Is public R&D a complement or substitute for private R&D ...bhhall/papers/DavidHallToole RP00.pdf · into the provision of government-provided services, both the defense-related and

Ž .Research Policy 29 2000 497–529www.elsevier.nlrlocatereconbase

Is public R&D a complement or substitute for private R&D? Areview of the econometric evidence

Paul A. David a,c,1, Bronwyn H. Hall b,d,e,), Andrew A. Toole c,f

a All Souls College, Oxford, UKb Oxford UniÕersity, Oxford, UK

c Stanford UniÕersity, Stanford, CA, USAd Nuffield College, Oxford, UK

e UniÕersity of California at Berkeley, Berkeley, CA, USAf Stanford Institute for Economic Policy Research, Stanford UniÕersity, Stanford, CA, USA

Abstract

Is public R&D spending complementary and thus ‘‘additional’’ to private R&D spending, or does it substitute for andtend to ‘‘crowd out’’ private R&D? Conflicting answers are given to this question. We survey the body of availableeconometric evidence accumulated over the past 35 years. A framework for analysis of the problem is developed to helporganize and summarize the findings of econometric studies based on time series and cross-section data from various levels

Ž .of aggregation laboratory, firm, industry, country . The findings overall are ambivalent and the existing literature as a wholeŽ .is subject to the criticism that the nature of the ‘‘experiment s ’’ that the investigators envisage is not adequately specified.

We conclude by offering suggestions for improving future empirical research on this issue. q 2000 Elsevier Science B.V.All rights reserved.

JEL classification: O30; O38; H40; H54; L10Keywords: R&D; Fiscal policy; Government subsidy; Technology policy

1. Introduction

The opening of the new millennium finds a na-tional public-sector civilian research enterprise whosescale and scope in most of the world’s countriessurpasses that of any previous period of their history.Among the leading industrial nations this may beseen as the outcome of a long historical process

) Corresponding author. Department of Economics, Universityof California at Berkeley, 549 Evans- Hall a 3880, Berkeley, CA94720-3880, USA. E-mail: [email protected]

1 Also corresponding author. E-mail:[email protected].

initiated with state patronage during the scientificrevolution of the seventeenth century. But, only sincethe closing decades of the nineteenth century haveorganized research and development activities begunto make appreciable claims upon the productive re-sources of those societies. 2 Since then, however, thefraction of real gross national product being directedby both private and governmental agencies towardexpanding the base of scientific and technologicalknowledge for non-defense purposes has trended up-

2 Ž . Ž .See, e.g., David 1998a; b , Lenoir 1998 , and sources citedtherein.

0048-7333r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.Ž .PII: S0048-7333 99 00087-6

( )P.A. DaÕid et al.rResearch Policy 29 2000 497–529498

wards — halting at times, yet never reversing signif-icantly.

Most of the growth in the relative importance ofthis intangible form of capital accumulation has comewithin the past half-century: even under the stimulusof military preparations during the 1930s, total R&Dexpenditures in countries such as the US, the UK andJapan remained in the range between two-thirds andone-quarter of one percentage point of their respec-tive national product figures. 3 In the aftermath ofWorld War II, the belief that organized research anddevelopment could stimulate economic growth andcontribute to improving economic welfare led to thecreation of many new public institutions supportingcivilian science and engineering, and pushed the civilR&D fraction upwards towards the one percentagepoint level in a growing number of countries. TheCold War Era fostered a further expansion of gov-ernment agency research programs in non-defense aswell as military technologies, and established modelsfor the performance of government-funded R&D byprivate sector contractors. Thus, accompanying theinstitutional expansion in public sector production ofscientific and technological knowledge, there wereenormous increases in the scale of public financialobligations for R&D activities performed primarilyby non-governmental agents. Added to these, a vari-ety of tax and subsidy measures was introduced withthe intention of encouraging private firms to under-take R&D projects at their own expense. 4

Although most people believe that governmentR&D activities contribute to innovation and produc-

3 Ž .See the estimates in Edgerton 1996 , Table 5.8, but note thatŽthe R&D fraction shown for Japan as 0.22% of national income

.in 1934 actually refers to share of GNP and is based on theresults of a 1930 survey; a Japanese survey taken in 1942 returnedexpenditures on the order of 1.5% of GNP. See Odagiri and GotoŽ .1993 , p. 84.

4 For an historical account focusing upon the US in the twenti-Ž .eth century, see Mowery and Rosenberg 1989 and the contribu-

tions dealing with particular sectors and industries in NelsonŽ .1982 . For the post-Cold War climate affecting government

Ž .support, especially in the US, see Cohen and Noll 1997 . NelsonŽ .1993 brings together profiles of the evolution of the ‘nationalinnovation systems’ of a variety of industrialized and some devel-

Ž .oping countries, mainly since the 1960s; Soligen 1994 providesa broader, internationally comparative treatment of relations be-tween scientific research and the twentieth century state.

tivity, many economists and policymakers havegrown frustrated with the paucity of systematic sta-tistical evidence documenting a direct contributionfrom public R&D. The burden of econometric find-ings concerning the productivity growth effects ofR&D seems to be that there is a significantly posi-tive and relatively high rate of return to R&Dinvestments at both the private and social levels. Yet,quite generally, privately funded R&D in manufac-turing industries is found to yield a substantial pre-mium over the rates of return from ‘‘own productiv-ity improvements’’ derived from R&D performedwith government funding. 5 In a recent survey,

Ž .Griliches 1995, p. 82 , suggests that the especiallypronounced differential over the returns on tangiblecapital investments observed at the private level mayreflect individual firms’ perceptions of especiallyhigh private risk in the case of R&D. The latterwould, of course, lead to the imposition of higherhurdle rates of return for firms’ individual fundingdecisions; whereas, by comparison, government-funded industrial R&D projects would be seen as

Ž .carrying less private risk, especially as much of itis devoted to ‘‘product innovation’’ for ‘‘output’’that eventually is to be sold back to the governmentprocurement agency under the terms of ‘‘cost plus’’contracts. In such circumstances there is little basisfor expecting that the R&D it performed with publicmonies would have a substantial direct impact on thecontracting firm’s own productivity.

1.1. The issue: substitution Õs. complementarity inpublic and priÕate R&D inÕestments

Having a direct impact on innovation that showsup as industrial productivity growth, however, is notthe only way in which public R&D may enhance

5 Ž . Ž .See Griliches 1995 and Hall 1996 for recent surveys.Negative findings on the productivity growth payoff from govern-ment expenditures for industrial R&D emerged from an earlier

Ž .econometric studies by Griliches 1980 , Griliches and Lichten-Ž . Ž . Ž .berg 1984 , Bartelsman 1990 and Lichtenberg and Siegel 1991 ,

some of which obtained coefficients on federally funded R&Dthat were close to zero as well as statistically insignificant.

( )P.A. DaÕid et al.rResearch Policy 29 2000 497–529 499

economic performance. Public funding of R&D cancontribute indirectly, by complementing and hencestimulating private R&D expenditures, even if it hasbeen undertaken with other purposes in view. Gov-ernment agencies sponsor some research and devel-opment projects and programs because the knowl-edge gained is expected to be germane to theirrespective mission capabilities, as often is the case,for example, in areas such as military technology andlogistics, and public health. This kind of R&D worksometimes will be assigned to the staffs of publicinstitutes and national laboratories, although itequally may be procured through government con-tracts with R&D-performing firms in the privatesector. 6 Beyond its putative direct value as an inputinto the provision of government-provided services,both the defense-related and civilian R&D expendi-tures funded through public agencies may generatesocial benefits, in the form of knowledge and train-ing ‘‘spillovers.’’ These often are held to enhanceprivate sector productive capabilities, and, specifi-cally, to encourage applied R&D investments byfirms that lead to technological innovations — fromwhich will flow future streams of producer andconsumer surpluses. 7

The theoretical plausibility of such claimsnotwithstanding, available empirical evidence on theissue remains rather short of being conclusive, to say

6 For example, the 1996r1997 data compiled by StonemanŽ .1999, Tables 1 and 6 show that the UK government funded

Ž .31.8% of total military and civil R&D, somewhat less than halfŽ .of which 14.4% of total R&D was performed in government

departments and laboratories run by the Research Councils. FromŽ .the National Science Board 1998, Appendix Table 4-3 figures

Ž .for the US in 1996, the corresponding federal and non-federalgovernment shares in funding and performance are seen to be32.5% and 8.9%, respectively. By adding to the latter figure the3.3% of federal government-funded R&D that was performed innon-profit federally funded research and development centersŽ .analogous to the research units of the UK Research Councils , wearrive at 12.2% for the overall share of total R&D that wasperformed in US government research facilities. The latter, likethe US governmental share in total R&D funding, rather closelyresembles the contemporary situation prevailing in the UK.

7 See, for example, the recent formulation of the economic casefor public support of research, in National Research CouncilŽ .1999 , especially Chaps. 1–2, with historical case studies drawnfrom the US federal government’s role in the development ofcomputing and communications technologies.

the least. Economists, continuing in the traditionŽ .pioneered by the research of Blank and Stigler 1957 ,

recurrently examine a variety of data for signs as towhether the relationship between public and privateR&D investments is on balance characterized by‘‘complementarity,’’ or by ‘‘substitution.’’ Severalrecent econometric studies, for example, documentpositive, statistically significant ‘‘spillover’’ effectsvia the stimulation of private R&D investment bypublicly funded additions to the stock of scientificknowledge. 8 The same might be said regarding aconsiderably more extensive body of historical casestudies, detailing the influence of government-spon-sored research programs and projects on commercialtechnological innovation. 9 Many among the latterstudies, however, focus on US federally funded re-search performed in academic institutions or quasi-academic public institutes, and so do not bear imme-diately on the questions raised concerning the im-pacts of publicly sponsored R&D conducted undercontract by industrial corporations. Nor do they in-form us about the effects of publicly fundedmission-oriented commercial research in the rest ofthe world. Moreover, while some studies in this areahave been able to support claims of positive spilloversfrom public to private expenditures, there is noshortage of investigations that arrive at the contraryconclusion. Thus, it is found that some public R&Dcontracts actually have done little or nothing topromote the efficient functioning of the governmentagencies involved, and yet also failed to providesignificant commercial ‘‘spillovers.’’ In still otherinstances, the benefits that private companies derivedfrom the public R&D expenditures are said to havebeen both predictable and large enough to haveelicited financing by profit-seeking firms, had thepolitical process not invoked subsidization of thoseprojects at the tax-payers’ expense. 10 Whereverpublicly funded R&D is seen to be simply substitut-

8 That, at least, is the inferential interpretation of the resultsŽ . Ž . Ž .reported by Jaffe 1989 , Adams 1990 , Acs et al. 1991 and

Ž .Toole 1999a; b . See further discussion in Section 3.4.9 Among recent, sophisticated contributions to this literature,

Ž . Ž .see Link and Scott 1998 and National Research Council 1999 .10 Ž .See, e.g., the examination of Cohen and Noll 1991 of a

selection of large-scale mission-oriented commercial R&D pro-grams that were funded by the US federal government.


ing for, or actually ‘‘crowding-out’’ private R&Dinvestment, it obviously is hard to justify such ex-penditures on the grounds that they exerted an imme-diate net positive impact upon industrial innovationand productivity growth. 11

Simply counting up the numbers of findings proand con that have accumulated on the issue ofpublic–private R & D complementarity since themid-1960s, however, cannot be very informative.Our approach instead will be to survey the availablebody of econometric work systematically, and insome detail, from an analytical perspective. Al-though we take notice of a number of time-seriesstudies that have been carried out at the macroeco-nomic level, most of this inquest is concerned withresearch that focuses on the impact of public R&Dcontracts and grants upon private R&D investmentby manufacturing firms and industries. It is there thatthe bulk of R&D expenditures by the world’s devel-oped economies continues to be concentrated. Ourpurpose in this is to assess the reliability of thestatistical findings and to arrive at a better under-standing of the reasons for the persisting lack of aclear-cut empirical consensus in the literature.

