versus nonroutine man-ual versus information processing) and states our hypothesisabout the impact of computerization for each cell The next sec-tion formalizes these ideas and derives empirical implications7
6 Software that recognizes patterns (eg neural networks) or solves prob-lems based upon inductive reasoning from well-specied models is under devel-opment But these technologies have had little role in the ldquocomputer revolutionrdquo ofthe last several decades As one example current speech recognition softwarebased on pattern recognition can recognize words and short phrases but can onlyprocess rudimentary conversational speech [Zue and Glass 2000]
7 Our focus on task shifts in the process of production within given jobsoverlooks two other potentially complementary avenues by which technical
1285SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
[1999] ldquoThe goal of understanding intelligence from a computa-tional point of view remains elusive Reasoning programs still ex-hibit little or no common sense Todayrsquos language programs trans-late simple sentences into database queries but those languageprograms are derailed by idioms metaphors convoluted syntax orungrammatical expressionsrdquo6
The implication of our discussion is that because presentcomputer technology is more substitutable for workers in carry-ing out routine tasks than nonroutine tasks it is a relativecomplement to workers in carrying out nonroutine tasks From aproduction function standpoint outward shifts in the supply ofroutine informational inputs both in quantity and quality in-crease the marginal productivity of workers performing nonrou-tine tasks that demand these inputs For example comprehen-sive bibliographic searches increase the quality of legal researchand timely market information improves the efciency of mana-gerial decision-making More tangibly because repetitive pre-dictable tasks are readily automated computerization of theworkplace raises demand for problem-solving and communica-tions tasks such as responding to discrepancies improving pro-duction processes and coordinating and managing the activitiesof others This changing allocation of tasks was anticipated byDrucker [1954] in the 1950s ldquoThe technological changes nowoccurring will carry [the Industrial Revolution] a big step furtherThey will not make human labor superuous On the contrarythey will require tremendous numbers of highly skilled andhighly trained menmdashmanagers to think through and plan highlytrained technicians and workers to design the new tools to pro-duce them to maintain them to direct themrdquo [p 22 bracketsadded]
Table I provides examples of tasks in each cell of our two-by-two matrix of workplace tasks (routine versus nonroutine man-ual versus information processing) and states our hypothesisabout the impact of computerization for each cell The next sec-tion formalizes these ideas and derives empirical implications7
6 Software that recognizes patterns (eg neural networks) or solves prob-lems based upon inductive reasoning from well-specied models is under devel-opment But these technologies have had little role in the ldquocomputer revolutionrdquo ofthe last several decades As one example current speech recognition softwarebased on pattern recognition can recognize words and short phrases but can onlyprocess rudimentary conversational speech [Zue and Glass 2000]
7 Our focus on task shifts in the process of production within given jobsoverlooks two other potentially complementary avenues by which technical
1285SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
IA The Demand for Routine and Nonroutine Tasks
The informal task framework above implies three postulatesabout how computer capital interacts with human labor input
A1 Computer capital is more substitutable for human laborin carrying out routine tasks than nonroutine tasks
A2 Routine and nonroutine tasks are themselves imperfectsubstitutes
A3 Greater intensity of routine inputs increases the mar-ginal productivity of nonroutine inputs
To develop the formal implications of these assumptions wewrite a simple general equilibrium production model with two
change impacts job task demands First innovations in the organization of pro-duction reinforce the task-level shifts that we describe above See Adler [1986]Zuboff [1988] Levy and Murnane [1996] Acemoglu [1999] Bresnahan [1999]Bartel Ichniowski and Shaw [2000] Brynjolfsson and Hitt [2000] Lindbeck andSnower [2000] Mobius [2000] Thesmar and Thoenig [2000] Caroli and VanReenen [2001] Fernandez [2001] Autor Levy and Murnane [2002] and Bresna-han Brynjolfsson and Hitt [2002] for examples Second distinct from our focus onprocess innovations Xiang [2002] presents evidence that product innovations overthe past 25 years have also raised skill demands
TABLE IPREDICTIONS OF TASK MODEL FOR THE IMPACT OF COMPUTERIZATION ON FOUR
CATEGORIES OF WORKPLACE TASKS
Routine tasks Nonroutine tasks
Analytic and interactive tasks
Examples Record-keeping Formingtesting hypotheses Calculation Medical diagnosis Repetitive customer service
(eg bank teller) Legal writing Persuadingselling Managing others
Computer impact Substantial substitution Strong complementarities
Manual tasks
Examples Picking or sorting Janitorial services Repetitive assembly Truck driving
Computer impact Substantial substitution Limited opportunities forsubstitution orcomplementarity
1286 QUARTERLY JOURNAL OF ECONOMICS
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
task inputs routine and nonroutine that are used to produceoutput Q which sells at price one Because our discussionstresses that computers neither strongly substitute nor stronglycomplement nonroutine manual tasks we consider this model topertain primarily to routine cognitive and routine manual tasksand nonroutine analytic and nonroutine interactive tasks
We assume for tractability an aggregate constant returns toscale Cobb-Douglas production function of the form
(1) Q 5 ~LR 1 C12bLNb b [ ~01
where LR and LN are routine and nonroutine labor inputs and Cis computer capital all measured in efciency units Computercapital is supplied perfectly elastically at market price r perefciency unit where r is falling exogenously with time due totechnical advances The declining price of computer capital is thecausal force in our model8
We assume that computer capital and labor are perfect sub-stitutes in carrying out routine tasks Cobb-Douglas technologyfurther implies that the elasticity of substitution between routineand nonroutine tasks is one and hence computer capital andnonroutine task inputs are relative complements While the as-sumption of perfect substitutability between computer capitaland routine task input places assumptions A1 and A2 in boldrelief the only substantive requirement for our model is thatcomputer capital is more substitutable for routine than nonrou-tine tasks Observe that routine and nonroutine tasks are q-complements the marginal productivity of nonroutine tasks riseswith the quantity of routine task input consistent with assump-tion A39
We assume a large number of income-maximizing workerseach of whom inelastically supplies one unit of labor Workershave heterogeneous productivity endowments in both routine andnonroutine tasks with Ei 5 [rini] and 1 $ r i ni 0 i A givenworker can choose to supply ri efciency units of routine taskinput n i efciency units of nonroutine task input or any convex
8 Borghans and ter Weel [2002] offer a related model exploring how thedeclining price of computer capital affects the diffusion of computers and thedistribution of wages A key difference is that the tasks performed by computersand workers are inseparable in the Borghans-ter Weel model Accordingly com-puterization alters wage levels but does not directly change the allocation ofhuman labor input across task types This latter point is the focus of our modeland empirical analysis
9 Specically ]2Q]LN](LR 1 C) 5 b(1 2 b) LNb2 1 (LR 1 C)b 0
1287SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
combination of the two Hence Li 5 [lir i (1 2 l i)ni] where 0 li 1 These assumptions imply that workers will choose tasksaccording to comparative advantage as in Roy [1951] We adoptthe Roy framework because it implies that relative task supplywill respond elastically to relative wage levels If instead work-ers were bound to given tasks the implications of our model fortask productivity would be unchanged but technical progressreected by a decline in r would not generate re-sorting of work-ers across jobs
Two main conditions govern market equilibrium in thismodel First given the perfect substitutability of routine tasksand computer capital the wage per efciency unit of routine taskinput is pinned down by the price of computer capital10
(2) wR 5 r
Second worker self-selection among occupationsmdashroutine versusnonroutinemdashclears the labor market
Dene the relative efciency of individual i at nonroutineversus routine tasks as h i 5 nir1 Our assumptions above implythat h i [ (0`) At the labor market equilibrium the marginalworker with relative efciency units h is indifferent betweenperforming routine and nonroutine tasks when
(3) h 5 wRwN
Individual i supplies routine labor (l i 5 1) if hi h andsupplies nonroutine labor otherwise (li 5 0)
To quantify labor supply write the functions g(h) h(h)which sum population endowments in efciency units of routineand nonroutine tasks respectively at each value of h Henceg(h) 5 yenir i z I[hi h] and h(h) 5 yenin i z I[h i $ h] where I[ z ]is the indicator function We further assume that h i has nonzerosupport at all h i [ (0`) so that that h(h) is continuouslyupward sloping in h and g(h) is continuously downward sloping
Assuming that the economy operates on the demand curveproductive efciency requires
(4) wR 5]Q]LR
5 ~1 2 bu2b and wN 5]Q]LN
5 bu12b
10 We implicitly assume that the shadow price of nonroutine tasks absentcomputer capital exceeds r and hence equation (2) holds with equality In theprecomputer era it is likely that wR r
1288 QUARTERLY JOURNAL OF ECONOMICS
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
where u in this expression is the ratio of routine to nonroutinetask input in production
(5) u ~C 1 g~hh~h
These equations provide equilibrium conditions for the mod-elrsquos ve endogenous variables (wR wN uCh) We use them toanalyze how a decline in the price of computer capital affects taskinput wages and labor supply beginning with the wage paid toroutine task input
It is immediate from (2) that a decline in the price of com-puter capital reduces wR one-for-one ](ln wR)](ln r) 5 1 andhence demand for routine task input rises
(6)] ln u
] ln r5 2
1b
From the perspective of producers the rise in routine taskdemand could be met by either an increase in C or an increase inLR (or both) Only the rst of these will occur however Becauseroutine and nonroutine tasks are productive complements (spe-cically q-complements) the relative wage paid to nonroutinetasks rises as r declines
(7)] ln~wNwR
] ln r5 2
1b
and] ln h] ln r
51b
Consequently marginal workers will reallocate their labor inputfrom routine to nonroutine tasks Increased demand for routinetasks must be met entirely by an inux of computer capital
Hence an exogenous decline in the price of computer capitalraises the marginal productivity of nonroutine tasks causingworkers to reallocate labor supply from routine to nonroutinetask input Although routine labor input declines an inow ofcomputer capital more than compensates yielding a net increasein the intensity of routine task input in production
IB Industry Level Implications
Does this model have testable microeconomic implicationsBecause the causal force in the model is the price of computercapital in one sense we have only a single macroeconomic timeseries available However additional leverage may be gained by
1289SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
considering equation (1) as representing the production functionof a single industry with distinct industries j producing outputsqj that demand different mixes of routine and nonroutine tasksWe write industry j rsquos production function as
(8) q j 5 rj12bjn j
bj b j [ ~01
where bj is the industry-specic factor share of nonroutine tasksand rj nj denote the industryrsquos task inputs All industries useCobb-Douglas technology but industries with smaller bj are moreroutine task intensive
We assume that consumer preferences in this economy maybe represented with a Dixit-Stiglitz [1977] utility function
(9) U~q1q2 qj 5 ~Oj
qj12v1~12v
where 0 v 1 The elasticity of demand for each good is2(1v) with the market clearing price inversely proportional tothe quantity produced pj(qj) qj
2 v Industry prot maximization yields the following rst-order
conditions for wages
(10) r 5 njbjrj
2bj~1 2 b j~1 2 v~njbjrj
12bj2v and
wN 5 nbj21r12bjbj~1 2 v~n jbjr j
12bj2v
Rearranging to obtain factor demands gives
(11) nj 5 wN21v~bj ~1 2 v1v SwN
rz
~1 2 bj
bjD ~~12b j~12vv
and
rj 5 r21v~~1 2 b j~1 2 v1vS wN
rz
~1 2 b j
bjD ~bj~v21v
Using these equations we obtain the following three proposi-tions which we test empirically below
P1 Although all industries face the same price of computercapital r the degree to which industries adopt thiscapital as its price declines depends upon bj For a givenprice decline the proportionate increase in demand forroutine task input is larger in routine-task-intensive (bj
small) industries as may be seen by taking the cross-partial derivative of routine task demand with respect tor and b j
1290 QUARTERLY JOURNAL OF ECONOMICS
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
d ln rj
dr5
bj~1 2 v 2 1vr
0 andd2 ln rj
drdbj5
1 2 vvr
0
Although we cannot observe b j a logical proxy for it isthe observed industry level of routine task input in theprecomputerization era We therefore test whether in-dustries that were historically (ie precomputer era)intensive in routine tasks adopted computer capital to agreater extent than industries that were not
P2 Due to the complementarity between routine and non-routine inputs a decline in the price of computer capitalalso raises demand for nonroutine task input This de-mand increase is proportionately larger in routine-task-intensive industries
d ln nj
dr5
~bj 2 1~1 2 v
vr 0
d2 ln ndrdb 5
1 2 vvr
0
Recall however that labor supply to routine tasks de-clines with r Rising routine task demand must thereforebe satised with computer capital Hence sectors thatinvest relatively more in computer capital will show alarger rise in nonroutine labor input and a larger declinein routine labor input
P3 The previous propositions refer to industry demandsAnalogously we expect that occupations that make rela-tively larger investments in computer capital will showlarger increases in labor input of nonroutine tasks andlarger decreases in labor input of routine tasks