Three quite restricted questions will be askedregarding those investigations. First, is the design ofthe statistical analysis such that it can yield anyreliable findings on the question of whether govern-ment R&D expenditures do or do not have a signifi-cant and economically palpable impact upon theirprivate sector counterparts? Secondly, where thefindings are credible, may we conclude that govern-ment subsidy programs do not displace private R&Dinvestment, but instead have the complementary ef-fect of inducing additional company-funded R&Dactivities? Thirdly, how can the econometric findingsbe reconciled with those of other well-designed stud-ies that addressed ostensibly the same question, yetarrived at different conclusions?

At this time, the econometric results obtainedfrom careful studies at both the micro- and macro-

11 It remains conceivable, however, that some special features ofthe government-sponsored projects create capabilities in the per-forming firms that are conducive in the longer run to increasedprivate R&D investment, to higher marginal innovation yields, orto both.

levels tend to be running in favor of findings ofcomplementarity between public and private R&Dinvestments. But, that reading is simply an un-weighted summary based upon some 30 diverse stud-ies; it is not a conclusion derived from a formalstatistical ‘‘meta-analysis,’’ and in no sense is itoffered here as a judgement that would pretend tosettle the issue definitively. To formally weigh up

Žand aggregate the available and still-growing array.of statistical analyses seems to us a virtually impos-

sible task in this case. We are not dealing withstatistical results that have been generated by prop-erly designed ‘‘experiments’’, where provision wasmade in the policy process for replication and ‘‘con-trols.’’ Instead, we are dealing with ex-post in-quiries, and the results reported by many of theindividual papers that constitute the literature on thistopic reflect a convolution of many counterbalancingeffects that are further compounded with the effectsof a varying mix of public funding and other incen-tives for R&D activities. The ability of the econome-tricians to impose ex-post statistical controls varieswidely among these studies. Moreover, they aredistributed over differing time periods, and across avariety of scientific and technological fields, as wellas diverse sectors and different economies.

Inasmuch as the spheres of investigation as wellas the findings considered here are far from uniform,it is difficult to see what good would be served bystriving for a broad empirical generalization thatmight mask clear-cut instances, however few, wherepublicly funded R&D is found substantially to dis-place private investment. Indeed, the better way ofproceeding would seem to lie in trying more pre-cisely to identify and delineate the characteristics ofthe circumstances in which ‘‘substitution’’ effectspredominate. Policy making in this area of growinglong-term importance calls for more specific empiri-cal support and guidance if it is to advance beyondgeneral theoretical arguments, intuitive practicaljudgements, and political rhetoric.

1.2. Cautions regarding the surÕey’s scope and limi-tations

It should be made explicit at the outset that thepresent review is addressed to only one aspect of the


broader empirical picture that is of interest for publicpolicy formation. For one thing, we do not examinethe large body of evidence on the relative productiv-ity impacts of public and private R&D. 12 A secondissue that we do not treat in any detail is the otherside of the interdependence of public and privateR&D spending, namely, the latter’s impact upon theformer. Many of the micro-level empirical studieswe have surveyed treat public R&D either explicitlyor implicitly as an exogenous influence on privateR&D within an investment framework. Conse-quently, in a number of instances we do find itnecessary to point out the econometric consequencesof ignoring the existence of latent variables that mayjointly effect both public- and private-sector deci-sions to allocate R&D funding to specific industrialareas, and to have the work performed by particularŽ .rather than randomly drawn firms. Our discussionrecurrently touches on this point, arguing that moreattention to structural modeling of governmentagency behavior as well as industrial R&D re-sponses is needed for a proper interpretation of theoverall, reduced-form findings. 13 This may be seenas an instance of the more general case that David

Ž .and Hall 1999 advance for taking an explicit struc-tural modeling approach to mitigate the frequency ofapparent contradictions and ambiguities in theeconometric literature.

1.3. Organization

The remaining presentation is organized as fol-lows. Section 2 presents a conceptual framework forunderstanding the net effects of public R&D uponprivate R&D investment activities. Section 3 re-views and critiques the available econometric re-search findings, beginning with studies carried outusing data for the line-of-business and laboratory

Ž .level Section 3.1 , and progressing upwards to those

12 More recent work on this question is surveyed by Klette et al.Ž .2000 .

13 Ž .The studies by Leyden and Link 1991 and Leyden et al.Ž .1989 are noticed as having set a good example for future workin this regard.

Žconcerned with effects at the level of the firm Sec-. Ž .tion 3.2 , the industry Section 3.3 and the aggre-

Ž .gate economy Section 3.4 . Because the bulk of theeconomic research on this question looks at publiclyfunded R&D that is being performed under theterms of government contracts with commercialfirms, the main focus of discussion in this surveyfalls upon programs of that kind. In Section 3.5,however, notice is taken also of a small body ofeconometric studies that examine the impacts onprivate R&D of publicly fundedrpublicly performedresearch or publicly fundedrnon-profit performedresearch. 14 We conclude in Section 4 with severalmethodological observations and suggestions for fu-ture research in this area of perennial policy rele-vance.

2. ‘‘Net’’ private R&D effects of public R&D: aconceptual framework

Before proceeding further, it is appropriate topause to ask what modern ‘‘technology policy mea-sures’’ have been meant to achieve. Presumably thecentral rationale for government support of R&D isthe correction of the market failures in the produc-tion of scientific and technological knowledge, aris-ing from the ‘‘incomplete private appropriability’’

Ž .problems identified by Nelson 1959 and ArrowŽ .1962 . Economists have indicated two main policyresponses to the resulting tendency towards under-provision of knowledge-based innovative effort onthe part of profit-seeking business entities: directprocurement andror production in public facilities,and incentives for a greater amount of private invest-ment. We have here eschewed issues concerning thefirst of these, primarily those involving the perfor-mance of public research institutes and national labo-

14 Ž .Recent studies by Jaffe and Trajtenberg 1996 and Jaffe et al.Ž .1998 use patent citations to investigate the flow of knowledgeand commercial technology out of federal labs, but our discussionis restricted to papers that explicitly deal with the impacts uponprivate R&D investment.


ratories, and have restricted our review exclusivelyto the second class of policy responses.

2.1. Tax incentiÕes Õs. direct subsidies for R&D

Under that heading two main policy instrumentsmay be identified: tax incentives that reduce the costof R&D, and direct subsidies that raise the private

Ž .marginal rate of return MRR on investment in suchactivities. Although not strictly necessary, the pri-mary difference in execution between these two pol-icy instruments is that the former typically allows theprivate firms to choose projects, whereas the latterusually is accompanied by a government directedproject choice, either because the government spendsthe funds directly or because the funds are dis-tributed via grants to firms for specific projects orresearch areas.

The effectiveness of R&D tax credits in increas-ing R&D in private firms is surveyed in Hall and

Ž .van Reenen 1999, this issue . Because a tax creditdirectly reduces the marginal cost of R&D, onewould not expect to see ‘‘crowding out’’ effects onindustrial R&D, spending except via the channel of

Žan increase in the real cost of R&D if the inputs are.in inelastic supply . This implies that crowding out

of nominal private-sector R&D expenditures wouldnot be observed, even though it is entirely possiblethat there could be displacement of private realR&D investment if the prices of the inputs increasedsufficiently. 15 Tax credits, however, do not leavethe composition of R&D unaffected. As firms ex-pand their R&D activity in response to linked taxoffsets against earnings, the incentives are likely tofavor projects that will generate greater profits in theshort-run. Consequently, projects with high socialrates of return, and long-run exploratory projects and‘‘research infrastructure’’ investments in particular,may be less favored by the expansion of privatefunding. In this way, rather weaker ‘‘spillover’’ ben-efits to other firms and industries would be generated

15 Ž .See David and Hall 1999 for a full discussion of the impactof inelastic R&D inputs on the effects of R&D subsidies.

by the private response to extensive reliance uponthis particular pro-R&D policy instrument.

By contrast, direct funding of R&D programsdesignated by government agencies allows publicR&D subsidies to be targeted toward projects thatare perceived to offer high marginal social rates ofreturn to investments in knowledge. At least in prin-ciple, such funding could be concentrated in areaswhere there was a large gap between the social andthe private rate of return. For this reason, directR&D subsidies or government spending on basicresearch activities should not be expected to displaceprivate real R&D investment, except via its genericimpacts on the price of research and developmentinputs that are in inelastic supply. Yet, the possibilityremains that in the politics of technology policyformation, there will be strong pressures to providesubsidies for projects with high private marginalrates of return — possibly to assure the appearanceof successful public ‘‘launch aid’’, or simply be-cause the prospective private payoffs make lobbyingfor subsidies an attractive undertaking. In such cir-cumstances, it is more likely that increased directgovernment funding for industrial R&D projectswould enable firms correspondingly to reduce theirown outlays. This form of ‘‘investment displace-ment’’ arises primarily because R&D activities areheterogeneous rather than homogeneous, and it isdistinguishable from, and additional to the form of

Ž .‘‘crowding out’’ identified by David and Hall 1999as operating through the R&D input market effectsof public expenditures.

2.2. The need for a structural framework

When considered as a whole, the literature underreview here may be characterized as predominantlyinductive in its approach to considering the effects ofgovernment R&D funding upon the level of busi-ness R&D investment behavior. That is not simplyto say that the orientation of this work has beenprimarily empirical rather than analytical, but, ratherthan the empirical approach pursued is essentiallydescriptive in nature, aiming at establishing the signand magnitude of the overall or ‘‘net’’ effect inquestion. Few studies in this area have been espe-


cially concerned to delineate the different channelsof influence that may connect R&D resource alloca-tion in the two spheres, and fewer still attemptstructural estimation in order to ascertain the charac-ter and strength of the underlying effects associatedwith each such channel.

Although all these empirical studies acknowledgebeing motivated by the same over-arching policyissue, the widespread tendency to eschew explicitstructural modeling has reinforced the common fail-ing of econometric work in this field to specifyadequately what ‘‘experiment’’ the investigators im-plicitly envisage conducting. In other words, suppos-ing one could vary government policy, what ob-

Ž .served pairings of policy action s and sequelaewould establish whether public and private R&Dwere ‘‘complements’’ or ‘‘substitutes?’’ Taken to-gether with the fact that a large number of verydifferent empirical ‘‘experiments’’ at various levels

Ž .of aggregation are contemplated in effect by theresearch literature, this lack of specification con-tributes to the difficulties of interpreting the individ-ual findings and reconciling the seeming contradic-tions among them.

For the foregoing reasons, rather than out of anymethodological precommitment to favor structuralmodeling approaches in econometric analysis, webelieve it will be best to review the evidence pre-sented by the individual studies only after having setout a general conceptual framework that identifiesthe array of hypothesized micro-level determinantsof private sector R&D investment, and relates thesein turn to relationships that hold and manifest them-selves at the macro-level. That is our task in thepresent section.