II EMPIRICAL IMPLEMENTATION
Our analysis requires measures of tasks performed in par-ticular jobs and their changes over time We draw on informationfrom the Fourth [1977] Edition and Revised Fourth [1991] editionof the U S Department of Laborrsquos Dictionary of OccupationalTitles (DOT) Many of the details of our data construction areprovided in the Data Appendix Here we discuss the most salientfeatures The U S Department of Labor released the rst editionof the DOT in 1939 to ldquofurnish public employment ofces withinformation and techniques [to] facilitate proper classicationand placement of work seekersrdquo [U S Department of Labor1939xi as quoted in Miller et al 1980] Although the DOT was
1291SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
updated four times in the ensuing 60 years [1949 1965 1977 and1991] its structure was little altered Based upon rst-handobservations of workplaces Department of Labor examinersmdashusing guidelines supplied by the Handbook for Analyzing Jobs[U S Department of Labor 1972]mdashevaluate more than 12000highly detailed occupations along 44 objective and subjectivedimensions including training times physical demands and re-quired worker aptitudes temperaments and interests11
Our DOT data are based on an aggregation of these detailedoccupations into three-digit Census Occupation Codes (COC) ofwhich there are approximately 450 We append DOT occupationcharacteristics to the Census Integrated Public Micro Samples[IPUMS Ruggles and Sobeck 1997] one percent extracts for 19601970 1980 and 1990 and to CPS Merged Outgoing RotationGroup (MORG) les for 1980 1990 and 1998 We use all obser-vations for noninstitutionalized employed workers ages 18 to 64For the industry analysis these individual worker observationsare aggregated to the level of 140 consistent Census industriesspanning all sectors of the economy in each year of the sample Allanalyses are performed using full-time equivalent hours (FTEs)of labor supply as weights The latter is the product of the indi-vidual Census or CPS sampling weight times hours of work inthe sample reference week and for Census samples weeks ofwork in the previous year
We exploit two sources of variation for measuring changingjob task requirements The rst consists of changes over time inthe occupational distribution of employment holding constanttask content within occupations at the DOT 1977 level We referto cross-occupation employment changes as ldquoextensiverdquo marginshifts which we can measure consistently over the period 1960 to1998 This variation does not however account for changes intask content within occupations [Levy and Murnane 1996] whichwe label the ldquointensiverdquo margin To measure intensive marginshifts we analyze changes in task content measures within oc-cupations over the period 1977 to 1991 using occupationsmatched between the Fourth Edition and Revised Fourth Editionof the DOT
Although the DOT contains the best time-series on job task
11 The Department of Laborrsquos recent successor to the DOT ONET providespotentially more up-to-date information but is not suitable for time-seriesanalysis
1292 QUARTERLY JOURNAL OF ECONOMICS
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
requirements for detailed occupations in the United States it alsohas well-known limitations [Miller et al 1980] These includelimited sampling of occupations (particularly in the service sec-tor) imprecise denitions of measured constructs and omissionof important job skills These shortcomings are likely to reducethe precision of our analysis12
IIA Selecting Measures of Routine and Nonroutine Tasks
To identify variables that best approximate our task con-structs we reduced the DOT variables to a relevant subset usingDOT textual denitions means of DOT measures by occupationfrom 1970 and detailed examples of DOT evaluations from theHandbook for Analyzing Jobs We selected two variables to mea-sure nonroutine cognitive tasks one to capture interactive com-munication and managerial skills and the other to capture ana-lytic reasoning skills The rst codes the extent to which occupa-tions involve Direction Control and Planning of activities (DCP)It takes on consistently high values in occupations requiringmanagerial and interpersonal tasks The second variable GED-MATH measures quantitative reasoning requirements For rou-tine cognitive tasks we employ the variable STS which measuresadaptability to work requiring Set limits Tolerances or Stan-dards As a measure of routine manual activity we selected thevariable FINGDEX an abbreviation for Finger Dexterity Tomeasure nonroutine manual task requirements we employ thevariable EYEHAND an abbreviation for Eye-Hand-Foot coordi-nation Denitions and example tasks from the Handbook forAnalyzing Jobs are provided in Appendix 1
A limitation of the DOT variables is that they do not have anatural scale and moreover cannot condently be treated ascardinal To address this limitation we transformed the DOTmeasures into percentile values corresponding to their rank inthe 1960 distribution of task input We choose 1960 as the baseperiod for this standardization because it should primarily reectthe distribution of tasks prior to the computer era Consequentlyall of our outcome measures may be interpreted as levels or
12 Researchers who have used the DOT to analyze changing job skill re-quirements include Rumberger [1981] Spenner [1983 1990] Howell and Wolff[1991] Wolff [1996 2002] Handel [2000] and Ingram and Neumann [2000] Whatis unique to our work is the focus on routine and nonroutine tasks and the jointanalyses of the effect of computerization on task changes between and withinoccupations
1293SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
changes in task input relative to the 1960 task distributionmeasured in ldquocentilesrdquo13
IIB A Predictive Test
As an initial check on our data and conceptual framework wetest the rst proposition of our theoretical model industries his-torically intensive in routine tasks should have adopted computercapital relatively rapidly as its price fell To operationalize thistest we form an index of industry-level routine task intensityduring the precomputer era Using the 1960 Census data pairedto our selected DOT task measures we calculate the percentageshare of routine task input in industry jrsquos total task input asRoutine Task Sharej 1960 5 100 3 rj 1960 (r j 1960 1 n j 1960)where all task measures are standardized with equal mean andvariance The numerator of this index is the sum of industryroutine cognitive and routine manual task inputs while the de-nominator is the sum of all ve task inputs routine cognitive andmanual nonroutine analytic interactive and manual This in-dex which has mean 400 and standard deviation 50 shouldroughly correspond to (1 2 b j) in our model
To proxy computer adoption after 1960 we use the CurrentPopulation Survey to calculate industriesrsquo percentile rank of com-puter use in 1997 Although we do not have a measure of industrycomputer use in 1960 this was likely close to zero in all casesConsequently the 1997 measure should closely reect post-1960computer adoption
We t the following equation
(12) Computer adoptionj1960ndash1997 5
22456~1918
1 185~048
3 Routine Task Sharej1960
~n 5 140 R2 5 010
The point estimate of 185 (standard error 048) for the routinetask share variable conrms that an industryrsquos routine task in-tensity in 1960 is strongly predictive of its subsequent computeradoption Comparing two industries that in 1960 were 10 per-centage points (2 standard deviations) apart in routine task in-put the model predicts that by 1997 these industries would be 19percentage points apart in the distribution of computer adop-
13 An earlier version of this paper [Autor Levy and Murnane 2001] em-ployed raw DOT scores rather than the percentile measures used here Resultswere qualitatively identical
1294 QUARTERLY JOURNAL OF ECONOMICS
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
tionmdashapproximately 13 percentage points apart in on-the-jobcomputer use
We have estimated many variations of this basic model toverify its robustness including specifying the dependent variableas the level or percentile rank of industry computer use in 19841997 or the average of both scaling the routine task sharemeasure in percentiles of the 1960 task distribution calculatingthe routine task share index using task percentiles rather thantask levels and replacing the routine task index with its loga-rithm These many tests provide robust support for the rstproposition of our theoretical model demand for computer capitalis greatest in industries that were historically routine taskintensive
III TRENDS IN JOB TASK INPUT 1960 ndash1998
Our model implies that the rapidly declining price of com-puter capital should have reduced aggregate demand for laborinput of routine tasks and increased demand for labor input ofnonroutine cognitive tasks This section analyzes the evidence forsuch shifts
IIIA Aggregate Trends
Figure I illustrates the extent to which changes in the occu-pational distribution over the period 1960 to 1998 resulted inchanges in the tasks performed by the U S labor force Thisgure is constructed by pairing the selected DOT 1977 taskmeasures with Census and CPS employment data for each de-cade By construction each task variable has a mean of 50 cen-tiles in 1960 Subsequent points depict the employment-weightedmean of each assigned percentile over each decade14
As is evident in the gure the share of the labor forceemployed in occupations that made intensive use of nonroutineanalytic and nonroutine interactive tasks increased substantiallyduring the last four decades Although both of these measures ofnonroutine tasks increased in the 1960smdashthat is during the
14 We do not impose an adding-up constraint across task measuresmdashwhereby total task allocation must sum to one within jobs or time periodsmdashsincethis structure is not intrinsic to the DOT It is therefore possible for the econ-omywide average of total task input to either rise or fall This over-time variationis modest in practice The mean of all ve task measures equal to 50 by construc-tion in 1960 rose slightly to 525 in 1980 and fell to 512 in 1998
1295SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
precomputer eramdashthe upward trend in each accelerated thereaf-ter By 1998 nonroutine analytic task input averaged 68 centilesabove its 1970 level and nonroutine interactive input averaged115 centiles above its 1970 level
By contrast the share of the labor force employed in occupa-tions intensive in routine cognitive and routine manual tasksdeclined substantially Between 1970 and 1998 routine cognitivetasks declined 87 centiles and routine manual tasks declined by43 centiles Notably these declines reversed an upward trend inboth forms of routine task input during the 1960s For routinecognitive tasks this trend reversed in the 1970s and for routinemanual tasks the trend halted in the 1970s and reversed in the1980s
FIGURE ITrends in Routine and Nonroutine Task Input 1960 to 1998
Figure I is constructed using Dictionary of Occupational Titles [1977] taskmeasures by gender and occupation paired to employment data for 1960 and 1970Census and 1980 1990 and 1998 Current Population Survey (CPS) samplesData are aggregated to 1120 industry-gender-education cells by year and eachcell is assigned a value corresponding to its rank in the 1960 distribution of taskinput (calculated across the 1120 1960 task cells) Plotted values depict theemployment-weighted mean of each assigned percentile in the indicated year SeeTable I and Appendix 1 for denitions and examples of task variables
1296 QUARTERLY JOURNAL OF ECONOMICS
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
Finally the share of the labor force employed in occupationsintensive in nonroutine manual tasks showed a secular declineacross all decades This decline was most rapid in the 1960s andslowed considerably in subsequent decades
Panel A of Table II provides the corresponding means of taskinput by decade both in aggregate and by gender For both malesand females there are pronounced shifts against routine cogni-tive routine manual and nonroutine manual task inputs andpronounced shifts favoring nonroutine analytic and interactiveinputs These shifts however are numerically larger for femalesGiven the rapid entry of women into the labor force in recentdecades it appears plausible that demand shifts for workplacetasks would impact the stock of job tasks more rapidly for femalesthan males [Goldin 1990 Weinberg 2000 Blau Ferber and Wink-ler 2002 Chapter 4] To assess the importance of these genderdifferences we estimated all of our main results separately formales and females Because we found quite similar results forboth genders we focus below on the pooled gender samples
To complete the picture provided by the decadal meansFigure II depicts smoothed changes in the density of the tworoutine and two nonroutine cognitive task measures between1960 and subsequent decades Three series are plotted for eachtask measure Two depict extensive margin task shifts at approxi-mately twenty-year intervals These are measured using the 1977DOT task measures paired to the 1960 1980 and 1998 employ-ment data The third series adds intensive margin task shifts bypairing the 1991 DOT task measures with the 1998 employmentdata By construction task input is uniformly distributed acrossall percentiles in 1960 Hence the height of each line in the gurerepresents the difference in the share of overall employment in1980 or 1998 at each centile of 1960 task input15 To conservespace we do not provide a plot of the nonroutine manual mea-sure since it is not the subject of subsequent analysis
As shown in panels A and B of the gure the distribution ofnonroutine analytic and nonroutine interactive task input shiftedmarkedly rightward after 1960 In particular there was substan-tial growth in the share of employment requiring nonroutine taskinput above the 1960 median and a corresponding decline below
15 We apply an Epanechnikov kernel with bandwidth h 5 090sn2 1 5 where n is the number of observations and s is the standard deviation For oursamples this yields bandwidths between 5 and 7 centiles
1297SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EII
ME
AN
SO
FT
AS
KIN
PU
TB
YD
EC
AD
EA
ND
DE
CO
MP
OS
ITIO
NIN
TO
WIT
HIN
AN
DB
ET
WE
EN
IND
US
TR
YC
OM
PO
NE
NT
S1
960
ndash199
8
A
Wei
ghte
dm
ean
sof
econ
omyw
ide
task
inpu
tby
deca
de(p
erce
nti
les
of19
60ta
skdi
stri
buti
on)
1N
onro
utin
ean
alyt
ic2
Non
rout
ine
inte
ract
ive
3R
outi
neco
gnit
ive
4R
outi
ne
man
ual
5N