2.3. Determinants of priÕate R&D inÕestment at themicro-leÕel

A useful framework for understanding how publicR&D affects R&D funding decisions in the privatesector is provided by an adaptation of a familiar,rather elementary model of firm-level investmentbehavior. To our knowledge, such a framework wasfirst employed for this purpose by Howe and

Ž .McFetridge 1976 . It postulates that, at each point inŽ .time or for each planning period , an array of

Fig. 1.

potential R&D investment projects is available. 16

The firm is assumed to rationally consider the ex-pected cost and benefit streams for each project, inorder to calculate its expected rate of return. Undercertain conditions, these can be thought of as internalrates of return and therefore used by the firm inquestion to rank the associated projects in descend-ing order of anticipated yield, thereby forming itsMRR schedule.

A downward-sloping schedule of this kind ap-Žpears in Fig. 1, where the marginal yield and the

.marginal cost of capital is plotted on the verticalaxis, and the horizontal axis gives the cumulatedamount of investment required as one proceeds down

Žthe list of projects. Following expositional conven-tion, each project is implicitly taken as being finelydivisible, so that the resulting MRR schedule is

.continuous and continuously differentiable . Under

16 This formulation abstracts from important issues concerningthe determinants of the firm’s access to the scientific and engi-neering knowledge base that is relevant for formulating plausiblyfeasible R&D projects, and estimating the time distribution of thecosts and benefits of the innovations they would generate. There isa well-known recursion problem here, inasmuch as among theresearch projects that a rational decision process would need toconsider is the project for gaining the knowledge required toconstruct and evaluate its current ‘‘innovation possibility set’’.But, in a full dynamic specification, it is straightforward analyti-cally to treat the latter as a lagged endogeneous variable.


this construction, as one moves along a given sched-ule describing the distribution of projects in thefirm’s prevailing ‘‘technological innovation possibil-ity set,’’ there is no alteration in the constellation ofother variables that would influence the rates ofreturn on the array of R&D projects in the firm’spotential portfolio. The net impact of any and allalternations in those other conditions, therefore, mustshow up as shifts in the MRR schedule.

As also may be seen from Fig. 1, the firm faces aŽ .marginal cost of capital MCC schedule, which

reflects the opportunity cost of investment funds atdifferent levels of R&D investment. 17 The upwardslope of this schedule over its full range is at-tributable to the fact that as the volume of R&Dinvestment is increased the firm will have to movefrom financing projects with internally generated

Ž .funds i.e., retained earnings to calling upon exter-Ž .nal equity and debt funding. Use of retained earn-

ings for R&D accounts for the flat range at the leftof the MCC schedule, whereas the firm’s increasedrecourse to external financing would tend to push itsmarginal costs of capital upwards. 18

It should be apparent that as the MCC schedule inFig. 1 describes the opportunity cost of capital, itwould slope upwards eventually. This must be soeven were it the case that all of the firm’s R&Dinvestment remained financed out of retained earn-

17 This implicitly holds constant the amount of other, tangiblecapital formation expenditures that the firm has scheduled for theplanning period in question. Although the assumption of riskneutrality on the part of the firm is implied by the use of theexpected MRR as a sufficient statistic to describe each project inthe portfolio, it should be recognized that in practical capital-budgeting exercises firms add premia to their marginal costs ofcapital, forming ‘‘hurdle rates of return’’ that allow for theriskiness of various classes of investment. Although, for exposi-tional simplicity it has been supposed that the tangible capitalformation budget has been predetermined, at the margin R&Dshould compete with all other capital projects on the basis of theirrisk-adjusted internal rates of return.

18 Obviously, in the case of R&D intensive ‘‘start-ups’’ thereare no retained earnings upon which to draw. But, the possibilityexists of borrowing capital from employees by paying them withstock option, which may keep the marginal cost of capital downso long as there is an adequate supply of qualified personnel whoalso happen to have a high tolerance for risk, although issuance ofstock options does have a cost due to its effect in diluting equity.

ings; at the margin, expansion of the R&D invest-ment budget would force the firm to turn to externalfinancing for its tangible capital acquisition projects.

The foregoing simplified schema can be repre-sented by the following equations:

MRRs f R , X , 1Ž . Ž .

MCCsg R ,Z , 2Ž . Ž .

where R is the level of R&D expenditure, and Xand Z are vectors of other ‘‘shift variables’’ thatdetermine the distribution of project rates of returnand the associated marginal costs of capital, respec-tively. The X-variables reflect:Ž .i The ‘‘technological opportunities’’ governingthe ease with which it is possible to generate

Ž .innovations relevant to the firm’s market area ;Ž .ii The state of demand in its potential marketarea or line-of-business;Ž .iii Institutional and other conditions affecting the‘‘appropriability’’ of innovation benefits.

Correspondingly, the Z-variables include:Ž .i Technology Policy measures that affect the

Žprivate cost of R&D projects such as the taxtreatment of that class of investment, R&D subsi-dies, and cost-sharing programs of government

.procurement agencies ;Ž .ii Macroeconomic conditions and expectationsaffecting the internal cost of funds, via the generalstate of price-earnings ratios in equity markets;Ž .iii Bond market conditions affecting the externalcost of funds;Ž .iv The availability and terms of venture-capitalfinance, as influenced by institutional conditionsŽ .such as the development of IPO markets and thetax treatment of capital gains.As is depicted by Fig. 1, in the firm’s profit

maximizing equilibrium the optimal level of R&Dinvestment is found at RU , where the MRR and themarginal cost of funds are equalized:

RU sh X ,Z . 3Ž . Ž .

Several points are now obvious about the relation-ship between private R&D investment and public


R&D funding. First, if we take the provision ofpublic funds to be exogenous, the effects of such‘‘shocks’’ would show up as a shift of either thefirm’s MRR schedule or its MCC schedule, or ofboth. For example, direct R&D subsidies, and cost-sharing arrangements by public agencies, by reliev-ing the firm of some joint costs of research anddevelopment activities would be tantamount to shift-ing the position of its MCC schedule to the right.Had the firm initially been facing increasing marginalcosts of capital, this change would permit the under-taking of additional projects with its own money —other things being equal, of course. 19 To cite an-other specific illustration, the award of governmentR&D contracts to a small firm also might have theeffect of lowering the recipient’s capital costs at themargin; especially in the case of start-up enterprises,where this could act as a signal for external fundingsources to apply a smaller risk premium when settingtheir lending terms.

A number of other potential positive micro-leveleffects of government contracts for industrially per-formed R&D, also have been noticed in the litera-

Ž .ture following Blank and Stigler 1957 . Each of thefollowing three factors would appear as an outwardshift of the MRR schedule in Fig. 1.

Ž .a Publicly subsidized R&D activity can yieldlearning and training effects that acquaint the enter-prise with the latest advances in scientific and engi-neering knowledge, and so enhance its efficiency inconducting its own R&D programs.

Ž .b Where public funds are made available forconstruction of test facilities and the acquisition ofdurable research equipment, and also pay the fixedcosts of assembling specialized research teams, thefirm involved may be able to conduct further R&D

Ž .projects of its own at lower incremental cost, andthereby derive higher expected internal rates of re-turn on its R&D investments.

Ž .c Government contract R&D, by signaling fu-ture public sector product demand, and private sector

19 It should be obvious, however, that were the firm facing acompletely inelastic MCC constraint, the public contract fundingwould not have any positive incremental impact upon the level ofcompany-funded R&D.

demand in markets for dual-use goods and services,may raise the expected marginal rates of return onproduct or process innovation targeted to those mar-kets.

There is a distinct possibility that in the case ofŽ . Ž .the above-mentioned effects a and c , the techno-

logical knowledge and market information associatedwith publicly funded R&D performed by one firmcould result in ‘‘spillovers.’’ The latter would, simi-larly, raise the expected marginal rates of return forother firms in the same industry, and also for firmsin other industries. Public R&D performed in aca-demic and other non-profit institutions, includinggovernment laboratories, also could have correspond-ingly positive spillover effects. This is so particularlywhere the research resulted in the development of‘‘infrastructural knowledge’’ — general principles,research tools and techniques, and skill acquisitionthat raised the expected rates of return on commer-cially oriented, applied R&D projects. 20

2.4. Distinguishing between goÕernment R&D con-tracts and grants

When considering the potential ‘‘net’’ effects ofpublic R&D activities, it is likely to be important todistinguish between public contracts and grants.Government R&D contracts in most instances arefinancial outlays to procure research results that areexpected to assist the public agency in better defin-ing and fulfilling its mission objectives. Such con-tracts are the largest component of public awardsmade to private for-profit firms, and also include allarrangements to purchase R&D- intensive publicgoods. This category covers much of the publicaerospace and defense expenditure. Public grants, onthe other hand, are usually competitive financialawards that do not carry any future public commit-ment to purchase. They are the primary mechanism

20 Ž . Ž .Leyden and Link 1991 and David et al. 1992 suggest avariety of ‘‘spillover’’ channels through which the infrastructure-forming aspects of so-called ‘‘basic’’ research funded — and insome cases performed — in the public sector has a complemen-tary impact upon private sector R&D investment.


for funding exploratory research for the advancementof knowledge and fostering emerging technologies.

Since the path-breaking work of Blank and StiglerŽ .1957 , four channels have been identified in theliterature through which public contracts for industri-ally performed R&D could stimulate ‘‘complemen-tary’’ private R&D expenditures in the short run:Ž .i Public R&D contracts increase the efficiencyof the firm’s R&D by lowering common cost orincreasing ‘‘absorptive capacity;’’Ž .ii Public R&D contracts signal future demand;Ž .iii Public R&D contracts may improve thechances for success on the firm’s other projects;Ž .iv Public R&D contracts allow firms to over-come fixed R&D startup costs — ‘‘pump-prim-ing.’’These imply that public subsidies either shift the

Ž .firm’s marginal returns schedule in Fig. 1 out to theright, andror that the firm’s opportunity cost ofcapital schedule is shifted to the right, thereby even-tually lowering the firms’ MCC for the higher levelsof R&D expenditure. The first three of the foregoingputative effects would shift the firm’s MRR schedulerightwards either by increasing the expected rev-enues or decreasing the expected costs of the firm’savailable projects. The fourth effect, by decreasing afirm’s fixed costs, lowers the opportunity cost ofcapital and shifts this curve out to the right. All ofthese ‘‘complementarity effects’’ suggest that publicR&D contracts stimulate additional private R&Dinvestment.

From the preceding discussion, it might appearthat the micro-level impact on private R&D invest-ment of both government contracts for industrialR&D and grants awarded to non-profit organizationswould be unambiguously positive. But, there are twosorts of countervailing influences, both of which arelikely to operate more strongly in the case of con-tracts. First, the performance of contract-specifiedlines of R&D with public funding may simply sub-

Ž .stitute for some if not all of the investment that theperforming firms otherwise would have been pre-pared to undertake in order to be in a position to bidsuccessfully for related government procurementcontracts.

Secondly, publicly funded R&D also may mili-tate against private sector investments in the sametechnological areas, because the expected rates of

return to investments by firms that do not receivecontracts tend to be lowered by the prospect thatgovernment contractors would succeed in producingcommercially exploitable innovations. Doing socould leave them well positioned to enter the finalproduct market with significant first-mover advan-tages. Non-contract receivers also might be discour-aged from undertaking their own R&D by the antici-pation that the government procurement agency inquestion would have an incentive to disseminatecost-saving and quality-enhancing innovations, as ameans of enabling entry and greater competition inthe end-product market. When viewed from the latterperspective, ‘‘dual-use’’ programs of governmentprocurement of R&D-intensive goods take on theappearance of a two-edged sword. 21

Both the direct and the indirect ‘‘displacement’’effects just considered may be conceptualized asaltering the shape of the firms’ respective MRRschedules, reducing expected marginal returns onR&D projects belonging to particular technologicalareas that were ‘‘targeted’’ for public contract sup-port. Although it is clear in principle that the policyprescription should be for the government to selectprojects for subsidization that the private sector isnot likely to undertake, or not undertake in sufficientvolume, matters may be otherwise in actual practice.Pressures within public agencies for high ‘‘successrates’’ in contract awards may lead to the use inR&D funding decisions of selection criteria that putheavy weight on factors that are correlated positivelywith high expected rates of return to private R&Dfunding. Therefore, when investigating the net ef-fects of government-funded R&D at the micro-level,it is important to distinguish between programs that

21 Government procurement costs may be reduced by takingadvantage of spillovers from industry-funded R&D directed to-wards civilian products. But the potential for the R&D performedunder cost-plus procurement contracts to have spillover effects oncompany financed R&D might, correspondingly, be nullified bythe heightened anticipation of competitive entry into the businessof exploiting dual-use designs. Such opportunities may exist wherenew high-tech systems required by government have components,or utilize methods applicable to the production of goods forprivate purchasers. On the benefits of dual-use technology pro-

Ž .grams, see Branscomb and Parker 1993 .


provide grant funding and those that involve con-tracts. Likewise, when dealing with questions con-cerning aggregate level effects of changes in policiesaffecting public R&D, one should make allowancefor the effects of any significant alterations in thedistribution of funding between those two modes.

Although government grants typically do not haveŽa final product demand-increasing component such

as is frequently present in the case of public con-.tracts for R&D performance , they may cause the

MRR schedule to be shifted upwards nonetheless.This would occur because a program of grant-fundedresearch had raised firms’ R&D efficiency, or hadimproved the risk-return pattern on other projects.The convention in the literature, however, has beento abstract from the ways in which these effects ofgrant-type funding for industrial R&D would im-pinge upon the shape or position of firms’ MRRschedules, and so to identify whatever effects ensueas produced by shifts in the MCC schedule.

Three main analytical cases have been delineated.In the first, it is assumed that the firm is asset-con-

Ž .strained and thus faces a perfectly inelastic verticalMCC schedule at its current level of R&D invest-ment. The award of a subsidy in the form of a publicgrant then shifts the MCC curve to the right, increas-ing the firm’s performance of R&D by just the fullamount of the subsidy. The second case postulatesthat the public grant shifts an upward sloping MCCcurve to the right, so that the amount of the incre-mental increase in the amount of R&D the firmundertakes increases by less than the grant award.The third case considers that the MCC schedule is

Ž .perfectly elastic horizontal at the pre-grant equilib-rium, but is shifted downwards because the signal toequity holders provided by the public grant awardlowers the firm’s internal cost of funds. The magni-tude of the increase in private R&D investment willthen depend on both the strength of the ‘‘signal’’,which is likely to vary directly with the relative sizeof the grant, and on the slope of the firm’s MRRschedule. Other things being equal, the ‘‘flatter’’ isthe MRR schedule, the greater will be the increase inthe induced amount of private R&D investment.

Thus, only the last of these speculative situationsenvisages the possibility of a complementarity effectof public grants for industrial R&D, i.e., one thatwould elicit additional private R&D expenditures.

Grants to firms for R&D are likely to be used inways that assure greater private appropriability of thebenefits than is the norm for grant-funded research inacademic institutions, and similar non-profit researchinstitutes. There consequently may be justificationfor having presumed, in the foregoing analysis, thatcontracts yield no positive spillover effects that wouldinduce significantly increased private R&D invest-ment. But, by the same token, the same presumptionshould not be extended to considerations of theimpact of all public sector grant funding.

2.5. Short-run ‘‘net’’ impacts on priÕate R&D: frommicro- to macro-leÕel effects

In general, the likely direction of net effects ofpublic R&D contracts on private R&D investmentremains ambiguous. The previous discussion has re-viewed an array of channels at the micro-levelthrough which public contracts as well as grantswould have positive effects on the level of privatelyfunded R&D activity. On the other side of theledger, however, two principal arguments have beenadvanced on behalf of supposing that public expendi-tures for industrial R&D would exert a ‘‘crowdingout’’ effect on private R&D investment. The first ofthese is simply the micro-level displacement of fund-ing, previously discussed. This would occur wherecontracts are targeted in areas of technological devel-opment that firms otherwise would still find it worth-while undertaking; the resulting alteration in theshape of the MRR schedule may be such as to pushit downwards and to the left.

The second argument introduces macro-level con-siderations. There is likely to be upward pressure onthe prices of R&D inputs when the provision offunding to a particular firm or group of firms occursin the context of an expanded government R&Dprogram that absorbs substantial scientific and engi-neering personnel, along with other specialized mate-rials and facilities. The resulting increased costsassociated with the array of potential private R&Dprojects implies a lowering of the MRR schedule;and, other things being unchanged, that translatesinto a reduced level of business R&D investment.

Where will the balance be struck between theopposing forces arising from increased public sectorR&D expenditures at higher levels of aggregation


— i.e., in technologically related industry groups,and the economy as a whole? 22 This question re-cently has been examined analytically by David and

Ž .Hall 1999 . Their basic proposition is simple: when-ever the market supply of R&D inputs is less thaninfinitely elastic, as is likely to be the case in theshort-run, increased public sector demands for thoseresources must displace private R&D spending, un-less it gives rise to ‘‘spillovers’’ that also raise theaggregate private derived demand for R&D inputs.In the simple two-sector model developed by David

Ž .and Hall 1999 , the nature of the macro-level rela-tionship between private and public R&D invest-ment depends upon four parameters of the system.Complementarity, rather than substitution effects arelikely to dominate where the relative size of thepublic sector in total R&D input use is smaller,where the elasticity of the labor supply of qualifiedR&D personnel is higher, where the grant–contractmix of public outlays for R&D performance isskewed more towards the former, and where the rateat which the private marginal yield of R&D de-creases more gradually with increased R&D expen-ditures.

Without having fully specified both the magni-tudes of the elasticities, and the shifts in schedulesdue to spillover effects, in an analogous manner forthe microeconomic framework depicted by Fig. 1,the foregoing review of the static qualitative argu-

22 Attention should be called to the analytical difficulty ofpassing explicitly from the micro-level framework of the previoussubsections to the macro-level. In principle it would be possible toconstruct an aggregate private marginal efficiency of R&D invest-

Žment schedule, from the union of all the projects each with their.individually perceived expected internal rates of return . As that

might well involve duplicative investments in some projects, itshould be evident that the private expectations would generallynot be realized. The consistent aggregate private marginal effi-ciency of investment schedule would be lower, even with spilloversit is likely to lie below the aggregate social marginal efficiency ofinvestment schedule. But the real difficulty lies in passing fromthe aggregate private MRR schedule to the aggregate demand forR&D investment when firms are realistically heterogeneous. Thedistribution of projects is not identical across firms, and neither dothey all face the same MCC schedule. This means that there isnothing to guarantee that the same ranking of all the projectswould be selected by a central profit-maximizing agent chargedwith allocating the private sector’s total R&D funding.

ments for and against complementarity leaves oneunable to determine the sign of the net impact ofpublic subsidies on the level of business R&D ex-penditures. The general point that the foregoing dis-cussion does bring out clearly, however, is the pres-ence of identification problems due to the fact thatthe MRR and MCC curves may be shifting simulta-neously. This must be dealt with if econometricstudies are to succeed either in providing reliableestimates for the critical underlying elasticity param-eters, or in simply ascertaining the sign of the neteffect of public R&D contracts. It has been pointedout that both fixed costs associated with R&Dstartup, and resource constraint effects on input pricesthat are correlated with individual firms’ receipts offunding, may be shifting the MCC schedule. Byholding those effects essentially constant by the useof an appropriate econometric specification, includ-ing proper instrumental variables, it should be possi-ble to evaluate the net impact of public R&D con-tracts on private R&D investment demand. Theidentified effect would measure the net movement ofthe firm’s MRR schedule, holding fixed the opportu-nity cost of capital.

2.6. ‘‘Dynamic’’ or long-run effects of R&D subsi-dies

Even though most of the empirical literature hasbeen devoted to quantifying these presumed short-runŽ .‘‘static’’ effects, we should recognize at least two‘‘dynamic’’ or long-run effects that are partially theoutcome of public R&D funding. First, informa-tional spillovers from the advance of public scienceand engineering knowledge, much of which is madepossible by government R&D activities, will likelyshift the firm’s MRR schedule outward over time.Since new knowledge is the main source of newtechnological opportunities, the outward shift of theMRR curÕe assumes that these opportunities takethe form of higher project returns. Of course, sucheffects would be felt with variable lags and are likelyto be localized among some subset of the underlyingprojects. So, the relevant schedules would undergochanges in shape as well as position and the impactswill not necessarily be felt symmetrically throughoutthe population of firms.


A second dynamic effect stems from the trainingof new scientists and engineers. There is a strong andimportant emphasis in the US and UK public re-search enterprise on training, particularly within theresearch universities. Due to the trend toward in-creasingly heavy reliance upon foreign graduate stu-dents as research assistants on grant-funded aca-demic research projects, one must not simply pre-sume the existence of tight coupling between train-ing activities and the future availability of qualifiedresearch personnel in the labor markets where suchtraining occurred. But, insofar as there is a laggedinput supply response from expanded public funding

Ž .of R&D grants , this could show up at the microe-conomic level as an outward drift of the MRRschedule over time. 23 But, if one considers thesituation in the market for industrial R&D person-nel, it is more natural to conceptualize the aggregateeffect in terms of a downward shift of the laborsupply schedule. The latter would thus be a factormitigating such demand-driven upward pressures onreal unit costs of R&D that were set in motion bythe expansion of government funding. In general,then, the balance of the long-run dynamic effectsseems to favor the emergence at higher levels ofaggregation of net complementarities, rather than arelationship dominated by ‘‘crowding out’’, or thesubstitution of public for private R&D investment.

2.7. Endogeneity, and common latent Õariables ef-fects

Using this simple investment framework to clarifythe expected channels of influence helps to formulate

23 On the economic significance of the rising numerical impor-tance of foreign graduate students and post-doctoral fellows in

Ž .university research systems, see Dasgupta and David 1994 .There is substantial evidence, most of it accumulated from surveysof company executives responsible for research and development,that firms actively search for recent trainees of university depart-ments that have successfully drawn public funding for basicresearch; that they regard recent graduates as an important sourceof practical knowledge about the use of new techniques. See, e.g.,

Ž . Ž .Levin et al. 1987 and Pavitt 1991 . It remains unclear to whatextent this public goods spillover effect via researcher training isequally characteristic of university-based research supported byindustry funds, and conducted under restrictions typical of propri-etary R&D.

and evaluate the empirical ‘‘tests’’ that exist in theliterature. It does not, however, allow us to addressthe possible mutual interdependence of public andprivate R&D expenditures. This may present anissue for econometric analysis, either because ofsimultaneity and selection bias in the funding pro-cess, or because there are omitted latent variablesthat are correlated with both the public and privateR&D investment decisions. Endogeneity due to se-lection biases in R&D grants and subsidized loans tosmall and medium size firms has been addressed

Ž .recently in the work of Busom 1999 and WallstenŽ .1999 , whereas the possibility of omitted time-con-stant firm effects in the awarding of governmentR&D contracts was pointed out many years ago by

Ž .Lichtenberg 1984 . Lichtenberg also remarked onthe problem presented by the fact that firms under-take a significant amount of preparatory R&D inorder to qualify for government contracts and grants:this may create a situation in which the firm’s MRRschedule already has shifted outward in anticipationof a public R&D contract or grant, and consequentlythe firm’s response to the award may be more diffi-cult to detect in the data.