onro
uti
ne
man
ual
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
emA
llM
ale
Fem
All
Mal
eF
em
1960
(Cen
sus)
500
546
370
500
584
264
500
486
538
500
432
692
500
565
316
1970
(Cen
sus)
525
575
415
511
617
275
522
498
575
540
449
744
469
541
308
1970
(Cen
sus)
519
564
418
507
608
281
531
508
582
535
447
733
462
532
307
1980
(Cen
sus)
549
582
492
554
633
418
529
500
580
551
446
733
440
532
281
1980
(CP
S)
532
566
479
533
614
404
518
500
548
538
433
704
444
550
275
1990
(CP
S)
562
574
546
586
627
530
483
481
486
523
428
655
418
531
263
1998
(CP
S)
587
593
580
622
639
599
444
466
416
492
415
593
413
526
265
B
Dec
ompo
siti
onof
task
shif
tsin
tobe
twee
nan
dw
ith
inin
dust
ryco
mpo
nen
tsfo
rco
mbi
ned
gen
ders
(10
3an
nu
alch
ange
sin
mea
nta
skpe
rcen
tile
)
Tot
alB
twn
Wth
nT
otal
Btw
nW
thn
Tot
alB
wtn
Wth
nT
otal
Btw
nW
thn
Tot
alB
twn
Wth
n
1960
ndash197
02
571
740
831
152
034
149
220
114
106
401
239
162
23
032
228
20
7419
70ndash1
980
302
154
148
468
026
442
20
140
332
047
163
079
084
22
252
100
21
2519
80ndash1
990
297
092
205
531
052
479
23
482
142
22
072
147
20
162
131
22
582
127
21
3119
90ndash1
998
312
067
245
448
054
394
24
882
131
23
572
388
20
382
350
20
632
031
20
31
Sou
rces
Dic
tion
ary
ofO
ccup
atio
nal
Tit
les
[197
7]a
ndal
lem
ploy
edw
orke
rsag
es18
ndash64
Cen
sus
IPU
MS
1960
197
019
80C
PS
MO
RG
1980
199
0an
d19
98S
ampl
esu
sed
for
deca
dal
chan
ges
inpa
nel
Bar
e19
60ndash1
970
1960
and
1970
Cen
sus
1970
ndash198
019
70an
d19
80C
ensu
s19
80ndash1
990
1980
and
1990
CP
SM
OR
G
1990
ndash199
819
90an
d19
98C
PS
MO
RG
Tw
oC
ensu
s19
70sa
mpl
esar
eus
edin
pane
lsA
and
Bo
ne
code
dfo
rco
nsi
sten
cyw
ith
the
1960
Cen
sus
occu
pati
onco
des
and
ase
con
dco
ded
for
cons
iste
ncy
wit
hth
e19
80C
ensu
soc
cupa
tion
code
sD
ata
are
aggr
egat
edto
1120
indu
stry
-gen
der-
educ
atio
nce
lls
byye
aran
dea
chce
llis
assi
gned
ava
lue
corr
espo
ndi
ngto
its
ran
kin
the
1960
dist
ribu
tion
ofta
skin
put
(cal
cula
ted
acro
ssth
e11
201
960
task
cell
s)P
anel
Aco
ntai
nsth
eem
ploy
men
t-w
eigh
ted
mea
nof
each
assi
gned
perc
enti
lein
the
indi
cate
dye
arP
anel
Bpr
esen
tsa
deco
mpo
siti
onof
the
aggr
egat
ech
ange
inta
skin
put
over
the
indi
cate
dye
ars
into
wit
hin
and
betw
een
indu
stry
com
pon
ents
for
140
cons
iste
ntC
ensu
sIn
dust
ryC
ode
(CIC
)ind
ustr
ies
(59
inm
anuf
actu
rin
g81
inno
nman
ufa
ctur
ing)
See
Tab
leI
and
App
endi
x1
for
den
itio
nsan
dex
ampl
esof
task
vari
able
s
1298 QUARTERLY JOURNAL OF ECONOMICS
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
the 1960 median These shifts are visible from 1960 to 1980 andbecome even more pronounced by 1998 Adding variation alongthe intensive margin augments the rightward shift particularlyfor the nonroutine interactive measure
Panels C and D of Figure II plot the corresponding densitiesfor routine cognitive and routine manual task input Consistentwith the theoretical model the distribution of labor input of bothroutine cognitive and routine manual tasks shifted sharply left-ward after 1960 mdashopposite to the case for nonroutine tasks Theshift is particularly pronounced for the routine cognitive taskmeasure and becomes even more apparent when the intensivemargin is added As suggested by Figure I the decline in theinput of routine manual tasks is less dramatic though stillvisible
In sum the evidence in Figures I and II supports our modelrsquosprimary macroeconomic implications Between 1970 and 1998there were secular declines in labor input of routine cognitive androutine manual tasks and corresponding increases in labor inputof nonroutine analytic and interactive tasks We next analyze thesources of these task shifts at the industry level
IIIB Task Changes within and between Industries
The changes in economywide labor input of routine and non-routine tasks documented in Figure I and Table II could stemfrom substitution of computer capital for routine labor inputswithin detailed industries as our model suggests Alternativelythey could stem from changes in the composition of nal demandSince much of our detailed analysis focuses on changes in taskinput at the industry level we explore briey the extent to whichchanges in job content are due to within-industry task shifts
Panel B of Table II presents a standard decomposition of taskchanges into within- and between-industry components16 Thisdecomposition shows quite consistent patterns of task change
16 We decompose the use of task k in aggregate employment between years tand t (DTkt 5 Tkt 2 Tkt) into a term reecting the reallocation of employment acrosssectors and a term reecting changes in task j input within industries usingthe equation DTkt 5 yenj(DEjtgjk) 1 yenj(Dg jktE j) 5 DTkt
b 1 DTktw where j indexes
industries Ejkt is the employment of workers in task k in industry j inyear t as a share of aggregate employment in year t Ejt is total employment(in FTEs) in industry j in year t gjkt is the mean of task k in industry j inyear t gj k 5 (gjkt 1 g jk t) 2 and E j 5 (Ejt 1 E jt) 2 The rst term (DTkt
b ) reectsthe change in aggregate employment of task k attributable to changesin employment shares between industries that utilize different intensities oftask k The second term (DTkt
w ) reects within-industry task change
1299SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
FIGURE IISmoothed Differences between the Density of Nonroutine Task Input
in 1960 and Subsequent YearsFigure II is constructed using Dictionary of Occupational Titles (DOT) task
measures by gender and occupation paired to employment data from 1960 1980and 1998 Census and Current Population Survey samples Plots depict the changein the share of employment between 1960 and the indicated year at each 1960percentile of task input All series use DOT 1977 data paired to employment datafor the indicated year except for series marked ldquo1991 task measuresrdquo which usetask data from 1991 DOT See Table I and Appendix 1 for denitions andexamples of task variables
1300 QUARTERLY JOURNAL OF ECONOMICS
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
Both the nonroutine analytic and nonroutine interactive taskmeasures show strong within-industry growth in each decadefollowing the 1960s Moreover the rate of within-industry growthof each input increases in each subsequent decade Although asnoted above nonroutine analytic input also increased during the1960s Table II shows that this was primarily a cross-industry
1301SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
phenomenonmdashie due to sectoral shifts After the 1960s bycontrast the growth in nonroutine task input was dominated bywithin-industry task shifts
Trends in both routine cognitive and routine manual tasksshow a similarly striking pattern Both types of routine taskinput increased during the 1960s due to a combination of be-tween- and within-industry shifts In the decades following how-ever input of both routine cognitive and routine manual taskssharply declined and the bulk of these declines was due to with-in-industry shifts Moreover the rate of within-industry declineincreased in each subsequent decade
As distinct from the other four task measures we observesteady within- and between-industry shifts against nonroutinemanual tasks for the entire four decades of our sample Since ourconceptual framework indicates that nonroutine manual tasksare largely orthogonal to computerization we view this pattern asneither supportive nor at odds with our model
In summary the trends against routine cognitive and man-ual tasks and favoring nonroutine cognitive tasks that we seek toanalyze are dominated by within-industry shifts particularlyfrom the 1970s forward We next analyze whether computeriza-tion can explain these task shifts17 Because our model makes noprediction for how computerizing industries will adjust demandfor nonroutine manual tasks we do not include this variable inour industry-level analysis below (see Autor Levy and Murnane[2001] for detailed analysis)
IV COMPUTERIZATION AND TASK CHANGEINDUSTRY LEVEL RELATIONSHIPS
As industries adopt computer technology our model predictsthat they will simultaneously reduce labor input of routine cog-nitive and manual tasks and increase labor input of nonroutinecognitive tasks We test these hypotheses below
17 Our model also implies that the expenditure shares of routine-task-intensive industries should have increased as r declined By contrast the predic-tion for the employment share of routine-task-intensive industries is ambiguoussince these industries should have differentially substituted computer capital forlabor input Because our data measure employment not expenditures we areunable to test the implication for expenditure shares Closely related computer-intensive industries should have experienced relatively larger gains in laborproductivity as r declined Stiroh [2002] presents evidence that this occurred inthe 1990s
1302 QUARTERLY JOURNAL OF ECONOMICS
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
IVA Industry Computerization and Task Trendsover Four Decades
We begin by estimating a model for the within-industry re-lationship between computer adoption and task change over fourdecades Specically we t the equation
(13) DT jkt 5 a 1 fDC j 1 ejkt
where DTjk t 5 Tjk t 2 Tjk t is the change in industry jrsquos input oftask k between years t and t and DCj is the annual change in thepercentage of industry workers using a computer at their jobsbetween 1984 and 1997 (estimated from the October CurrentPopulation Survey supplements) While our model predicts thatcomputer adoption should be highly correlated with industry taskchange during the computer era we would not anticipate a simi-lar relationship for the precomputer era of the 1960s Accord-ingly we estimate equation (13) separately for each of the fourdecades in our sample This allows us to conrm that the rela-tionship between computerization and industry task change inthe 1970sndash1990s does not reect trends that predate the com-puter era18
Estimates of equation (13) for each of the four task measuresare given in Table III The rst two panels show that during the1970s 1980s and 1990s rapidly computerizing industries raisedtheir input of nonroutine analytic and interactive tasks signi-cantly more than others For example the point estimate of 1204(standard error of 474) in column 1 of panel A indicates thatbetween 1990 and 1998 a 10 percentage point increase in com-puter use was associated with a 12 centile annualized increase inlabor input of nonroutine analytic tasks To evaluate the magni-tude of this relationship note that the mean annualized industrylevel rise in nonroutine analytic task input during 1990 ndash1998tabulated immediately below the regression estimate was 25percentiles By comparison the intercept of the bivariate regres-sion for this period is 01 Hence almost the entirety of theobserved within-industry change in nonroutine analytic input isldquoexplainedrdquo by the computerization measure Similar compari-sons conrm that the relationship between industry computer-ization and rising input of nonroutine interactive and analytic
18 As noted by Autor Katz and Krueger [1998] and Bresnahan [1999] theera of rapid computer investment began in the 1970s Desktop computing becamewidespread in the 1980s and 1990s
1303SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
tasks is economically large and statistically signicant at conven-tional levels in each of the three most recent decades
Panels C and D of the table provide analogous estimates forthe two routine task measures As predicted by the conceptualmodel the relationships between industry computerization andchanges in routine task input are uniformly negative in the1970s 1980s and 1990s These relationships are also economi-
TABLE IIICOMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 1204 1402 911 7491984ndash1997 (474) (497) (417) (528)Intercept 007 2066 2026 2055
(100) (103) (086) (105)R2 004 005 003 001Weighted mean D 245 205 148 083
B D Nonroutineinteractive
D Computer use 1478 1721 1081 7551984ndash1997 (548) (632) (571) (664)Intercept 102 146 235 010
(115) (131) (117) (132)R2 005 005 003 001Weighted mean D 394 479 442 149
C D Routinecognitive
D Computer use 21757 21394 21100 23901984ndash1997 (554) (572) (540) (448)Intercept 2011 063 163 178
(117) (119) (111) (089)R2 007 004 003 001Weighted mean D 2357 2207 2047 106
D D Routinemanual
D Computer use 22472 2594 2656 4151984ndash1997 (577) (564) (484) (350)Intercept 138 2016 209 085
(122) (117) (099) (070)R2 012 001 001 001Weighted mean D 2350 2131 084 162
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See Table I and Appendix 1 for denitions and examples of taskvariables
1304 QUARTERLY JOURNAL OF ECONOMICS
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
cally large and in most cases statistically signicant For exam-ple the computerization measure explains the entirety of thewithin-industry decline in routine task input during the 1990sand more than explains this decline in the 1980s and 1970s
A notable pattern for all four task measures is that therelationship between computerization and industry task changetends to become larger in absolute magnitude with each passingdecade This suggests a secularly rising relationship betweencomputerization and task change The nal column of Table IIItests for this rise by estimating equation (13) for the 1960s adecade during which computerization is unlikely to have stronglyinuenced task demands Reassuringly there are no signicantrelationships between computerization and task change in thisdecade And in one case the coefcient is of the opposite sign asin later decades Hence these estimates suggest that the rela-tionship between industry task shifts and computer adoptioneither commenced or substantially accelerated during the com-puter era and not before19
IVB Using Composite DOT Variables
Though we view the selected task measures as the mostappropriate available from the DOT we are sensitive to theconcern that the choice of variables could be viewed as arbitraryOne way to test their appropriateness is to use alternative com-posite variables We used principal components analyses (PCA) topool variation from each selected DOT task measure with severalother plausible alternatives and estimated equation (13) usingthese composites20 The details of our compositing exercise areprovided in the Data Appendix and the results of the compositeestimation are found in Appendix 2 A limitation of this exerciseis that the variables used in the composites do not in our view
19 We also estimated the models in Table III separately by gender and formanufacturing and nonmanufacturing sectors The pattern of results is similar inall cases For both genders computer investment is a signicant predictor ofreductions in routine labor input of cognitive and manual tasks and increases innonroutine analytic input For females the relationship between computerizationand nonroutine interactive tasks is positive but insignicant The magnitude ofthe relationship between computerization and nonroutine tasks is somewhatlarger in manufacturing than nonmanufacturing and the reverse is true forroutine tasks Further details are available from the authors
20 The PCA extracts eigenvectors that maximize common variation amongselected measures each of which is standardized with mean zero and varianceone subject to the constraint that the sum of squared weights in the eigenvectorequals one It can be shown that if measurement error in the selected variables isclassical (ie white noise) the PCA extracts maximal nonerror variation
1305SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
correspond as closely to the intended construct as our primarymeasures
As visible in the table the qualitative trends in the compositerelationships are comparable to those using our preferred mea-sures in Table III In particular industry computerization isassociated with sharp declines in routine cognitive and manuallabor inputs and growth in nonroutine analytic and interactivelabor inputs Moreover these relationships typically becomestronger in successive decades Contrary to expectations how-ever the composite measure for routine cognitive input is onlysignicant in the most recent decade and the composite measurefor nonroutine interactive input is statistically signicant in the1960s Thus while our results are generally robust to variablechoice this exercise underscores that variable choice does matterA data source specically designed to measure changes in work-place input of routine and nonroutine cognitive tasks over a longtime horizon would clearly provide a more complete test of themodel Given the absence of such a data source for the UnitedStates we view the evidence provided by the DOT as uniquelyinformative21
IVC Employing Contemporaneous Measures of Computerand Capital Investment
A limitation of the CPS computer measure used so far is thatit is only available for the 1980s and 1990s To provide morecomprehensive measures of computer and capital investmentavailable for the entire 1959 ndash1998 period we draw on the Na-tional Income and Product Accounts (NIPA) which provides de-tailed data on capital stocks across 42 major industries excludinggovernment [U S Department of Commerce 2002a 2002b] As ameasure of industry computerization we calculated the log of realinvestment in computer hardware software and peripherals perfull-time equivalent employee (FTE) over the course of each de-cade To distinguish the relationship between task change andcomputerization from overall capital-skill complementarity[Griliches 1969] we construct two variables to control for capital
21 Spitz [2003] studies the predictions of our task model using German datafrom 1979ndash1999 which contains far more detailed and precise information onworkplace tasks than is available from the DOT Consistent with the predictionsof the model Spitz reports that computer capital substitutes for repetitive manualand repetitive cognitive skills and complements analytical and interactive skillsSee also Bartel Ichniowski and Shaw [2000] and Ichniowski and Shaw [2003] forquantitative and case study evidence
1306 QUARTERLY JOURNAL OF ECONOMICS
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
deepening the log of capital investment ow per worker and thelog capital to labor ratio
Using these data we t stacked rst-difference industry taskshift models of the form
(14) DT jkt 5 a 1 d70ndash80 1 d80ndash90 1 d90ndash98 1 wCI jt 1 uKIjt 1 ejkt
where CI jt is industry jrsquos real log investment in computer capitalper FTE over the contemporaneous decade in industry KIjt is theanalogous measure for real capital investment the drsquos are timedummies equal to one in each of the decades post-1960 corre-sponding to their subscripts and a is a common intercept In thisequation the drsquos measure the trend change in industry task inputin the 1970s 1980s and 1990s relative to the base period of the1960s
Estimates of equation (14) are found in Table IV Two sets ofHuber-White standard errors are tabulated for each model Thosein parentheses account for the fact that the NIPA capital mea-sures are observed at a more aggregate level than the dependentvariables measured from the CPS and Census (42 sectors versus123 sectors for this exercise) The standard errors in bracketsadditionally account for potential serial correlation in industrytask changes over succeeding decades (cf Bertrand Duo andMullainathan [2004])
As is visible in the table the NIPA measure of computerinvestment consistently predicts relative declines in industry in-put of both routine cognitive and manual tasks and growth ininput of nonroutine analytic and interactive tasks How large arethese relationships We can gauge the modelrsquos explanatory powerby comparing the magnitude of the estimated drsquos conditional oncomputer investment with the unconditional within-industrytrends in task input observed for each decade To facilitate thiscomparison the bottom panel of Table IV tabulates the uncondi-tional decadal trends As with the Table III estimates we ndthat industries making relatively greater investments in com-puter capital are responsible for the bulk of the observed substi-tution away from routine cognitive and manual tasks and towardnonroutine analytic and interactive tasks Holding computer in-vestment constant we can explain more than 100 percent of theoverall trend increase in nonroutine cognitiveanalytic task inputa substantial part of the trend increase in nonroutine cognitiveinteractive input and substantial parts of the trend decreases inroutine cognitive and routine manual inputs
1307SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
TABLE IVCOMPUTER INVESTMENT CAPITAL INTENSITY AND TASK INPUT IN THREE-DIGIT
INDUSTRIES 1960ndash1998 STACKED FIRST-DIFFERENCE ESTIMATES
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
A DNonroutine
analytic
B DNonroutineinteractive
C DRoutinecognitive
D DRoutinemanual
(1) (2) (1) (2) (1) (2) (1) (2)
Log(ClL) 665 676 1159 1003 2827 2830 2911 2820(413) (397) (321) (331) (363) (326) (257) (229)[636] [590] [397] [450] [474] [376] [327] [286]
Log(KlL) 122 2341 2293 2242(436) (458) (476) (395)[636] [445] [731] [587]
D Log(KL) 024 301 2132 2389(235) (224) (212) (192)[236] [219] [218] [238]
1970ndash1980 dummy 2064 2054 138 249 2032 2084 068 2080(108) (129) (150) (162) (131) (158) (096) (117)[102] [123] [207] [212] [113] [093] [094] [103]
1980ndash1990 dummy 2034 2025 058 183 2162 2214 2132 2290(157) (160) (181) (170) (156) (186) (111) (138)[143] [167] [158] [146] [133] [135] [086] [107]
1990ndash1998 dummy 2119 2113 2191 2090 2133 2171 2115 2236(155) (162) (183) (170) (166) (164) (132) (147)[177] [193] [185] [188] [193] [163] [095] [113]
Intercept 889 823 1240 1130 2929 2709 2962 2555(408) (442) (425) (359) (412) (363) (314) (267)[545] [638] [476] [480] [448] [436] [375] [330]
R2 006 006 011 012 014 014 020 021
Weighted mean of dependent variable
1960ndash1970 116 174 130 1631970ndash1980 123 459 2020 0981980ndash1990 207 469 2205 21741990ndash1998 215 376 2303 2282
n 5 492 Robust standard errors in parentheses are heteroskedasticity consistent and account forclustering of errors within 42 consistent NIPA sectors in each decade (168 clusters) Standard errors inbrackets additionally account for potential serial correlation within sectors (42 clusters) Each columnpresents a separate OLS regression of ten times annual industry changes in task input on the indicatedcovariates Sample is 123 consistent CIC industries with four observations per industry 1960ndash1970 and1970ndash1980 changes use Census IPUMS samples and 1980ndash1990 and 1990ndash1998 use CPS MORG samplesEstimates are weighted by mean industry share of total employment (in FTEs) over the endpoints of the yearsused to form the dependent variable All capital measures are in millions of real 1996 dollars
Log(ClL) and Log(KlL) are respectively one-tenth the log of annual computer investment per FTE andtotal capital investment per FTE between the two end years used to form the dependent variable Means ofLog(ClL) are 2108 2095 2087 and 2073 in 1960ndash1970 1970ndash1980 1980ndash1990 and 1990ndash1998respectively Means of Log(Kl L) are 2057 2054 2054 and 2052 in the corresponding years
D Log(KL) is ten times the annual change in log capitalFTE over the two end years used to form thedependent variable Means are 043 010 010 and 024 in the corresponding years
1308 QUARTERLY JOURNAL OF ECONOMICS
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
A notable pattern in Table IV is that the coefcients on thecapital investment and capital intensity variables are economi-cally small and in most cases insignicant This indicates thataggregate capital deepeningmdashapart from computer investmentmdashexplains little of the observed change in task input22 We havealso explored a large number of variations of these models includ-ing estimating separate models for manufacturing and nonmanu-facturing industries and for male and female task input subdi-viding capital investment into equipment and structures invest-ment subdividing computer investment into hardware andsoftware investment and controlling for industriesrsquo log outputcapital-output ratios and import and export penetration Thesetests conrm the overall robustness of our results
V TASK CHANGE WITHIN EDUCATION GROUPS AND OCCUPATIONS
The analyses above establish that industries undergoingrapid computerization reduced labor input of routine cognitiveand manual tasks and increased labor input of nonroutine inter-active and analytic tasks Since better educated workers arelikely to hold a comparative advantage in nonroutine versusroutine tasks one interpretation of these results is that theyconrm the established pattern of increasing relative educationalintensity in computerizing industries over the past several de-cades While we do not disagree with this interpretation our taskframework makes a broader claim namely that changes in thedemand for workplace tasks stemming from technologicalchange are an underlying causemdashnot merely a reectionmdash ofrelative demand shifts favoring educated labor To test thisbroader implication we exploit the unique features of the DOT toanalyze two novel margins of task change changes within edu-cation groups and changes within occupations
VA Within-Industry Task Shifts by Education Group1980 ndash1998
We showed in Tables III and IV that increased industrycomputerization predicts increased nonroutine cognitive activityand reduced routine cognitive and manual activities Why does
22 This pattern echoes the ndings of Berman Bound and Griliches [1994]Autor Katz and Krueger [1998] and Bresnahan Brynjolfsson and Hitt [2002]for skill upgrading
1309SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
this occur One possibility is that as industries purchase com-puter capital they hire better educated workers who specialize inthese tasks Alternatively industries may change the task as-signments of workers with given educational attainments reduc-ing their allocation to routine tasks and raising it to nonroutinetasks We explore the relative importance of these two channelsby estimating a variant of equation (13) for within-industry taskupgrading by education group Specically we estimate themodel
(15) DT ijkt 5 a i 1 f iDC j 1 eijkt
where the dependent variable is the within-industry change inthe mean of each DOT task measured in centiles of the 1960distribution among workers of the same educational attainmentIn this equation i indexes each of four education groupsmdashhighschool dropouts high school graduates some-college completersand college graduatesmdashand subscripts j k and t refer to indus-tries tasks and time periods as above We estimate this modelusing industry task data for 1980 ndash1998 to exploit the (almost)contemporaneous industry computer use data for 1984 ndash1997
To establish a baseline for comparison we initially estimateequation (15) for aggregate within-industry task changes over1980 ndash1998 (ie incorporating both between- and within-educa-tion group task shifts) Consistent with earlier ndings theseestimates in panel A of Table V show striking correlations be-tween industry computerization rising labor input of routinecognitive and manual tasks and declining labor input of nonrou-tine interactive and analytic tasks
Panels B through E of Table V present analogous modelsestimated separately for the four education groups Here mea-sured changes in task input stem solely from within-educationgroup shifts in occupational distributions within industriesThese estimates reveal that industry-level computerization isstrongly predictive of shifts toward nonroutine and against rou-tine tasks within essentially all education groups For the twogroups at the middle of the education distributionmdash high schoolgraduates and those with some collegemdashchanging employmentpatterns within rapidly computerizing sectors entirely accountfor observed task shifts More precisely holding computer adop-tion xed our estimates would not predict any signicant within-industry task change for either education group
For the education groups at the bottom and top of the distri-
1310 QUARTERLY JOURNAL OF ECONOMICS
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
TABLE VCOMPUTERIZATION AND INDUSTRY TASK INPUT 1980ndash1998
OVERALL AND BY EDUCATION GROUP
DEPENDENT VARIABLE 10 3 ANNUAL CHANGE IN QUANTILES OF TASK MEASUREMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 D Nonroutineanalytic
2 D Nonroutineinteractive
3 D Routinecognitive
4 D Routinemanual
A Aggregate within-industry change
D Computer use 1295 1597 21584 214321984ndash1997 (368) (432) (473) (473)Intercept 2033 127 038 054
(077) (090) (099) (099)Weighted mean task D 220 439 2271 2225
B Within industry High school dropouts
D Computer use 464 1192 2264 28851984ndash1997 (607) (873) (795) (676)Intercept 2251 2439 002 111
(126) (182) (166) (141)Weighted mean task D 2161 2207 2049 2062
C Within industry High school graduates
D Computer use 004 1349 22818 225501984ndash1997 (417) (540) (613) (605)Intercept 2149 107 155 048
(087) (113) (128) (126)Weighted mean task D 2148 370 2395 2449
D Within industry Some college
D Computer use 795 1814 21568 217771984ndash1997 (503) (554) (527) (561)Intercept 2188 2058 035 139
(105) (115) (110) (117)Weighted mean task D 2033 296 2271 2208
E Within industry College graduates
D Computer use 161 557 2078 24461984ndash1997 (342) (335) (485) (570)Intercept 025 010 2096 2012