More generally, it may well be that there arestrong selectivity biases that lead firms which have arecognized competence for certain kinds of R&D toreceive public contracts as well as to fund suchactivities with their own money. Beyond that, incross-section analyses at the industry level, a distinctbut related econometric problem may arise whereboth private and public investment decisions areresponding to the same latent variable, namely, theinter-industry variation in the ‘‘technological oppor-tunity set.’’ The possibility of exogenous changes inthe state of the opportunities created for commer-cially attractive innovation — such as those openedup by developments in fiber optics, high-temperaturesuperconductivity, or the availability of restrictionenzyme techniques for ‘‘gene-splicing’’ — may con-found efforts to identify the causal impact of publicR&D allocations upon the pattern of private invest-ment.

Even though the framework presented here hasnot undertaken to formalize these and other, more‘‘politically implicated’’ sources of endogeneity inpublic R&D expenditures, it has assisted us in un-derscoring the need for empirical studies to be ex-


plicit about their identifying assumptions, and toinclude proper ‘‘control’’ variables. The foregoingdiscussion also has highlighted the point that publicR&D contracts are likely to have a much strongerimmediate effect on the firm’s marginal returnsschedule than is the case with public R&D grants.For that reason alone, public contracts are moredifficult to evaluate because those effects are readilyconfounded with the many other factors that shift theMRR schedule, including the nature of changes inthe production and product technology, appropriabil-ity conditions, the type of product market competi-tion, and so on.

As the previous discussion of identification prob-lems may have suggested, to undertake to estimatethe magnitude of the effect of public R&D spendingupon private R&D investment at different levels ofaggregation is tantamount to conducting rather widelydiffering ‘‘experiments’’ in the hope of determining‘‘the’’ response. Some considerable doubt must sur-round the very idea that there is a universal relation-ship of that kind, and so it will be better to avoidcasual comparisons and juxtapositions of findings,striving to compare like with like where that isfeasible. Macro-level time-series studies have to con-sider feedback effects operating through price move-ment in the markets for R&D inputs, whereas inmicro-level analyses the findings should reflect‘‘real’’ rather than nominal expenditure relationshipsbetween public and private R&D. At least thatwould be so once controls had been entered for time

Ž .effects in the form of year dummies . Of course, insaying this we assume that there is substantial poten-tial mobility on the part of R&D personnel amongfirms and industries. In other words, when ‘‘theexperiment’’ under analysis involves the provision ofa subsidy for R&D conducted by a particular firm, itis reasonable to assume that the firm faces a highlyelastic supply of R&D inputs, and therefore thatinput prices and unit costs cannot be materiallyaffected by the subsidy. 24 The implication is thatthe observed effect on the level of private R&D

24 This is precisely true only for the log–log specification,where the aggregate price effects are in the constant term, or whenthe number of firms receiving subsidies is a small fraction of thetotal so that there is no price effect in the cross-section.

spending should be somewhat weaker at the microe-conomic level than that found by studies conducted

Žusing aggregate data since the effects in the former.case are real rather than nominal .

Nevertheless, at the lower level of aggregationthere are likely to be additional complications due tothe presence of significant cross-sectional differencesin technological opportunity or innovation capabili-

Žties ‘‘competences’’ in the terminology preferred by.the recent management literature . Some controls for

‘‘fixed effects’’ may be appropriate in such cases.But, in studies based on firm- and industry-levelpanel data, over time the ‘‘innovation opportunitysets’’ may be undergoing differential alterationsamong the various technological and market areas.Simple, fixed effects methods are then likely toprove inadequate to the task, and more complexeconometric tactics would seem to be called for.

Thus alerted to the numerous underlying complex-ities that beset those in search of a straightforwardempirical answer to the question of whether ‘‘com-plementarity’’ or ‘‘substitution’’ prevails in this do-main, we may now be in a position to criticallyexamine the econometric findings which the litera-ture presently has to offer.

3. The empirical literature: review and critique

Our survey is organized in subsections, distin-guishing among the econometric studies according totheir choice of the unit of observation, and by thetype of data analyzed. Four types of observationalunits have been studied: line-of-business or labora-

Ž .tory, firm, industry, and national or domestic econ-omy. The typical econometric approach is to regresssome measure of private R&D on the governmentR&D, along with some other ‘‘control’’ variables.When a positive coefficient on the public R&Dvariable is found, this is interpreted as revealing thepredominance of complementarity between publicand private investments. On the flip-side, a negativecoefficient is taken to imply that public R&D andprivate R&D are substitutes. In several studies, theauthors use the magnitude of the estimate to makestatements to the effect that ‘‘a one dollar increase inpublic R&D funding leads to an X dollar increase


Ž .decrease in private R&D investment.’’ On thewhole, however, the magnitudes of the publishedestimates are very diverse. 25 In some instances, theyare very difficult to compare directly, owing tovariations in the specification of the estimated equa-tion, and the absence of collateral statistical informa-tion needed to calculate dimension-free parameterestimates, such as elasticities. Our summary tablespresent the estimated elasticity of private R&D withrespect to public R&D wherever it has been possibleto obtain this magnitude.

To achieve appropriate comparability among thefindings, we divide the studies into four main groups,according to the nature of the statistical observationsused and the econometric approach complied withthe following.

Ž .1 Pure cross-section studies at the micro level,where firms or industries with different levels ofgovernment R&D funding are compared. Here it iscrucial to control for differences in demand condi-tions, technological opportunity, and appropriability.

Ž .2 Panel data studies at the micro level within agiven industry, in which there are controls for time-invariant differences among firms — each firm ineffect serving as its own control, so that the resultsreflect individual firms’ time-series responses tochanges in government funding.

Ž .3 Aggregate or macroeconomic studies, wherethe response is identified by changes in private R&Dfunding over time as a function of government R&Dfunding. Here it is important to control for macroe-conomic influences that may be driving both vari-ables. In addition, it is likely that results based onthese studies will contain R&D input supply effects

Ž .of the sort that are identified by Goolsbee 1998 andŽ .David and Hall 1999 .

Ž .4 Studies, whether micro or macro, that attemptto control for the simultaneity between private andpublic R&D spending using instrumental variables.It is probable that the results from these studies will

25 Ž .For example, Wallsten 1999 concludes that there is a one-for-one crowding-out of private investment, whereas RobsonŽ .1993 concludes that there is a one-for-one stimulus of privateR&D investment. To be sure, these studies were not analyzing thesame dataset.

differ from those in the other studies if commonomitted variables are a problem.

The discussion in Section 2 urges caution wheninterpreting the results of some of these regressionstudies. For example, when we look across firms orindustries in a cross-section, we are seeing a set of

Ž Uequilibrium choices for the level of R&D R in.Fig. 1 , rather than tracing out the derived demand

curve for R&D as a function of the position of theMCC curve. In fact, each of the firms and industriesin question is likely to face a different demand curvefor R&D investment, as well as a different MCCschedule. In addition, some of the effects of thepublic support for R&D will be to shift or otherwisechange that curve. Many, but not all, researchershave attempted to control for the variability in theMRR curve when estimating the relationship, and inwhat follows we will try to assess the success of theapproaches that they have taken in doing so.

Ž .Blank and Stigler 1957 suggested in their origi-nal formulation of the question that the most effi-cient and direct way to test for a complementary orsubstitution relationship is to analyze a sample ofresearch programs over time within a ‘‘suitable’’sample of companies. Due to data limitations, how-ever, they were forced to rely on a cross-section ofmanufacturing firms in their analysis. Their methodwas to compare the ratios of scientific workers to allemployment for firms with and without governmentcontracts. For a total sample of 1564 firms in 1951,they found that firms with public contracts also hadlower scientific personnel intensity working on pri-vate research. This result supports ‘‘substitution,’’ atleast in the sense that public contracts were associ-ated with fewer research personnel on private pro-jects. When Blank and Stigler changed the samplinguniverse, however, and took it to be US manufactur-ing firms engaged in any R&D, public androrprivate, they reported that ‘‘in general substitution isnow almost absent.’’ 26 Finally, using their mostreliable data for firms with more than 5000 employ-ees, they found evidence for complementarity. Theseresults are consistent with the view that althoughmany individual firms may find it attractive to sub-

26 Ž .Blank and Stigler 1957 , p. 61.


stitute government funding for their own R&D bud-gets, the large firms are better able to take advantageof complementarities, due to knowledge spilloversand pump-priming effects. The latter may operateacross lines-of-business and even standard industrialclassification lines, so that the size of those firmsmay really be a surrogate for the product diversifica-tion that enables them to appropriate benefits fromthe less predictable range of their R&D projects.Blank and Stigler, however, warned that their esti-mate based on variation across firms and industriesat a point in time could be seriously biased by othersources of heterogeneity, for example, variations intechnological conditions faced by different firms.

3.1. Line of business and laboratory studies

Ž .The subsequent research of Scott 1984 and Ley-Ž . Ž .den et al. 1989 ; Leyden and Link 1991 , together

with very recent work using Norwegian data byŽ .Klette and Moen 1998 , are the only studies that

follow the recommendation of Blank and StiglerŽ .1957 to look ‘‘below’’ the firm level. For each ofthese studies, Table 1 lists the sample used, theeconometric methodology and type of data, the formof the private R&D variable to be explained, theform of the public R&D explanatory variable, aswell as the ‘‘net’’ findings reported by the authors.

Ž .Scott 1984 performs a cross-sectional analysison FTC line-of-business data for 437 firms in 259four-digit industries. He finds that private R&D ispositively associated with government financed R&Dusing both a restricted intensity version of the rela-tionship, and a log-level version. In both cases, hisestimates are robust to the inclusion of firm dummyvariables and four-digit industry dummy variables.The results also hold up whether or not firms thathave zero company financed R&D are includedfrom the sample. As is the case for many of thestudies surveyed here, his analysis may be biasedbecause of endogeneity between public and privateR&D, due to omitted variables that drive both setsof funding decisions. At least part of the ‘‘below-firm’’ variation across lines-of-business probably isattributable to variation in technological opportuni-ties and appropriability conditions that are affectingthe marginal returns. Scott’s use of industry dum-mies should capture some of these omitted variables.

Ž .But, the recent works of Trajtenberg 1989 andŽ .Toole 1999a; b suggest that variation in technologi-

cal opportunities is very important even at the prod-uct class level.

Ž . Ž .Leyden and Link 1991 and Leyden et al. 1989develop a more elaborate structural model in order toprovide insight into the empirical relationship be-tween public and private R&D. Since the approachof these two studies is very similar, we will focus on

Ž .the empirical results of Leyden and Link 1991 . Theauthors estimate a three-equation system using

Ž .three-stage least squares 3SLS , using a cross-sec-tion of firm R&D laboratory data for 1987. Theirendogenous variables are laboratory total private R&D budget, laboratory knowledge sharing effort, andgovernment total spending to acquire technicalknowledge. For each lab, total government spendingis defined broadly to include government contracts,grants, and the value of government financed equip-ment and facilities.

Leyden and Link’s reduced form equations in-Ž .clude the following predetermined variables: 1 a

dummy variable indicating that the lab has coopera-Ž .tive sharing arrangements; 2 a dummy variable

Ž .indicating that the lab does basic research; 3 adummy variable indicating that the lab is oriented

Ž .toward biological or chemical research; 4 the two-Ž .digit industry level R&Drsales ratio; and 5 an

interaction term between the R&Drsales ratio andthe presence of cooperative sharing arrangements.The excluded exogenous variables in the privateR & D equation are essentially the dummy ofbiorchem research and the dummy for basic re-

Žsearch the dummy for cooperative sharing arrange-ments is present in the equation, multiplying theindustry R&D to sales, so it is not adding any

.identification .The regression model allows government R&D to

influence the private R&D budget directly, as wellas indirectly through the lab’s knowledge-sharingactivities. Further, the authors account for potentialsimultaneity of private and public funding decisionsby using the predicted values from a first stagereduced form regression as instrumental variables.They find a positive and significant coefficient onthe government R&D in both their private R&Dand knowledge sharing equations. Accounting forvarious feedback effects, this implies an elasticity of

()

P.A

.DaÕid

etal.r

Research

Policy

292000

497–

529513

Table 1Line of business and laboratory studiesSee text for details.OLSsOrdinary least squares; 3SLSs three-stage least squares; FEsFixed effects.Where the regression is in levels, the elasticity is derived using the mean levels of R&D spending for the sample.

Author Time Data type Number of Explained variable Explanatory variable Controls Method ‘‘Net’’ findingsŽ . Ž . Ž .period observations private R&D public R&D elasticity

Ž . Ž . Ž . ŽScott 1984 1974 LB cross-section 3338 log private R&D Log Government R&D Size, firm or OLS firm, Complementarity. Ž .industry dummies industry effects 0.06–0.08

Leyden et al. 1987 Lab. cross-section 120 US$ Private US$ Government R&D Size, lab K-sharing, 3SLS InsignificantŽ . Ž . Ž .1989 lab budget funding to lab D R&D industry 0.145Leyden and 1987 Lab. cross-section 137 US$ Private US$ Government R&D RrS, lab K-sharing, 3SLS Complementarity

Ž . Ž . Ž .Link 1991 lab budget and equipment D chemrbio , 0.336Ž .D basic R

Ž .Klette and Moen 1982–1995 Panel within industry 192=3.6 US$ Private US$ Government Sales, sales sq., FE OLS Neither 1 for 1Ž . Ž . Ž .1998 Norway Mach., elec., inst. R&D Log R&D subsidy log cash flow, complementarity

Ž . Ž . Ž .private R&D Government R&D time dummies 0.06


0.34 between private and public R&D. Since, in thisstudy, identification of the effect of governmentR&D on private spending comes primarily from thecomparison of labs of different types, it is not sur-prising that it turns out to be fairly weak.

A different approach is taken in a new study byŽ .Klette and Moen 1998 , based on panel data on

lines of business in Norwegian high-technologyfirms. Specifically, they looked at the effects ofgovernment matching grants in the electronics andelectrical equipment sector on the R&D undertakenin the lines of business between 1982 and 1995. Thisstudy is one of the few to control for the potentialendogeneity of receiving a grant that arises fromfirm self-selection into the program. 27 Klette andMoen find that government R&D is complementaryto private R&D, in the sense that the lines ofbusiness do not reduce their R&D effort, but that noadditional R&D is induced beyond the level thatwould have been performed anyway. That is, the50% that the firm has to put up to receive the grantcomes out of their normal R&D budget. Because theauthors have a fairly long panel, they are able to lookmore closely at the dynamic effects of this R&Dsubsidy program and this yields their most interest-ing finding: the government subsidy does appear toinduce a permanent increase in the level of theline-of-business R&D.

3.2. Firm-leÕel studies

A summary of the firm level studies can be foundŽ .in Table 2. Hamberg 1966 was the first researcher

to use a regression approach on firm level cross-sec-tional data to address the relationship between publicR&D and private R&D. Within the context of abroader exploration of firm R&D activity, Hambergincluded a government R&D variable defined to bethe total value of Department of Defense contractsdivided by gross fixed assets. Separating his sampleof 405 firms into industry groups, he regressed thefirm’s proportion of private R&D employment on itsgovernment contracts and other variables for each

27 Ž . Ž . Ž .See Lichtenberg 1988 , Busom 1999 and Wallsten 1999 ,which are discussed in Sec. 4.2, for other studies that make thiscorrection.

industry using ordinary least squares. He reportedthat the government contracts are positively related

Žto private R&D for six out of eight industries four.are significant and negatively but insignificant for

the other two. The four industries showing comple-Ž . Ž .mentarity are: 1 industrial chemicals; 2 electronic

Ž .components and communications equipment; 3Ž .other electrical equipment; and 4 office machines.

A surprising result was that the coefficient on De-fense Department contracts was positive but notsignificant for the aircraft and missiles industry,because this industry had nearly 90% of its R&Dfunded by defense contracts during the sample pe-riod. Exploration of the timing of the response tothese contracts failed to overturn this result. Note,

Ž .however, that David and Hall 1999 show that aweak response of private R&D spending to publicR&D is more likely when the government share of

Žtotal R&D spending is large in a specialized tech-.nological or labor market area , or when government

R&D does not enhance private productivity verymuch. Both conditions may well apply in the case of

Ž .this industry particularly the former .Ž .The analysis of Hamberg 1966 represented a

step forward in two important ways. First, he fo-cused on DOD contracts awarded to individual in-dustry groups. The institutional homogeneity of thefunding agency and the industry group help limitpotential omitted variables bias from other sources ofvariation like technological opportunities. Second, heincluded an impressive combination of control vari-ables such as profits, sales, depreciation, gross in-vestment, and a lag of R&D personnel. As we sawin Section 3, these variables help hold constant otherfactors that may be shifting the firm’s marginalreturns and cost of funds schedules. However, thecost of the homogeneity of his sample is a smallsample size for each regression. His largest industrygroup had only 34 firms.

Ž .Howe and McFetridge 1976 conduct a carefulanalysis of the impact of Canadian public R&Dincentive grants on Canadian private R&D invest-ment. Canadian incentive grants are a cost-sharingarrangement in which public funds are used to sup-port half of the total cost of a private R&D projectthat the government has decided to support. Motivat-ing their specification using the same investmentframework described in Section 3, the authors esti-


mate a reduced form model of private R&D inten-sity on government incentive grants and other vari-ables. They use a sample of 81 firms over the period1967–1971 separated into three industry groups. As

Ž .in Hamberg 1966 , this helps protect against biasedestimates stemming from inter-industry differencesin product characteristics, technology, appropriabil-ity, and other factors. They find a positive coefficienton public incentive grants for all three industries;however, it is significant only for the electrical in-dustry.

Ž . Ž .Higgins and Link 1981 and Link 1982 analyzehow the composition of private R&D investmentresponds to public R&D support. Using a cross-sec-tion of 174 manufacturing firms in 1977, Higgins

Ž .and Link 1981 find that the percentage of privateR&D investment devoted to research falls as thefirm receives more public R&D funds. Althoughthey interpret this as evidence for crowding-out, itshould be remembered that this is only one portionof a firm’s R&D budget. In fact, in an expanded

Ž .version of this study, Link 1982 finds that federalR&D stimulates total private R&D intensity. Whenprivate R&D investment is decomposed into basic,applied, and development components, Link findsthat public R&D reduces private basic research in-tensity. Public funds, however, stimulate private de-velopment intensity and have no significant effect onprivate applied research intensity. While both Hig-

Ž . Ž .gins and Link 1981 and Link 1982 include con-trol variables accounting for firm profits, diversifica-tion, and high-technology orientation, their resultsare probably affected by unobserved inter-industryvariation. Nonetheless, these studies attest to theimportant point that the type of public R&D supportmatters for the private R&D investment decision.The distinction made between public R&D contractsand grants in Section 3 is based on the same observa-tion.

Ž .Lichtenberg 1984; 1987; 1988 has stressed theeconometric issues involved in the analysis of therelationship between public contract R&D and pri-vate R&D investment. He has employed a demand

Žand supply framework similar in spirit to the.micro-level investment framework presented here in

order to make two basic points: first, that publicR&D contracts should be treated as endogenous atthe firm level; second, that sales to government are

more R&D intensive and have an effect on privatemarginal returns that should be included in the speci-fication. In his 1984 paper, Lichtenberg also suggeststhat the failure of previous studies to account fortime constant and unobserved firm characteristicshas led to an upward bias on the coefficient esti-mates. Using data on a cross-section of firms in threedifferent time periods, he presents regression resultsfor private R&D intensity in both level and first-dif-

Žference specifications i.e., estimates that control for.permanent differences across firms . The mostly pos-

itive and significant coefficients on public R&Dbecome negative and significant in the latter regres-sions, which implies that the observed complemen-tarity was due to firm-level differences in R&Dintensities that are correlated with the award ofpublic R&D contracts. Increases or decreases inpublic R&D funding within the firm were not asso-ciated with increases or decreases of private funding.As in all such differenced regressions, however, thepotential for downward bias due to measurementerror in the independent variable must be kept inmind.

ŽThe subsequent papers of Lichtenberg 1987;.1988 address the issues posed by the tendency of

federal procurement demand to concentrate on moreR&D-intensive products, and the potential endo-geneity of federal contracts. His 1987 paper com-bines firm data from Compustat and the FederalProcurement Data System to construct a panel of 187firms over the period 1979–1984. He demonstratesthat the value of government contracts variable be-comes insignificant once government purchases areincluded separately in the regression equation.

Using a sample of 169 firms from the sameŽ .database, Lichtenberg 1988 proceeds to further par-

tition federal contracts into ‘‘competitive’’ and‘‘non-competitive.’’ While his results show a posi-tive and significant government R&D coefficient forthe total and ‘‘within’’ OLS regressions, the coeffi-cient becomes negative and significant when he ac-counts for endogeneity using as an instrumental vari-able potential government contracts for products thatthe firm produces. It is important to note that thegovernment sales variable in the regression includesthe value of government contracts. As he makesclear, his results imply that there is no additionaleffect of government contracts beyond their impact

()

P.A

.DaÕid

etal.r

Research

Policy

292000

497–

529516

Table 2Ž . Ž .Firm-level studies — a US data; b Data from other countries

Ž .Definitions: Esemployment. CSsCross-section. FEsFixed effect within or firm dummies . FDsFirst differences. IVs Instrumental variables.