(071) (070) (101) (119)Weighted mean task D 057 222 2148 2198
F Decomposition into within and between education group components
Explained task D 252 311 2309 2279Within educ groups () 237 779 917 1111Between educ groups () 763 221 83 2111
n in panels AndashE is 140 139 140 140 and 139 consistent CIC industries Standard errors are in parenthesesEach column of panels AndashE presents a separate OLS regression of ten times the annual change in industry-leveltask input for the relevant education group (measured in centiles of the 1960 task distribution) during 1980ndash1998on the annual percentage point change in industry computer use during 1984ndash1997 (weighted mean 0198) anda constant Estimates are weighted by mean industry share of total employment (in FTEs) in 1980 and 1988Industries with no employment in the relevant educational category in either 1980 or 1998 are excluded Datasources are CPS MORG 1980 and 1998 and DOT 77 job task measures The ldquoexplainedrdquo component in Panel F isthe within-industry change in the task measure predicted by computerizationin regression models in Panel A SeeTable I and Appendix 1 for denitions and examples of task variables
1311SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
butionmdashcollege graduates and high school dropoutsmdashsimilarpatterns prevail but they are less precisely estimated In allcases the estimates are of the expected sign but none is statis-tically signicant For college graduates this is likely to reectldquotopping outrdquo since this education group was already at theextreme of the distribution for all tasks We are less certain whythe relationships are weaker for high school dropouts but onepossibility is that this group has insufcient human capital to beeffectively redeployed to alternative job tasks
To assess whether these within-education group shifts are aquantitatively important component of the overall change in in-dustry task content panel F of Table V presents a decompositionof industry task changes into within and between educationgroup components This exercise shows that in every case within-education group task upgrading explains a substantial share 24to 111 percent of total task upgrading over these two decadesFor example the annual within-industry change in nonroutineinteractive tasks over the period 1980 ndash1998 is 44 centiles perdecade of which 31 centiles (71 percent) is accounted for bycontemporaneous industry computerization Within-educationgroup task changes explain the bulk of these shifts 78 percent ofthe explained component and 55 percent of the total Subdividingthe explained within-education group component further 59 per-cent is due to changes in task assignment among high schoolgraduates and those with some college and the rest is equallyaccounted for by task shifts among college graduates and highschool dropouts
This exercise demonstrates that within-education groupshifts in task content are the primary channel through which thestructure of workplace tasks has shifted over the past two de-cades Furthermore a large portion of the within-education groupchanges are accounted for by cross-industry patterns of computeradoption This suggests to us that task change is antecedent toeducational upgrading rather than merely a reection of it
VB Task Shifts within Occupations
The analyses above exploit shifts in occupational composi-tionmdashthe extensive marginmdashto quantify changes in task inputThis approach is imperfect since it assumes that the tasks per-formed within occupations are static which is unlikely to beaccurate over long time intervals Moreover our task frameworkimplies that this assumption should be violated in a specic
1312 QUARTERLY JOURNAL OF ECONOMICS
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
manner occupations undergoing rapid computerization shoulddifferentially reduce labor input of routine cognitive and manualtasks and increase labor input of nonroutine cognitive tasks Toprovide one example the 1976 edition of the Department ofLaborrsquos Occupation Outlook Handbook described the job of Sec-retary as ldquo Secretaries relieve their employers of routine du-ties so they can work on more important matters Although mostsecretaries type take shorthand and deal with callers the timespent on these duties varies in different types of organizationsrdquo[U S Department of Labor 1976 p 94] In 2000 the entry forSecretary reads ldquoAs technology continues to expand in ofcesacross the Nation the role of the secretary has greatly evolvedOfce automation and organizational restructuring have led sec-retaries to assume a wide range of new responsibilities oncereserved for managerial and professional staff Many secretariesnow provide training and orientation to new staff conduct re-search on the Internet and learn to operate new ofce technolo-giesrdquo [U S Department of Labor 2000 p 324]
To test whether this example captures a pervasive phenome-non we match occupations from the 1977 and 1991 revisions ofthe DOT to estimate the following equation
(16) DTmkt 5 a 1 jDCm 1 emkt
Here DTm k t is the change in occupational input of task k between1977 and 1991 in three-digit COC occupation m and DCm is thechange in occupational computer penetration measured by theCPS To provide a clean test our data set is constructed usingonly the subset of occupations appearing in the 1977 DOT whichwas used to create our original occupation crosswalk Accord-ingly the variation used to estimate equation (16) stems exclu-sively from DOT examinersrsquo reevaluation of the task content ofindividual occupations between 1977 and 199123
Table VI presents three estimates for each task measure Therst column of each panel presents a bivariate regression of thewithin-occupation change in task content on occupational com-puterization and a constant These estimates provide striking
23 The weighted fraction of employment reevaluated between 1978 and 1990in our data is 73 percent Occupations were chosen for reevaluation by DOTexaminers partly on the expectation that their content had changed Hence thisis not a random sample We assume that occupations that were not revisedbetween the 1977 and 1991 DOT experienced no task change Provided that theseoccupations did not experience offsetting shifts our approach will provide a lowerbound on the extent of task change
1313SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
TA
BL
EV
IC
OM
PU
TE
RIZ
AT
ION
AN
DC
HA
NG
ES
INJO
BT
AS
KC
ON
TE
NT
WIT
HIN
OC
CU
PA
TIO
NS
1977
ndash199
1D
EP
EN
DE
NT
VA
RIA
BL
E1
03
AN
NU
AL
WIT
HIN
-OC
CU
PA
TIO
NC
HA
NG
EIN
QU
AN
TIL
EO
FT
AS
KM
EA
SU
RE
M
EA
SU
RE
DIN
PE
RC
EN
TIL
ES
OF
1984
TA
SK
DIS
TR
IBU
TIO
N
AD
Non
rout
ine
anal
ytic
BD
Non
rou
tin
ein
tera
ctiv
eC
DR
outi
ne
cogn
itiv
eD
DR
outi
nem
anu
al
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
DC
ompu
ter
use
294
357
402
570
586
708
218
18
216
56
218
48
174
083
037
1984
ndash199
7(1
84)
(19
2)(2
06)
(18
8)(1
97)
(21
1)(3
29)
(34
1)(3
65)
(28
9)(3
01)
(32
3)D
Col
lege
grad
emp
24
792
483
24
472
458
225
922
76
216
07
216
03
1984
ndash199
7(5
54)
(55
4)(5
68)
(56
7)(9
86)
(98
5)(8
70)
(87
1)D
HS
grad
emp
283
309
20
190
5216
97
158
62
104
22
107
019
84ndash1
997
(37
8)(3
81)
(38
8)(3
90)
(67
3)(6
77)
(59
4)(5
99)
DF
emal
eem
p2
237
26
4710
14
247
1984
ndash199
7(3
94)
(40
3)(6
99)
(61
9)In
terc
ept
20
922
091
20
952
046
20
422
052
056
014
030
042
070
074
(04
0)(0
41)
(04
1)(0
41)
(04
2)(0
42)
(07
1)(0
72)
(07
3)(0
63)
(06
4)(0
64)
R2
001
001
001
002
002
003
006
008
008
000
001
001
Wei
ghte
dm
ean
D2
039
058
22
760
74
nis
470
cons
iste
ntth
ree-
digi
tC
ensu
sO
ccup
atio
nC
ode
(CO
C)
occu
pati
ons
Sta
ndar
der
rors
are
inpa
rent
hese
sE
ach
colu
mn
pres
ents
ase
para
teO
LS
regr
essi
onof
ten
tim
esth
ean
nual
chan
gein
the
occu
pati
onal
task
mea
sure
(mea
sure
din
cent
iles
ofth
e19
84di
stri
buti
on)b
etw
een
the
1977
and
1991
DO
Tre
visi
ons
Com
pute
rus
eco
lleg
egr
adua
teh
igh
scho
olgr
adua
tea
ndfe
mal
eem
ploy
men
tsha
res
are
mea
sure
das
ten
tim
esth
ean
nual
chan
gein
the
rele
van
tmea
sure
from
the
1984
and
1997
CP
SW
eigh
ted
mea
nsar
e0
183
001
72
001
5an
d0
017
resp
ecti
vely
O
mit
ted
edu
cati
onca
tego
ries
are
som
eco
lleg
ean
dh
igh
sch
ool
drop
out
Est
imat
esar
ew
eigh
ted
byth
eav
erag
eoc
cupa
tion
alsh
are
ofU
S
em
ploy
men
tin
1984
and
1997
Se
eT
able
Ian
dA
ppen
dix
1fo
rde
nit
ion
san
dex
ampl
esof
task
vari
able
s
1314 QUARTERLY JOURNAL OF ECONOMICS
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
conrmation of the predicted relationships between computeriza-tion and task change Occupations making relatively large in-creases in computer use saw relatively greater increases in laborinput of nonroutine cognitive analytic and interactive tasks andlarger declines in labor input of routine cognitive skills Each ofthese relationships is signicant at the 10 percent level orgreater and is of sizable magnitude for all three cognitive taskmeasures the computerization variable more than fully accountsfor the observed change in occupational task input Only in thecase of routine manual tasks where the point estimate is close tozero do we fail to nd the expected relationship
To examine whether intra-occupational task changes are im-plicitly captured by shifts in the educational and gender distri-bution of employees within an occupation we add controls for thecontemporaneous change in the percentage of workers in anoccupation who are college graduates high school graduates andfemales24 As is visible in specications 2 and 3 the relationshipbetween computerization and within-occupation task change issurprisingly insensitive to these controls In fact standard mea-sures of educational and gender composition are poor proxies forchanges in job tasks observed by DOT examiners In net thesendings demonstrate that shifts in job content away from routinetasks and toward nonroutine cognitive tasks are a pervasivefeature of the data and are concentrated in industries and occu-pations that adopted computer technology most rapidly
VI QUANTIFYING THE MAGNITUDE OF TASK STRUCTURE CHANGES
What is the economic signicance of the change in the tasksperformed by the U S labor force during the last three decadesThe answer is not immediately apparent since units of task inputdo not have a familiar scale To quantify task shifts in concreteeconomic terms we draw together task changes within indus-tries education groups and occupations to calculate their poten-tial contribution to the demand for college-educated labor during1970 to 1998 This analysis proceeds in three steps
We begin by estimating a ldquoxed coefcientsrdquo model of educa-
24 For consistency of measurement we employ CPS computerization edu-cation and gender means by occupation for 1984 to 1997 We cannot perform ananalogous exercise using the NIPA investment measures since they are notavailable for occupations
1315SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
tion requirements in industries and occupations as a function oftheir task inputs
(17) College Sharej 5 a 1 Ok51
4
pk z T jk 1 ej
In this equation College Sharej is the college-equivalent share ofemployment (in FTEs) in industry or occupation j at the midpointof our samples and the T j
k rsquos measure industry or occupation taskinput in centiles during the same period25 The coefcients pk obtained from (17) provide an estimate of college-equivalent labordemand as a function of industry or occupation task inputs Weestimate this model separately for industry and occupation taskdemands using employment data from our CPS samples from1980 and 1984 paired to the 1977 DOT job task measures26
We refer to equation (17) as a xed coefcients model becauseit neglects the impact of task prices on task demandsmdash or equiva-lently assumes that the elasticity of substitution between collegeand noncollege equivalent workers is zero This is an imperfectapproximation if the market price of nonroutine relative to rou-tine tasks has risen this calculation will understate demandshifts favoring nonroutine tasks and by implication collegegraduate employment
The second step of our methodology is to translate taskshifts into predicted changes in college employment We rstassemble changes in our four key task measures over 1970 to1998 DT1970ndash1998
k We then apply these task shift measures tothe ldquoxed coefcientsrdquo estimated from equation (17) to calculate
(18) DCollege Share1970ndash1998 5 Ok51
4
pk z DT 1970ndash1998k
Here DCollege Share1970 ndash1998 is the change in the college share ofaggregate employment predicted by task shifts over 1970 ndash1998
25 We follow Autor Katz and Krueger [1998] and Murphy Romer andRiddell [1998] in dening college equivalent workers as all those with a collegedegree or greater plus half of those with some college Results using exclusivelycollege graduates are quite similar
26 The industry task demand model is estimated using the 1980 MORGemployment data which is at the midpoint of our sample The occupation taskdemand model is estimated using the 1984 CPS sample which is at the midpointof our occupation sample For completeness estimates of equation (17) also controlfor input of nonroutine manual tasks Inclusion or exclusion of this covariate hasno substantive impact
1316 QUARTERLY JOURNAL OF ECONOMICS
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
The intuition for this calculation is that industries and occupa-tions that are high in nonroutine cognitive task input and low inroutine manual and routine cognitive task input employ collegegraduates relatively intensively Consequently secular increasein nonroutine cognitive task input and declines in nonroutinecognitive and manual task input over 1970 ndash1998 will cause equa-tion (18) to predict corresponding growth in the college graduateshare of aggregate employment
Table VII summarizes the changes in job task input due tocross-occupation (extensive margin) and within-occupation (inten-sive margin) shifts documented by our earlier analyses Panel Apresents observed economywide shifts in task input during the pe-riod 1970 to 1998 Panel B presents analogous numbers where inplace of observed task changes we tabulate changes in task inputpredicted by computerization Specically we use estimates of equa-tions (14)ndash(16) corresponding to the models in Tables IVndashVI tocalculate the predicted mean change in each task measure due tocontemporaneous industry or occupation computerization A limita-tion of this approach is that it treats computerization as an exoge-nous determinant of industry and occupation task change Since asstressed above we view computer adoption and task change assimultaneously determined we view this exercise as primarilyillustrative