( )aŽ . Ž .Hamberg 1966 1960 Firm CS within 8= ;20 Private R&D Er US$ Government Size, depreciation, Weighted Mixedr

industry Total E contractsrAssets investment, OLS complementaritylag R&D E

Ž .Shrieves 1978 1965 Firm CS across 411 Log % Government- Size, prod mkt, OLS SubstitutabilityŽ .industry private R&D E finaced R&D tech opportunity, C4

Ž .Carmichael 1981 1976–1977 Firm CS within 46=2 US$ Private R&D US$ Government Size Pooled Substitutabilityindustry expenditure R&D contracts OLSŽ .transprotation

Higgins and 1977 Firm CS across 174 % Research in US$ Government- ProfitrS, divers., OLS SubstitutabilityŽ . Ž . Ž .Link 1981 industry private R&D financed R&D D hitech y0.13Ž .Link 1982 1977 Firm CS across 275 Private R&Drsales Government- ProfitrS, divers., OLS Complementarity

Ž .industry financed C4, D governanceR&Drsales

Ž .Lichtenberg 1984 1967, 1972, Firm CS across 991 Change in private Change in Size Fixed Substitutability1977 industry R&Drsales Government effects

R&DrsalesŽ .Lichtenberg 1987 1979–1984 Panel across 187=6 US$ Private US$ Government- Year dummies, size, Pooled Insignificant

industry R&D expenditure financed R&D sales to government OLSŽ . Ž .Lichtenberg 1988 1979–1984 Panel across 167=6 US$ Private US$ Government- Year dummies, size, FE OLS, Substitutability IV

Ž .industry R&D expenditure financed R&D sales to government IV complementarity FEŽ .Wallsten 1999 1990–1992 Firm CS across 81 US$ Private R&D Number of SBIR Age, size, patents, OLS, Substitutability

Ž .industry experienced awards, total value R&D exp. 1990 , 3SLSŽ .in 1992 of SBIR awards D never apply ,

industry andgeography dummies

()

P.A

.DaÕid

etal.r

Research

Policy

292000

497–

529517

( )bŽ .Howe and 1967–1971 Firm panel 6=44 US$ Private US$ Government Size poly. , profit, Weighted Mixedr

McFetridge within industry R&D Expenditure R&D grants deprec, HHI, OLS complementarityŽ . Ž . Ž .1976 Canada D foreign

Ž . Ž . ŽHolemans and 1980–1984 Firm CS across 5= ;47 Log private R&D log Government Size, divers., HHI, FE OLS Complementarity. Ž . Ž .Sleuwaegen and within R&D grants D for. , log 0.25–0.48

Ž . Ž . Ž .1988 Belgium industry royaltiesŽ .Antonelli 1989 1983 Firm CS within 86 Private R&D % Government- Size, profit, OLS Complementarity

Ž . Ž . Ž . Ž .Italy industry MM lire , log financed R&D, log D div , share for. 031–0.37Ž . Žprivate R&D government R&Dr Sales, foreign RrS

.total R&DŽ . Ž Ž .Busom 1999 1988 Firm CS across 147 Private R&D D participation in Size, patents, OLS with Complementarity 0.2

Ž . . Ž .Spain industry expenditure, R&D subsidy loan program export share selection heterogeneousper employee industry Dummies correction

Toivanen and 1989, 1991, Panel across 133=3 US$ Private US$ Government- Investment, cash FD IV SubstitutabilityNiininen 1993 industry R&D expenditure financed R&D flow, interest, rate, –subsidies toŽ . Ž . Ž . Ž .1998 Finland loans and subsidies current and one lag large firms y0.10

of all variables loans and small firmsinsignificant


on the firm’s marginal returns. As opposed to substi-tution, it would appear that what Lichtenberg hasidentified is that the primary impact of governmentcontracts on private R&D investment works thoughtheir effect on the firm’s private marginal returns,rather than on the firm’s marginal cost of fundsschedule.

Ž . Ž .Toivanen and Niininen 1998 , Busom 1999 andŽ .Wallsten 1999 are the most recent firm-level stud-

ies on the publicrprivate R&D relationship. Wall-sten focuses on the effects of the US Small Business

Ž .Innovation Research Program SBIR . This is a com-petitive grant program designed to target R&D sub-sidies to smaller businesses, and Wallsten was ableto collect data on firms that have won SBIR awards,firms that applied but were rejected, and firms thatwere eligible to apply but did not apply. His sampleof awardees comes from the data systems of elevendifferent governmental agencies and covers a varietyof industrial areas over the 1990–1992 time period.In order to correct for the potential endogeneity ofthe public funding decision on the firm’s R&Dresponse, Wallsten uses a three-equation system in-tended to model the award granting process as wellas the firm’s response. Estimating the system by3SLS and using the total SBIR budget from whichfirms could have won awards as his instrument, hefinds that the number of SBIR awards won by a firmhas a positive but insignificant effect on firm em-ployment. In a separate set of regressions, he findsthat the number and value of SBIR awards signifi-cantly reduces firm R&D expenditure. In these latterregressions, however, Wallsten is forced to analyzeonly publicly owned firms and this reduces his sam-ple of firms from 481 observations to 81 observa-tions. Based on the typical size of an SBIR award,Wallsten concludes that public grants displace pri-vate R&D investment on a nearly one-for-one basis,although it must be noted that this finding pertainsonly to the publicly owned recipients of such fund-ing.

Ž .Busom 1999 analyzes the effect of Spanish gov-ernment subsidized loans on private R&D expendi-ture and employment using a sample of Spanish

Ž .firms. Like Wallsten 1999 , Busom is careful toaddress the potential endogeneity of the public fund-ing decision stemming from selection bias in thegrant process. She explores this problem by imple-

menting a two-step procedure that predicts the prob-ability of participation in the program in the firststep, and includes a correction for selection in thesecond step. Using a sample of 147 Spanish firms in1988, Busom finds that the hypothesis of no selec-tion bias cannot be rejected, even thought, the legiti-macy of pooling participant and non-participant firmsis rejected by a Chow test. Unfortunately, her datadoes not include the amount of the R&D subsidy,only the fact that it exists; so she is unable to providea quantitative estimate of the complementarity orcrowding-out effect. She finds, that on average, re-ceiving a government R&D subsidy induces moreprivate R&D effort than would be predicted on thebasis of an R&D effort equation for the ‘‘controls-

Ž .firms’’ those that did not receive a subsidy , whencorrections are made for sample selection biases. For30% of the firms, however, full crowding-out re-mained a possibility that could not be ruled out bythe data. It will be interesting to see results based ondata from this Spanish program in the future, espe-cially those that reveal the amount as well as thepresence of the R&D subsidies.

In addition to the typical econometric formula-tions that regress some measure of private R&D onpublic R&D, an increasing number of alternativeand more indirect approaches have been pursued inrecent years. Since these papers are less direct thanthose considered above, the following brief resumesreport only the ‘‘bottomline’’ from these efforts. Theinterested reader is urged to consult the originalstudies for more detail. First, Irwin and KlenowŽ .1996 analyze a sample of US semiconductor firmsto find out how private R&D investment respondswhen firms participate in a government supportedR&D consortium. By finding that membership re-duces private R&D investment, the authors suggestthat membership eliminates the need for duplicative

Ž .R&D. Lerner 1999 looks at the long-run impact ofthe SBIR program on firm sales and employment. Hefinds that these government grants have only a lim-ited positive impact, except in those areas wherethere also is substantial venture capital activity. Thespecification, however, does not allow one to distin-guish whether the presence of venture capital fund-ing exerts its effect through shifting the MCC sched-ule, or whether it is really a surrogate for technologi-cal opportunity set differences, i.e., proxying the


attraction of biotechnology and software innovationsthat can be pursued by small firms.

Ž .Cockburn and Henderson 1998 analyze the co-authorship of scientific papers between public andprivate institutions in the US. They find that privatefirm organization and research productivity are posi-tively related to the fraction of co-authorship withacademic institutions, which are largely supported byfederal funds. In a closely related study, Narin et al.Ž .1997 examine the contribution of public science toindustrial technology using patent citation measures.Their research finds a strong link between industrialpatents and publicly supported research. Feldman

Ž .and Lichtenberg 1998 report finding that for asample of European Union member countries there isa strong positive relationship between the number ofprivate institutions that specialize in a particularscientific field and the number of public institutionsspecializing in the same scientific field. They inter-pret this result as evidence of complementarity be-tween public and private investments in R&D.

3.3. Industry-leÕel studies

There have only been a handful of industry levelstudies on the relationship between public R&D andprivate R&D investment, probably because at thislevel of aggregation the absence of a clear ‘‘experi-ment’’ is most glaring. The most important proxi-mate source of variation in R&D intensity acrossfirms is differences in industry, and this is as true ofpublic R&D as of private R&D expenditures.Therefore we should not be surprised if industry-levelstudies show complementarity between the two, norshould we conclude anything other than the fact thatsome industries have greater technological opportu-nity than others.

The industry studies are summarized in Table 3.Ž . Ž .The Globerman 1973 and Buxton 1975 papers

examine industry cross-sectional variation and findcomplementarity. However, these studies use very

Ž .small samples, 15 data points in Globerman 1973Ž .and 11 data points of Buxton 1975 . Larger samples

and better specifications can be found in GoldbergŽ . Ž .1979 , Lichtenberg 1984 , and Levin and ReissŽ .1984 . Each of these studies use NSF data arrangedas a panel with observations on a cross-section of

Ž .industries over time. Goldberg 1979 , following aneoclassical investment approach, regresses privateR&D per unit of output on both current and laggedfederal R&D per unit of output, plus industry dum-mies and other control variables. He finds that cur-rent federal R&D has a negative and significantcoefficient while federal R&D lagged one periodhas a positive and significant coefficient. The sum ofthese coefficients indicates a small complementarityeffect of federal R&D on private R&D investment.

Ž .Lichtenberg 1984 , on the other hand, finds thatpublic R&D reduces private R&D investment andemployment at the industry level. His preferred spec-ification regresses the change in private R&D in-

Ž .vestment or employment on the change in contem-poraneous and lagged federal R&D plus industrydummies, and time dummies. For federal R&D, heconcludes that an additional dollar crowds out eightcents of private R&D investment. Introducing bettercontrols for cross-industry variation in technologicalopportunity and appropriability, however, would tendto remove the upward bias in the estimated coeffi-cient and thus reinforce the indications of crowding-out in this study.

Ž .The paper of Levin and Reiss 1984 stands out asthe most ambitious industry level study in the litera-ture. They develop a structural equation system thatrelates an industry’s concentration, R&D and adver-tising intensities to the industry’s structure of de-mand, technological opportunities, and appropriabil-ity conditions. In their model, government R&Dintensity enters both as a measure of technologicalopportunity and as a determinant of appropriabilityconditions. Rather than treat government R&D asexogenous, however, the authors specify a reducedform equation that models government R&D inten-sity as a function of the government share of indus-

Ž .try sales both defense and non-defense and of theextent of R&D ‘‘borrowing’’ via capital purchasesfrom other industries, along with the opportunity andappropriability variables. Using two stage leastsquares, Levin and Reiss find that government R&Dintensity has a positive and significant effect onprivate R&D intensity. In all specifications, the au-thors find a complementary relationship with a mag-nitude that implies each additional dollar of publicfunds stimulates from seven to seventy-four cents ofprivate R&D investment.

()

P.A

.DaÕid

etal.r

Research

Policy

292000

497–

529520

Table 3Industry-level studiesAll studies use US data unless otherwise noted.See Table 2 for definitions.