We implement these calculations in panels C and D Panel Cuses equation (18) to estimate the extent to which rising input ofnonroutine tasks and declining input of routine tasks raised thecollege share of aggregate employment over 1970 ndash1998 As seenin columns 1ndash4 observed cross-occupation (extensive margin)task changes raised college employment by 21 percentage pointsper decade between 1970 and 1998 Three-quarters of this con-tribution (15 percentage points) is due to shifts favoring nonrou-tine cognitive tasks The remainder is explained by shifts againstroutine cognitive and manual tasks
Columns 5ndash7 perform analogous calculations for 1980 to 1998Here we add within-occupation (intensive) margin task change for1977 to 1991 In net shifts favoring nonroutine over routine taskscontributed 25 percentage points growth per decade to college-equivalent employment over these eighteen years27
27 Observed intensive margin shifts did not contribute to this demandgrowth however due to the offsetting effects of routine cognitive and nonroutineanalytic tasks This stands in contrast to within-occupation task changes pre-dicted by computerization where intensive margin shifts are economically large
1317SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
TABLE VIISHIFTS IN COLLEGE-EQUIVALENT LABOR DEMAND IMPLIED
BY CHANGES IN JOB TASKS 1970ndash1998
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
A 10 3 observed annual changes in DOT task measures (percentile changesrelative to 1960 task distribution)
Nonroutineanalytic
302 297 312 304 305 2039 267
Nonroutineinteractive
468 531 448 484 485 058 543
Routinecognitive
2014 2348 2488 2303 2426 2276 2702
Routinemanual
163 2147 2388 2144 2281 074 2207
B 10 3 predicted annual changes in DOT task measures(percentile changes relative to 1960 task distribution)
NIPA computer input measure CPS computer use measure
Nonroutineanalytic
084 135 230 155 256 054 310
Nonroutineinteractive
147 236 401 270 316 104 420
Routinecognitive
2105 2168 2286 2192 2314 2332 2646
Routinemanual
2115 2186 2315 2212 2284 032 2252
C 10 3 predicted annual changes in college-equivalent share of employmentin percentage points due to observed task shifts (panel A)
Nonroutinetasks
153 140 163 151 153 2036 083
Routinetasks
2020 066 117 050 097 027 087
All tasks 133 206 280 201 249 2009 240
Panel A Observed extensive margin task shifts are dened as the change in economywide input of eachtask (in percentiles of the 1960 task distribution) estimated using DOT 1977 occupational task measuresapplied to Census and CPS samples for 1970 to 1998 and summarized in Table II Intensive margin shifts aremeasured as change in mean of DOT occupational task input (measured in centiles of the 1977 taskdistribution) between 1977 and 1991 DOT revisions using the 1980 and 1998 occupational distributions ofemployment from the CPS MORG samples See Table I and Appendix 1 for denitions and examples of taskvariables
Panel B Predicted task changes are calculated as the weighted mean of the NIPA computer investmentmeasure or CPS computer use measure (as noted) multiplied by the coefcient from a regression of changesin industry or occupation task input on the relevant computer measure NIPA coefcient estimates corre-spond to specication (1) of Table IV CPS extensive margin computer task estimates correspond to Panel Aof Table V CPS intensive margin computer task estimates correspond to specication (1) of Table VI (Anegligible interaction term is ignored)
1318 QUARTERLY JOURNAL OF ECONOMICS
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
TABLE VII(CONTINUED)
1 1970ndash1980
extensivemargin
2 1980ndash1990
extensivemargin
3 1990ndash1998
extensivemargin
4 1970ndash1998
extensivemargin
5 1980ndash1998
extensivemargin
6 1980ndash1998
intensivemargin
7 1980ndash1998
extensive 1intensive
D 103 annual changes in college-equivalent share of employment in percentagepoints predicted by impact of computerization on task input (panel B)
NIPA computer investment measure CPS computer use measure
Nonroutinetasks
040 065 110 069 141 040 181
Routinetasks
029 048 081 051 080 104 184
All tasks 070 112 191 119 221 144 365
E Estimated log demand shifts for college-equivalentnoncollege-equivalentlabor 1970ndash1998 (100 3 annual log changes)
Using constant-elasticity of substitution model to estimate changes in collegedemand
s 5 00 499 253 225 333 241s 5 14 395 465 276 386 381s 5 20 350 556 298 409 441
Using task model to predict changes in college demand
Total task D(panel C)
123 129 143 131 156 2006 151
Predicted bycomputer-ization(panel D)
064 070 098 076 139 091 229
Panels C and D Implied employment share changes for college-equivalent labor (in percentage points)are calculated as the inner product of observed or predicted changes in task input from panels A and B andthe coefcient vector from a xed coefcient model of educational input For extensive margin task shifts thiscoefcient vector is estimated from a regression of college equivalent employment (in FTEs) in 140 consistentCIC industries on the ve DOT measures of industry task input (in centiles) and a constant using the 1980MORG sample For intensive margin task shifts the coefcient vector is estimated from a regression ofcollege equivalent employment in 470 COC occupations on the ve DOT measures of occupational task input(in centiles) and a constant using the 1984 CPS sample College-equivalents labor is dened as all workerswith college or greater education plus half of those with some college
Panel E Fixed coefcients log relative demand shifts are calculated as the change in the log ratio ofcollege-equivalentnoncollege-equivalent employment using the initial (1970 1980 or 1990) college-equiva-lentequivalent employment share in full-time equivalents and the implied percentage point change in thisshare from Panels C and D
Constant Elasticity of Substitution (CES) implied relative demand shifts for college-equivalentlabor arecalculated following Autor Katz and Krueger [1998] using a CES aggregate production function with twoinputs college and high school equivalents and an elasticity of substitution denoted by s See Table II ofAutor Katz and Krueger for details
1319SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
We next estimate the contribution of task changes attribut-able to computerization to growth in college graduate employ-ment These estimates shown in panel D implement equation(18) using the within-industry and within-occupation task shiftspredicted by computerization (panel B) The estimates of theldquocomputer-inducedrdquo changes in college employment are compara-ble in magnitude to and in some cases larger than the analogousestimates based on observed task inputs We nd for examplethat computer-induced task changes along the extensive margincontributed 12 percentage points per decade to growth in collegeemployment during 1970 to 1998 and 22 percentage points dur-ing 1980 to 1998 Adding intensive margin task changes raisesthis estimate considerably In net we estimate that task shiftsattributable to computerization increased college employment by37 percentage points per decade from 1980 to 1998
The nal step of our estimation is to benchmark employmentchanges induced by shifting task demands against conventionalestimates of demand for college labor calculated for the sameperiod For this benchmarking exercise we use the familiar con-stant elasticity of substitution (CES) framework with two factorsof productionmdash college equivalent and high school equivalent la-bormdashto estimate log relative demand shifts favoring college laborduring 1970 to 1998 Under the assumptions that the economyoperates on the demand curve and that factors are paid theirmarginal products the CES model calculates the implied shift incollegenoncollege relative demand consistent with observedshifts in relative employment and earnings of college versusnoncollege workers28 This procedure also requires us to assumean elasticity of substitution s between college and high schoolequivalent workers Following a large literature we use values ofs ranging from 0 to 2 with a consensus estimate of s 5 14
Panel E of the table presents the benchmark CES estimatesfor the period 1970 to 1998 We calculate that relative demand for
28 The demand model is Q t 5 [at(atNc t)r 1 (1 2 at)(btNh t)r]1 r where Q isaggregate output Nct Nht are quantities of employed college and high schoolequivalent labor at b t are factor-augmenting technological parameters a is anindex of the share of work activities allocated to college versus high school laborand the elasticity of substitution is given by s 5 1(1 2 r) The demand index isDt 5 s ln(at[1 2 at]) 1 (s 2 1)ln(at bt) Katz and Murphy [1992] Johnson[1997] Murphy Romer and Riddell [1999] and Acemoglu [2002] implementsimilar models Our estimates are based on Table II of Autor Katz and Krueger[1998] and updated in 1998 (from 1996) for this analysis Unlike these authors wealso include estimates for s 5 0 since our xed coefcients model incorporates thisassumption
1320 QUARTERLY JOURNAL OF ECONOMICS
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
college-educated labor grew rapidly between 1970 and 1998 Us-ing the consensus value of s 5 14 we estimate that demand forcollege labor rose by 39 log points annually If we instead assumethat s 5 0 we obtain a smallermdashthough still quite rapidmdashdemand shift of 33 log points annually29
In the bottom of panel E we nally compare our task-baseddemand estimates with the CES-based numbers To facilitatecomparison we convert predicted changes in the college employ-ment level (panels C and D) into changes in log relative college-equivalentnoncollege employment As noted above this task-based calculation ignores wage changes and hence corresponds toan elasticity of substitution of s 5 0 It is therefore useful tocompare the task-based demand estimates to the correspondingCES numbers using both the consensus elasticity of s 5 14 andthe more comparable albeit less realistic value of s 5 0
As is visible in panel E the task model explains a sizablesharemdash25 to 65 percentmdash of the estimated growth in college-equivalentnoncollege-equivalent demand in each decade Consis-tent with the accelerating rate of task change shown in Figure Ithe smallest share of the overall demand shift is explained bytask shifts in the 1970s and the largest share in the 1990s
Comparing task changes potentially attributable to comput-erization to the CES demand index for these three decades wend that these extensive margin task changes explain 20 to 25percent of the estimated demand shift for college versus noncol-lege labor during 1970 to 1998 If we focus on only the two mostrecent decades and include both intensive and extensive marginchanges the task model can explain a large fractionmdash 60 to 90percentmdashof the estimated increase in relative demand for collegeemployment Notably almost 40 percent of the computer contri-bution to rising educational demand in the last two decades is dueto shifts in task composition within nominally unchangingoccupations
In net these illustrative calculations demonstrate thatchanges in task demands accompanying workplace computeriza-tion are economically large andmdashwith caveats notedmdashcould havecontributed substantially to relative demand shifts favoring edu-cated labor in the United States since 1970
29 The estimated demand shift is smaller in the latter calculation because ahigher value of s places greater weight on relative wage changes and the relativeearnings of college graduates rose rapidly after 1980
1321SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
VII CONCLUSION
What is it that computers domdash or what is it that people dowith computersmdashthat appears to increase demand for educatedworkers This paper formalizes and tests an intuitive answer tothis question that has been informally articulated by scholars ina number of disciplines over several decades Computer technol-ogy substitutes for workers in performing routine tasks that canbe readily described with programmed rules while complement-ing workers in executing nonroutine tasks demanding exibilitycreativity generalized problem-solving capabilities and complexcommunications As the price of computer capital fell precipi-tously in recent decades these two mechanismsmdashsubstitutionand complementaritymdash have raised relative demand for workerswho hold a comparative advantage in nonroutine tasks typicallycollege-educated workers
Our task framework emphasizes that the causal force bywhich advancing computer technology affects skill demand is thedeclining price of computer capitalmdashan economywide phenome-non We developed a simple model to explore how this pricedecline alters task demand within industries and occupationsThis model predicts that industries that were intensive in laborinput of routine tasks in the precomputer era would make rela-tively larger investments in computer capital Simultaneouslythey would reduce labor input of routine tasks for which com-puter capital substitutes and increase demand for nonroutinetask input which computer capital complements
Employing consistent representative time series observa-tions on the task composition of jobs from the Dictionary ofOccupational Titles we afrm these predictions across severalmargins of task change and estimate that they may have contrib-uted substantially to demand shifts favoring educated labor overthe past three decades We also considered several alternativeexplanations for our ndings most signicantly the rising humancapital and labor force attachment of women We nd that thedocumented task shifts and their associations with the adoptionof computer technology are as evident within gender educationand occupation groups as between them The pervasiveness ofthese shifts suggests to us that changes in job task contentmdashspurred by technological changemdashmay plausibly be viewed as anunderlying factor contributing to recent demand shifts favoringeducated labor
1322 QUARTERLY JOURNAL OF ECONOMICS
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
AP
PE
ND
IX1
DE
FIN
ITIO
NS
OF
TA
SK
ME
AS
UR
ES
FR
OM
TH
E19
77D
ICT
ION
AR
YO
FO
CC
UP
AT
ION
AL
TIT
LE
S
Var
iabl
eD
OT
de
niti
onT
ask
inte
rpre
tati
onE
xam
ple
task
sfr
omH
and
book
for
Ana
lyzi
ng
Job
s
1G
ED
Mat
h(M
AT
H)
Gen
eral
educ
atio
nal
deve
lopm
ent
mat
hem
atic
s
Mea
sure
ofno
nro
utin
ean
alyt
icta
sks
Low
est
leve
lA
dds
and
subt
ract
s2-
digi
tn
umbe
rsp
erfo
rms
oper
atio
nsw
ith
uni
tssu
chas
cup
pint
and
quar
tM
idle
vel
Com
pute
sdi
scou
nt
inte
rest
pro
ta
nd
loss
ins
pect
s
atgl
ass
and
com
pile
sde
fect
data
base
don
sam
ples
tode
term
ine
vari
ance
sfr
omac
cept
able
qual
ity
limit
sH
ighe
stle
vel
Con
duct
san
dov
erse
esan
alys
esof
aero
dyna
mic
and
ther
mod
ynam
icsy
stem
s
to
dete
rmin
esu
itab
ilit
yof
desi
gnfo
rai
rcra
ftan
dm
issi
les
2D
irec
tion
C
ontr
ol
Pla
nnin
g(D
CP
)
Ada
ptab
ility
toac
cept
ing
resp
onsi
bili
tyfo
rth
edi
rect
ion
cont
rol
orpl
anni
ng
ofan
acti
vity
Mea
sure
ofno
nro
utin
ein
tera
ctiv
eta
sks
Pla
ns
and
desi
gns
priv
ate
resi
denc
eso
fce
build
ings
fac
tori
esa
nd
othe
rst
ruct
ures
ap
plie
spr
inci
ples
ofac
coun
ting
toin
stal
lan
dm
ain
tain
oper
atio
nof
gene
ral
acco
unti
ngsy
stem
con
duct
spr
osec
uti
onin
cour
tpr
ocee
ding
s
gat
hers
and
anal
yzes
evid
ence
re
view
spe
rtin
ent
deci
sion
s
app
ears
agai
nst
accu
sed
inco
urt
ofla
wc
omm
ands
sh
ing
vess
elcr
ewen
gage
din
catc
hin
gs
han
dot
her
mar
ine
life
3
Set
Lim
its
Tol
eran
ces
orS
tan
dard
s(S
TS
)
Ada
ptab
ility
tosi
tuat
ion
sre
quir
ing
the
prec
ise
atta
inm
ent
ofse
tlim
its
tole
ran
ces
orst
anda
rds
Mea
sure
ofro
utin
eco
gnit
ive
task
sO
pera
tes
abi
llin
gm
achi
neto
tran
scri
befr
omof
ce
reco
rds
data
cal
cula
tes
degr
ees
min
utes
and
seco
ndof
lati
tude
and
long
itu
deu
sing
stan
dard
navi
gati
onai
ds
mea
sure
sdi
men
sion
sof
bott
leu
sing
gaug
esan
dm
icro
met
ers
tove
rify
that
setu
pof
bott
le-m
akin
gco
nfor
ms
tom
anuf
actu
rin
gsp
eci
cati
ons
prep
ares
and
veri
es
vote
rli
sts
from
ofc
ial
regi
stra
tion
reco
rds
4F
inge
rD
exte
rity
(FIN
GD
EX
)
Abi
lity
tom
ove
nge
rs
and
man
ipul
ate
smal
lob
ject
sw
ith
nge
rs
rapi
dly
orac
cura
tely
Mea
sure
ofro
utin
em
anu
alta
sks
Mix
esan
dba
kes
ingr
edie
nts
acco
rdin
gto
reci
pes
sew
sfa
sten
ers
and
deco
rati
vetr
imm
ings
toar
ticl
esf
eeds
tun
gste
nl
amen
tw
ire
coils
into
mac
hine
that
mou
nts
them
tost
ems
inel
ectr
icli
ght
bulb
sop
erat
esta
bula
ting
mac
hin
eth
atpr
oces
ses
data
from
tabu
lati
ng
card
sin
topr
inte
dre
cord
spa
cks
agri
cult
ural
prod
uce
such
asbu
lbs
frui
tsn
uts
egg
san
dve
geta
bles
for
stor
age
orsh
ipm
ent
atta
ches
hand
sto
face
sof
wat
ches
5
Eye
Han
dF
oot
Coo
rdin
atio
n(E
YE
HA
ND
)
Abi
lity
tom
ove
the
han
dan
dfo
otco
ordi
nate
lyw
ith
each
othe
rin
acco
rdan
cew
ith
visu
alst
imu
li
Mea
sure
ofno
nro
utin
em
anu
alta
sks
Low
est
leve
lT
ends
mac
hine
that
crim
psey
elet
sgr
omm
ets
next
leve
lat
tend
sto
beef
catt
leon
stoc
kra
nch
driv
esbu
sto
tran
spor
tpa
ssen
gers
nex
tle
vel
pilo
tsai
rpla
neto
tran
spor
tpa
ssen
gers
pru
nes
and
trea
tsor
nam
enta
lan
dsh
ade
tree
shi
ghes
tle
vel
perf
orm
sgy
mn
asti
cfe
ats
ofsk
illan
dba
lanc
e
Sou
rce
US
Dep
artm
ent
ofL
abor
M
anpo
wer
Adm
inis
trat
ion
Han
dboo
kfo
rA
naly
zin
gJ
obs
(Was
hing
ton
DC
197
2)
1323SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
DATA APPENDIX
A1 Samples Used from Current Population Survey and Censusof Populations
To calculate occupational and education distributions econo-mywide and within industries for 1960 to 1998 we used obser-vations on all noninstitutionalized employed workers ages
APPENDIX 2 COMPUTERIZATION AND INDUSTRY TASK INPUT 1960ndash1998USING COMPOSITE TASK MEASURES
DEPENDENT VARIABLE 10 3 ANNUAL WITHIN-INDUSTRY CHANGE IN TASK INPUTMEASURED IN PERCENTILES OF 1960 TASK DISTRIBUTION
1 1990ndash1998
2 1980ndash1990
3 1970ndash1980
4 1960ndash1970
A D Nonroutineanalytic
D Computer use 821 1209 650 8571984ndash1997 (455) (455) (421) (574)Intercept 064 2067 007 2007
(096) (094) (086) (114)R2 002 005 002 002Weighted mean D 226 167 132 151
B D Nonroutineinteractive
D Computer use 983 993 567 10451984ndash1997 (439) (474) (347) (500)Intercept 054 053 142 000
(092) (098) (071) (100)R2 004 003 002 003Weighted mean D 248 245 251 193
C D Routinecognitive
D Computer use 21340 2459 2476 27021984ndash1997 (444) (558) (370) (475)Intercept 023 2039 056 2005
(094) (116) (076) (095)R2 006 000 001 002Weighted mean D 2241 2128 2035 2135
D D Routinemanual
D Computer use 22317 21527 21325 3981984ndash1997 (699) (686) (534) (413)Intercept 2052 2080 197 135
(147) (142) (109) (082)R2 007 003 004 001Weighted mean D 2509 2376 2056 209
n is 140 consistent CIC industries Standard errors are in parentheses Each column of panels AndashDpresents a separate OLS regression of ten times the annual change in industry-level task input between theendpoints of the indicated time interval (measured in centiles of the 1960 task distribution) on the annualpercentage point change in industry computer use during 1984ndash1997 (mean 0193) and a constant Computeruse is the fraction of industry workers using a computer at their jobs estimated from the October 1984 and1997 CPS samples Estimates are weighted by mean industry share of total employment in FTEs over theendpoints of the years used to form the dependent variable Samples used are Census 1960 1970 and 1980and CPS MORG 1980 1990 and 1998 See the Data Appendix for details on construction of the compositetask variables See Table I for denitions and examples of task variables
1324 QUARTERLY JOURNAL OF ECONOMICS
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
18ndash64 from the Census PUMS one percent samples for 19601970 1980 and 1990 [Ruggles and Sobek 1997] and the MergedOutgoing Rotation Groups of the Current Population Survey forthe years 1980 1990 and 1998 All individual and industry levelanalyses are performed using as weights full-time equivalenthours of labor supply which is the product of the individualCensus or CPS sampling weight times hours of work in thesample reference week divided by 35 and for Census samplesweeks of work in the previous year Because hours are not re-ported for the self-employed in the CPS prior to 1994 we assignedself-employed workers in all CPS samples the average laborhours in their industry-education-year cell In cases where indus-try hours supplied by education category were unavailable (due toan empty industry-education-year cell) we assigned weeklyhours as the mean of workersrsquo education-year cells
To attain comparable educational categories across the re-denition of Census Bureaursquos education variable introduced in1990 in the Census and in 1992 in the CPS we use the methodproposed by Jaeger [1997] In data coded with the pre-1992 edu-cation question (Census PUMS 1960 1970 and 1980 and CPSMORG les 1980 and 1990) we dened high school dropouts asthose with fewer than twelve years of completed schooling highschool graduates as those having twelve years of completedschooling some college attendees as those with any schoolingbeyond twelve years (completed or not) and fewer than sixteencompleted years and college plus graduates as those with sixteenor more years of completed schooling In data coded with therevised education question (1990 Census PUMS and 1998 CPSMORG le) we dene high school dropouts as those with fewerthan twelve years of completed schooling high school graduatesas those with either twelve completed years of schooling or a highschool diploma or GED some college as those with some collegeor holding an Associatersquos Degree and college plus as those with aBA or higher
A2 Computing DOT Task Means for Census OccupationCategories (COCs)
To compute DOT Task Means for 1970 CIC Occupations weused the April 1971 CPS Monthly File issued by the NationalAcademy of Sciences [1981] in which experts assigned individualDOT occupation codes and associated DOT measures to each of60441 workers Because Census occupation categories are sig-
1325SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
nicantly coarser than DOT occupation categories the 411 1970census occupation codes represented in the 1971 CPS were as-signed a total of 3886 unique 1977 DOT occupations We used theCPS sampling weights to calculate means of each DOT taskmeasure by occupation Because the gender distribution of DOToccupations differs substantially within COC occupation cells weperformed this exercise separately by gender In cases where aCOC cell contained exclusively males or females we assigned thecell mean to both genders This provided a set of 822 DOT occu-pation means by 1970 COC and gender
To generate DOT means for 1960 occupations we developeda crosswalk from the 1970 to 1960 COC occupational classica-tion schemes using information in Priebe and Greene [1972] Ourcrosswalk (available on request) provides a set of 211 consistent1960 ndash1970 occupations representing the lowest common level ofaggregation needed to obtain a consistent series We applied the1970 COC means to our 1970 Census sample by occupation andgender and calculated weighted gender-occupation means acrossthe 211 consistent 1960 ndash1970 occupational categories
It was not possible to develop a bridging crosswalk between1970 and 1980 COC occupations due to the substantial differ-ences between these classications Instead we employed a Cen-sus sample prepared for the Committee on Occupational Classi-cation and Analysis chaired by Donald Treiman and provided tous by Michael Handel This le contains 122141 observationsfrom the 1980 Census that are individually dual coded with both1970 and 1980 COC occupation codes based on occupational andother demographic information supplied by Census respondentsTo calculate DOT means by 1980 occupation we merged the 1970COC-DOT means (above) to the Treiman le by gender and 1970COC occupation achieving a 97 percent match rate We appendedto the Treiman le consistent occupation codes for the years 1980to 1998 developed by Autor Katz and Krueger [1998] and cal-culated weighted means of each DOT measure within occupation-gender categories This yielded DOT means by gender for each of485 DOT occupations
A3 Computing DOT Task Means by Consistent1960 ndash1998 Industry
To compute DOT task means overall by industry and byindustry-education cell for 1960 ndash1998 we assigned the consis-tent DOT occupational task means for 1960 ndash1998 by gender and
1326 QUARTERLY JOURNAL OF ECONOMICS
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
occupation to each observation in our Census and CPS samplesfor 1960 ndash1998 Using labor supply in FTEs as weights we calcu-lated means of each DOT measure for each occupation-industry-education-year cell These means provide the primary outcomemeasures for our analysis To attain compatibility betweenchanging Census Industry Codes for 1960 ndash1998 we use a cross-walk developed by Autor Katz and Krueger [1998] providing 140consistent CIC industries spanning all sectors of the economyThis crosswalk includes all CIC industries and attains consis-tency by aggregating where necessary to the lowest common levelof consistent industry denition among 1970 1980 and 1990 CICstandards
A4 Composite Task Indicators from the DOT
To verify that our results are robust to plausible alternativeselections of DOT variables we formed composite indicators ofour intended constructs using Principal Components AnalysisWe chose a short list of alternative DOT variables that appearedrelevant to each of our conceptual categories These choices areNonroutine Analytic Tasks GED-MATH GED-REASON NUM-BER MVC Nonroutine Interactive Tasks DCP GED-LAN-GUAGE DEPL VARCH Routine Cognitive Tasks STS COL-ORDIS REPCON VOCPREP Routine Manual Tasks FING-DEX MOTOR FORM MANUAL Denitions and representativeexamples of these variables are found in U S Department ofLabor [1972] Using 1980 employment as weights we performedprincipal components analysis for each set of variables to identifythe linear combinations that maximized common variation sub-ject to the constraint that the sum of squared vector weights isequal to one In each case we used the rst principal component
A5 Calculating DOT Quantiles
To convert DOT measures into percentiles of the 1960 taskdistribution we used DOT 1977 task measures paired to the 1960Census to form an employment-weighted empirical cumulativedistribution function of task input for each task measure across1120 industry-education-gender cells 140 industries two genderand four education levels (high school dropout high school grad-uate some college and college graduate) We applied a smallamount of interpolation to remove at spots in the distributionWe inverted this empirical distribution and applied it to all DOTtask measures in subsequent years by industry-education-gender
1327SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
and year These centile values are the unit of measure for all lateranalyses A limitation of this methodology is that DOT valuesafter 1960 may potentially lie outside the support of the 1960distribution leading to truncation In practice this issue affectsat most 4 percent of the distribution of each task in any year Anearlier version of this paper [Autor Levy and Murnane 2001]employed raw DOT scores rather than the percentile measuresused here Results were qualitatively identical
A6 Calculating within-Occupation Changes in DOT TaskMeasures 1977ndash1991
To measure within-occupation changes in task content weemployed the 1991 Revised Fourth Edition of the Dictionary ofOccupational Titles available in electronic form from the Na-tional Academy of Sciences [1981] Based on a study of selectindustries to determine which jobs had undergone the most sig-nicant occupational changes since the 1977 publication of theDOT fourth edition DOT analysts introduced revised and elimi-nated occupational denitions for occupations that were observedto have most substantively changed between 1977 and 1991 Atotal of 2452 occupations were reviewed updated or were added646 nominal titles were revised 136 titles were combined and 75were deleted To provide a conservative measure of total taskchange we assume that any occupation that was not revised inthe 1991 DOT experienced no task change
We used documentation from the North Carolina Employ-ment Security Commission [1992a 1992b] to construct a cross-walk between the 1991 DOT and 1997 DOT occupation codesWith this crosswalk we applied DOT 1991 task variables to our1971 CPS le yielding a match rate of 999 percent Of thesematched occupations 73 percent (weighted by employment) hadbeen updated between 1977 and 1991 by DOT examiners Wethen calculated DOT means by 1970 and 1980 COC occupationsand gender using a procedure identical to that described in A3 toobtain 1991 DOT task means for the 1977 occupations The with-in-occupation variation that we exploit over 1977ndash1991 stemsexclusively from reevaluation of occupational content by DOTexaminers rather than from changes in the relative size of DOTsuboccupations within CIC occupations
We assigned the 1977 DOT and 1991 DOT task measures to470 consistent COC occupations in the 1984 and 1998 CPS sam-ples (corresponding to the years of our CPS computer measure)
1328 QUARTERLY JOURNAL OF ECONOMICS
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
As with the industry data we transformed the DOT task mea-sures into percentiles of the base year distribution in this caseusing occupational employment shares from 1984 to form employ-ment weights
A7 Computer Usage Data from the Current Population Survey
Industry computer use frequencies were calculated from theOctober 1984 and 1997 School Enrollment Supplements to theCurrent Population Survey (CPS) as the weighted fraction ofcurrently employed workers ages 18ndash 65 who answered yes to thequestion ldquoDo you use a computer directly at workrdquo within con-sistent CIC industries A computer is dened as a desktop termi-nal or PC with keyboard and monitor and does not include anelectronic cash register or a handheld data device To calculatethese frequencies in 1984 and 1997 61712 and 56247 observa-tions were used respectively
A8 Computer and Capital Investment Measures from theNational Income and Products Accounts
We used data on industry capital stocks and ows of equip-ment structures and computers from the National Income andProduct Accounts [U S Department of Commerce 2002a 2002b]for the years 1950 ndash1998 All NIPA stock and investment vari-ables are measured in real 1996 dollars Investment variablesmeasure cumulated real investment in the relevant asset over theprior ten years except in 1998 where we use 125 times cumu-lated investment over the previous eight years Deation of NIPAmeasures is performed by the Bureau of Economic Analysis usingprimarily Producer Price Indexes (PPIs) PPIs for computer in-vestment are based on quality adjustment price linking andhedonic regression methods As denominators for capitalFTEand computer investmentFTE variables we used Census andCPS samples to calculate FTEs by industry by year Computerinvestment is calculated as the sum of investment in mainframecomputers personal computers packaged and custom softwareprinters terminals storage devices and other integrated devicesStructures and equipment variables are dened in the NIPA
To match CPS and Census data to the NIPA we used acrosswalk developed by Autor Katz and Krueger [1998] andrevised for this analysis to accommodate small changes in theNIPA sector scheme made during the recent NIPA revision Theresulting aggregation of NIPA and CIC data contains 47 consis-
1329SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
tent industries covering all industrial sectors excluding Govern-ment and Private Households spanning 1960 ndash1998 Of these 47we exclude from our analysis agriculture and government domi-nated services (5 NIPA industries)
MASSACHUSETTS INSTITUTE OF TECHNOLOGY AND NATIONAL BUREAU
OF ECONOMIC RESEARCH
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RESEARCH
REFERENCES
Acemoglu Daron ldquoWhy Do New Technologies Complement Skills Directed Tech-nical Change and Wage Inequalityrdquo Quarterly Journal of Economic CXIII(1998) 1055ndash1089
mdashmdash ldquoChanges in Unemployment and Wage Inequality An Alternative Theoryand Some Evidencerdquo American Economic Review LXXXIX (1999)1259ndash1278
mdashmdash ldquoTechnical Change Inequality and the Labor Marketrdquo Journal of EconomicLiterature XL (2002) 7ndash72
Adler Paul S ldquoNew Technologies New Skillsrdquo California Management ReviewXXIX (1986) 9 ndash28
Autor David H Lawrence F Katz and Alan B Krueger ldquoComputing InequalityHave Computers Changed the Labor Marketrdquo Quarterly Journal of Econom-ics CXIII (1998) 1169ndash1214
Autor David H Frank Levy and Richard J Murnane ldquoThe Skill Content ofRecent Technological Change An Empirical Explorationrdquo National Bureau ofEconomics Research Working Paper No 8337 2001
Autor David H Frank Levy and Richard J Murnane ldquoUpstairs DownstairsComputers and Skills on Two Floors of a Large Bankrdquo Industrial and LaborRelations Review LV (2002) 432ndash 447
Bartel Anne P Casey Ichniowski and Kathryn Shaw ldquoNew Technology HumanResource Practices and Skill Requirements Evidence from Plant Visits inThree Industriesrdquo mimeograph Carnegie Mellon University 2000
Beamish Anne Frank Levy and Richard J Murnane ldquoComputerization andSkills Examples from a Car Dealershiprdquo mimeograph Massachusetts Insti-tute of Technology Department of Urban Studies 1999
Berman Eli John Bound and Zvi Griliches ldquoChanges in the Demand for SkilledLabor within U S Manufacturing Industries Evidence from the AnnualSurvey of Manufacturersrdquo Quarterly Journal of Economics CIX (1994)367ndash397
Berman Eli John Bound and Stephen Machin ldquoImplications of Skill-BiasedTechnological Change International Evidencerdquo Quarterly Journal of Eco-nomics CXIII (1998) 1245ndash1279
Berman Eli and Stephen Machin ldquoSkill-Biased Technology Transfer Evidence ofFactor Biased Technological Change in Developing Countriesrdquo mimeographBoston University 2000
Bertrand Marianne Esther Duo and Sendhil Mullainathan ldquoHow MuchShould we Trust Difference-in-Differences Estimatesrdquo Quarterly Journal ofEconomics CXIX (2004) forthcoming
Blau Francine Marianne Ferber and Anne Winkler The Economics of WomenMen and Work 4th ed (Upper Saddle River NJ Prentice-Hall 2002)
Borghans Lex and Bas ter Weel ldquoThe Diffusion of Computers and the Distribu-tion of Wagesrdquo mimeograph Maastricht University 2002
Bresnahan Timothy F ldquoComputerization and Wage Dispersion An AnalyticalReinterpretationrdquo Economic Journal CIX (1999) 390ndash 415
Bresnahan Timothy F Erik Brynjolfsson and Lorin M Hitt ldquoInformation Tech-
1330 QUARTERLY JOURNAL OF ECONOMICS
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
nology Workplace Organization and the Demand for Skilled Labor Firm-Level Evidencerdquo Quarterly Journal of Economics CXVII (2002) 339ndash376
Brynjolfsson Erik and Lorin M Hitt ldquoBeyond Computation Information Tech-nology Organizational Transformation and Business Performancerdquo Journalof Economic Perspectives XIV (2000) 23ndash48
Card David and John E DiNardo ldquoSkill-Biased Technological Change andRising Wage Inequality Some Problems and Puzzlesrdquo Journal of LaborEconomics XX (2002) 733ndash783
Caroli Eve and John Van Reenen ldquoSkill-Biased Organizational Change Evi-dence from a Panel of British and French Establishmentsrdquo Quarterly Journalof Economics CXVI (2001) 1449ndash1492
Chandler Alfred The Visible Hand The Managerial Revolution in AmericanBusiness (Cambridge MA Belknap Press 1977)
Dixit Avinish K and Joseph E Stiglitz ldquoMonopolistic Competition and OptimumProduct Diversityrdquo American Economic Review LXVII (1977) 297ndash308
Doms Mark Timothy Dunne and Kenneth R Troske ldquoWorkers Wages andTechnologyrdquo Quarterly Journal of Economics CXII (1997) 253ndash290
Drucker Peter F The Practice of Management (New York NY Harper Press1954)
Fernandez Roberto ldquoSkill-Biased Technological Change and Wage InequalityEvidence from a Plant Retoolingrdquo American Journal of Sociology CVII(2001) 273ndash320
Gera Surendra Wulong Gu and Zhengxi Lin ldquoTechnology and the Demand forSkills in Canada An Industry-Level Analysisrdquo Canadian Journal of Econom-ics XXXIV (2001) 132ndash148
Goldin Claudia Understanding the Gender Gap (New York NY Oxford Univer-sity Press 1990)
Goldin Claudia and Lawrence F Katz ldquoThe Decline of Non-Competing GroupsChanges in the Premium to Education 1890 to 1940rdquo National Bureau ofEconomics Research Working Paper No 5202 1995
Goldin Claudia and Lawrence F Katz ldquoThe Origins of Technology-Skill Comple-mentarityrdquo Quarterly Journal of Economics CXIII (1998) 693ndash732
Griliches Zvi ldquoCapital-Skill Complementarityrdquo Review of Economics and Statis-tics LI (1969) 465ndash 468
Handel Michael ldquoTrends in Direct Measures of Job Skill Requirementsrdquo JeromeLevy Economics Institute Working Paper No 301 2000
Hounshell David A From the American System to Mass Production 1800ndash1932The Development of Manufacturing Technology in the United States (Balti-more MD Johns Hopkins University Press 1985)
Howell David R and Edward N Wolff ldquoTrends in the Growth and Distributionof Skills in the U S Workplace 1960ndash1985rdquo Industrial and Labor RelationsReview XLI (1991) 486ndash502
Ichniowski Casey and Kathryn Shaw ldquoBeyond Incentive Pay Insidersrsquo Esti-mates of the Value of Complementary Human Resource Management Prac-ticesrdquo Journal of Economic Perspectives XVII (2003) 155ndash180
Ingram Beth F and George R Neumann ldquoThe Returns to Skillrdquo mimeographUniversity of Iowa 2000
Jaeger David A ldquoReconciling the Old and New Census Bureau Education Ques-tions Recommendations for Researchersrdquo Journal of Business and Econom-ics Statistics XV (1997) 300ndash309
Johnson George ldquoChanges in Earnings Inequality The Role of Demand ShiftsrdquoJournal of Economic Perspectives XI (1997) 41ndash54
Katz Lawrence F and David H Autor ldquoChanges in the Wage Structure andEarnings Inequalityrdquo in Handbook of Labor Economics Vol 3 O Ashenfelterand D Card eds (Amsterdam North-Holland and Elsevier 1999)
Katz Lawrence F and Kevin M Murphy ldquoChanges in Relative Wages 1963ndash1987 Supply and Demand Factorsrdquo Quarterly Journal of Economics CVII(1992) 35ndash78
Lang Kevin ldquoOf Pencils and Computersrdquo mimeograph Boston University 2002Levy Frank and Richard J Murnane ldquoWith What Skills Are Computers a
Complementrdquo American Economic Review Papers and Proceedings LXXXVI(1996) 258ndash262
1331SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
Lindbeck Assar and Dennis J Shower ldquoMultitask Learning and the Reorgani-zation of Work From Tayloristic to Holistic Organizationrdquo Journal of LaborEconomics XVIII (2000) 353ndash376
Machin Stephen and John Van Reenen ldquoTechnology and Changes in SkillStructure Evidence from Seven OECD Countriesrdquo Quarterly Journal ofEconomics CXIII (1998) 1215ndash1244
Miller Anne R Donald J Treiman Pamela S Cain and Patricia A Rooseeditors Work Jobs and Occupations A Critical Review of the Dictionary ofOccupational Titles (Washington DC National Academy Press 1980)
Mobius Markus ldquoThe Evolution of Workrdquo mimeograph Harvard University2000
Mokyr Joel The Lever of Riches Technological Creativity and Economic Progress(New York NY Oxford University Press 1990)
Murphy Kevin M Paul Romer and W Craig Riddell ldquoWages Skills and Tech-nology in the United States and Canadardquo in Elhanan Helpman ed GeneralPurpose Technologies and Economic Growth (Cambridge MA MIT Press1998)
National Academy of Science Committee on Occupational Classication andAnalysis ldquoFourth Edition Dictionary of DOT Scores for 1970 Census Catego-riesrdquo ICPSR Document No 7845 Ann Arbor MI 1981
Nelson Richard R and Sidney G Winter An Evolutionary Theory of EconomicChange (Cambridge MA Harvard University Press 1982)
Nordhaus William D ldquoThe Progress of Computingrdquo Yale Cowles FoundationDiscussion Paper No 1324 2001
North Carolina Employment Security Commission Occupational Analysis FieldCenter ldquoConversion Tables of Code and Title Changes Fourth to RevisedFourth Edition Dictionary of Occupational Titlesrdquo Raleigh NC 1992a
mdashmdash ldquoErrata Dictionary of Occupational Titles Fourth Edition Revised 1991rdquoRaleigh NC 1992b
Orr Julian E Talking about Machines An Ethnography of a Modern Job (IthacaNY Cornell University Press 1996)
Pinker Steven How the Mind Works (New York NY Norton 1997)Polanyi Michael The Tacit Dimension (New York NY Doubleday Press 1966)Priebe J A J Heinkel and S Greene ldquo1970 Occupation and Industry Classi-
cations in Terms of their 1960 Occupation and Industry Elementsrdquo CensusBureau Technical Paper No 26 1972
Roy A D ldquoSome Thoughts on the Distribution of Earningsrdquo Oxford EconomicPapers III (1951) 235ndash246
Ruggles Steven Matthew Sobek et al Integrated Public Use Microdata SeriesVersion 20 (Minneapolis MN Historical Census Projects University of Min-nesota 1997)
Rumberger Russell W ldquoThe Changing Skill Requirements of Jobs in the U SEconomyrdquo Industrial and Labor Relations Review XXXIV (1981) 578ndash590
Simon Herbert A ldquoThe Corporation Will it Be Managed by Machinesrdquo in M LAnshen and G L Bach eds Management and the Corporations 1985 (NewYork NY McGraw-Hill 1960)
Spenner Kenneth I ldquoDeciphering Prometheus Temporal Change in the SkillLevel of Workrdquo American Sociological Review LXVII (1983) 824ndash 837
mdashmdash ldquoSkill Meaning Methods and Measuresrdquo Work and Occupations XVII(1990) 399ndash421
Spitz Alexandra ldquoIT Capital Job Content and Educational Attainmentrdquo ZEWDiscussion Paper No 03-04 2003
Stiroh Kevin J ldquoInformation Technology and the U S Productivity RevivalWhat Do the Industry Data Sayrdquo American Economic Review XCII (2002)1559ndash1576
Thesmar David and Mathias Thoenig ldquoCreative Destruction and Firm Organi-zation Choice A New Look into the Growth-Inequality Relationshiprdquo Quar-terly Journal of Economics CXV (2000) 1201ndash1238
U S Department of Commerce Bureau of Economic Analysis ldquoDetailed Data forFixed Assets and Consumer Durable Goods 1947ndash2001rdquo 2002a available athttpwwwbeagov
mdashmdash ldquoIndustry Accounts Data 1947ndash2001rdquo 2002b available at httpwwwbeagov
1332 QUARTERLY JOURNAL OF ECONOMICS
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE
U S Department of Labor Manpower Administration Handbook for AnalyzingJobs (Washington DC 1972)
U S Department of Labor Dictionary of Occupational Titles First Edition(Washington DC 1939)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Fourth Edition (Washington DC 1977)
U S Department of Labor Employment and Training Administration Dictionaryof Occupational Titles Revised Fourth Edition (Washington DC 1991)
U S Department of Labor Occupational Outlook Handbook 1976ndash1977 EditionBulletin 1875 1976
U S Department of Labor Occupational Outlook Handbook 2000ndash2001 EditionBulletin 2520 2000
Weinberg Bruce ldquoComputer Use and the Demand for Female Workersrdquo Indus-trial and Labor Relations Review LIII (2000) 209ndash308
Winston Patrick H ldquoWhy We Should Start Overrdquo Keynote address AmericanAssociation for Articial Intelligence Orlando FL July 18 1999
Wolff Edward N ldquoThe Growth of Information Workers in the U S Economy1859ndash1990 The Role of Technological Change Computerization and Struc-tural Changerdquo C V Starr Center for Applied Economics New York Univer-sity RR No 96-41 1996
mdashmdash ldquoComputerization and Structural Changerdquo Review of Income and WealthXLVII (2002) 59ndash75
Xiang Chong ldquoNew Goods and Rising Skill Premium An Empirical Investiga-tionrdquo Research Seminar in International Economics Working Paper No 478University of Michigan 2002
Zuboff Shoshana In the Age of the Smart Machine (New York NY Basic Books1988)
Zue V W and J R Glass ldquoConversational Interfaces Advances and ChallengesrdquoProceedings of the IEEE LXXXVIII (2000) 1166ndash1180
1333SKILL CONTENT OF TECHNICAL CHANGE