Ž . Ž .Globerman 1973 1965–1969 Cross-section 15 R&D Er Government R&Dr D tech opportunity , OLS ComplementarityŽ .Canada Total E sales % foreign, industry

conc., sales growthŽ .Buxton 1975 1965 Cross-section 11 Private R&Dr Government R&Dr C4, Divers., OLS Complementarity

Ž .United Kingdom Gross output Gross output entry barriers?,Ž . ŽGoldberg 1979 1958–1975 Panel 18=14 Log private R&Dr Government R&Drsales Industry Dummies, FE OLS Complementarity

. Ž .output sum of lag 0 and 1 price of Rlag private Rroutput

Ž .Lichtenberg 1984 1963–1979 Panel 12=17 Change in Change in Year dummies, FE OLS Insignificantprivate R&D Government R&D industry dummies,

Levin and Reiss 1963, 1967, Panel 20=3 Private R&Dr Government R&Dr Tech dummies, 2SLS ComplementarityŽ .1984 1972 production costs shipments basic R share,

industry age, HHI


A handful of carefully executed case studies maybe noticed here, all of which report finding a com-plementary relationship between public R&D andprivate R&D investment. The best known study,

Ž .Mansfield and Switzer 1984 , uses interview datacollected from R&D executives in private firms.Their study focuses on R&D targeted on energytechnologies in the electrical equipment, petroleum,primary metals, and chemical industries. The authorscombine the responses from 25 firms to calculate thechange in company-financed R&D per dollar ofgovernment-financed R&D. They uncover an asym-metric relationship in which the effect of an increasein government R&D has a different magnitude andtime profile than a decrease in government R&D.An additional dollar of federal R&D stimulates anadditional six cents in the first 2 years and nothingthereafter. A one-dollar decrease, on the other hand,stimulates a twenty-five cent fall in private R&D inthe first 2 years and an additional nineteen-centdecrease in the third year. Support for a complemen-tary public–private R&D investment relationshipalso has emerged from studies of aircraft and civilian

Ž .space technology, by Hertzfeld 1985 and MoweryŽ .1985 .

3.4. Aggregate studies

We have identified seven aggregate macro econo-metric studies of the publicrprivate R&D relation-ship in the recent journal literature. These papers aresummarized in Table 4. Not only was the Levy and

Ž .Terleckyj 1983 paper the first of the macro levelstudies but it remains the most definitive of its kind.Using NSF data for the United States over the periodof 1949–1981, the authors explore the impact ofgovernment contract and ‘‘other’’ R&D on privateR&D investment and productivity. Their main find-

Ž .ings are: 1 government contract R&D is positivelyand significantly associated with private R&D in-

Ž .vestment and productivity; and 2 ‘‘other’’ govern-ment R&D has no contemporaneous relationship,but does complement private R&D with a lag of 3years, while reducing private productivity with a lagof 9 years. Even after correcting for governmentreimbursement of certain private R&D overheadexpenditures, Levy and Terleckyj find that an addi-tional dollar of public contract research added to the

stock of government R&D has the effect of inducingan additional twenty-seven cents of private R&D

Ž .investment. Moreover, Terleckyj 1985 shows thatthis effect remains after accounting for the R&Dintensity of governmental demand, to which attention

Ž .was drawn by Lichtenberg 1984 . Yet, LichtenbergŽ .1987 , also using NSF time-series data for the US,reports finding that when allowance is made for thehigher R&D intensity of sales to the federal govern-ment, there is no additional impact from publicR&D expenditures on private R&D investment. Thatstudy by Lichtenberg, however, does not replicatethe inclusion of controls for other determinants ofthe aggregate level of private investment and produc-

Ž .tivity by Levy and Terleckyj 1983 , nor does itfollow them in defining public and private R&D asstock variables. Although Levy and Terleckyj’s ap-proach in the latter regard is a conceptual improve-ment, because one would expect the effects of R&Dinvestment to persist for longer than 1 year, there isan practical econometric difficulty with its imple-mentation: working with stocks, rather than the an-nual flows, induces very strong positive serial corre-lation in the dependent and independent variables ofthe regression model.

Ž . Ž .Robson 1993 and Diamond 1998 also conductaggregate time-series analyses using NSF data forthe United States, but they restrict their focus toexamining the effects of basic research. This type offederal funding is of the ‘‘infrastructure’’ varietywhich can be expected to shift out the marginalproduct curve for private R&D; hence, as the analy-

Ž .sis in David and Hall 1999 shows, in the long run itis more likely to increase private and total invest-ment in R&D rather than reduce them. RobsonŽ .1993 focuses his attention on how private basicresearch investment responds to various forms offederally funded R&D, separating the aggregate intobasic research and government outlays forappliedrdevelopment R&D performed in industry.Using data for the period of 1956–1988, he regressesthe change in private basic research investment onthe change in federal basic research expenditures, thelevel of federal appliedrdevelopment R&D, thechange in private appliedrdevelopment R&D in-vestment, and the governmentrnon-government ratioof industry sales. Robson finds that both the changein federal basic and the level of federal appliedrde-

()

P.A

.DaÕid

etal.r

Research

Policy

292000

497–

529522

Table 4Aggregate studiesAll studies use US data unless otherwise noted.See Table 2 for definitions.

Author Time period Data type Number of Explained variable Explanatory variable Controls Method ‘‘Net’’Ž . Ž . Ž .observations private R&D public R&D findings elasticity

Levy and Terleckyj 1949–1981 Time-series 33 US$ Private R&D US$ Government Lag output, lag taxes, GLS ComplementarityŽ .1983 stock contracts to industry unemployment age R&D stock,

Ž .stock US$ Government R&D,US$ reimbursement

Ž .Terleckyj 1985 1964–1984 Time-series 21 US$ Private R&D US$ Government Output, government durables, GLS Complementarityexpenditure contracts to industry lag R&D in EuroperJapan

Ž .Lichtenberg 1987 1956–1983 Time-series 28 US$ Private R&D US$ Government Sales, sales OLS InsignificantŽ .expenditure contracts to industry to government 0.045

Ž .Levy 1990 1963–1984 Panel 9=21 US$ Private R&D US$ Government GDP, country dummies, Pooled ComplementarityŽ .cross-country expenditure contracts to industry pred. Europe and Japan GLS

private R&DŽ .Robson 1993 1955–1988 Time-series 33 Change in private Change in federal Level and chg private appl. R, OLS — Complementarity

basic research basic research Government appl. R, 1st-diffGovernment purchases, chgin non-government goodsin services

Ž .Diamond 1998 1953–1993 Time-series 41 US$ Private basic US$ Federal basic GDP, time trend OLS — ComplementarityŽ .research research 1st diff 1.04

Box–CoxVon Tunzelmann 1969–1995 Panel 22=27 Change in private Change in public Levels of private Fixed Complementarity

Ž .and Martin 1998 R&D R&D and public-funded R&D, effectsŽ .cross-country country dummies


velopment research have positive and significant co-Ž .efficients in this regression equation. Diamond 1998

similarly uses NSF data for the earlier portion of theŽ .same period 1953–1969 to examine the impact of

changes in federal basic research expenditures onchanges in basic research spending by industry. He,too, finds a positive and significant coefficient on thefederal spending variable.

Thus, both of the foregoing studies conclude thatat the aggregate level the overall, the net relationshipbetween public and private investments for basicresearch is one of complements, not substitutes. In-deed, the estimated the elasticity reported by Dia-

Ž .mond 1998 is very high, roughly unitary. It shouldbe noted, however, that as these results are obtainedfrom the co-variation in the aggregate time-seriesobservations, they could be reflecting the correlatedeffects of other macroeconomic variables. Neitherstudy makes use of instrumental variables to controlfor the influence of the US business cycle and shiftsin overall federal fiscal policy during the period inquestion.

As an alternative to attempting to control for thepossible influences of those omitted variables, adifferent approach is available where one can makeuse of cross-country panel data to deal with theendogeneity problem that is likely to plague time-series analyses of a single economy. Indeed, therehave been some recent attempts to exploit this possi-ble route to identifying the nature of the public–private R&D relationship, by employing aggregate-level time-series observations for OECD countries inpanel form. Starting with a sample of nine OECD

Ž .countries for the period of 1963–1984, Levy 1990works with a specification that distinguishes amongthree geographic regions within which it is assumedthat there would be strong spillovers of the effects ofpublic R&D expenditures: the US, Europe, andJapan. He therefore regresses national private R&Dinvestment on aggregate public R&D investment ineach region, aggregate regional GDP, and individualcountry dummy variables. Among the nine countries

Ž .in his panel, Levy 1990 finds that five countriesexhibit significant overall public–private comple-mentarity effects, whereas two countries show signif-icant substitution effects. The reasons for the differ-ences remain unexplored, which is understandable inview of the restricted size of the cross-national sam-

ple. Further progress along these lines may be possi-ble in the near future. Research by Von Tunzelmann

Ž .and Martin 1998 has ambitiously undertaken todevelop the R&D time-series for some 22 OECDcountries over the 1969–1995 period. Exploratoryanalysis of this data by von Tunzelmann and Martinis still in progress, but they report preliminary resultsof using the panel data to fit a linear model relatingchanges in industry-financed R&D to the changes inthe government-financed R&D, and the previouslevels of both private and public R&D expenditures,allowing country-specific differences in all the coef-ficients. In only 7 of the 22 countries do they findthat changes in government-funded R&D have anysignificant impact on changes in industry-funded R&D, with the sign being positive in five of those sevencases. 28 Rather more illuminating results are likelyto be obtained by exploiting the availability of thisenlarged panel to estimate models whose specifica-tions take account of cross-country differences in theset of structural characteristics that David and HallŽ .1999 suggests would affect the sign of the aggre-gate reduced-form relationship between public andprivate R&D investments.

Because it addresses an essentially macro-leveleffect, we also should take notice here of the investi-

Ž .gation by Goolsbee 1998 of the direct labor marketimpacts of US government R&D funding. Using

Ž .NSF data on scientists and engineers S&E , Gools-bee finds that increases in expenditures for publicR&D have a significant effect in raising averageS&E wages. He suggests that a major fraction ofpublic R&D yields windfall gains to S&E workers;and that, by raising the cost of technically skilledworkers used in private R&D laboratories, govern-ment funding tends to ‘‘crowd out’’ private R&D.

ŽAs we have noted in our earlier discussion see.Sections 2.5 and 2.6 of the R&D input market

effects of an exogenous increase — in either public

28 The countries exhibiting complementarity in this respect are:Germany, the Netherlands, Switzerland, New Zealand and theUSA. But the preliminary results overall are decidedly mixed:

Ž .leaving aside the three countries among the 22 for which thecoefficient on changes in government-financed R&D is bothnon-significant and close to zero, in 10 of the remaining 18 casesthe coefficient is found to be positive.


or private investment levels — the implication offindings such as Goolsbee’s is that aggregate-leveleconometric estimates are likely to overestimate thereal response of private R&D investment to publicR&D expenditures, because they will include thepositive price effects. This may help to explain whythe macro-level estimates implied by the findings inGoolsbee’s study, and others, tend to be morestrongly in favor of complementarity than the gen-eral run of the micro-level elasticity estimates sur-veyed here. 29

3.5. Micro-leÕel impacts of public R&D performedby non-profit organizations

Ž . Ž .Adams 1998 and Toole 1999a; b appear to bethe only econometric studies on the relationship be-tween publicly fundedrnon-profit performed re-search and private R&D investment in the manufac-turing sector. 30 These two studies, however, havevery different objectives and use different data and

Ž .specifications. Adams 1990 is interested in thefactors that mediate research spillovers between in-dustrial labs and outside firms and academic institu-tions. By separating funds devoted to learning activi-ties from the overall R&D budget, Adams is able toinvestigate the dependence of the level of privateR&D upon the stocks of industry R&D and feder-ally funded academic R&D. Using survey data from208 industrial laboratories in the chemical, machin-ery, transportation equipment, and electrical equip-ment industries, he finds that publicly supportedacademic research does not stimulate industriallearning R&D, although it does stimulate greaterexpenditures on learning about academic R&D. Eventhough this line inquiry is on-going, econometricresearch that separates private R&D investment intodifferent categories, in the same spirit as that of LinkŽ .1982 , can be expected to improve our understand-

29 Ž .See David and Hall 1999 for further discussion of theŽ .quantitative implications of the estimates of Goolsbee 1998 .

30 Ž . Ž .The papers by Jaffe 1989 and Ward and Dranove 1995 areŽ .part of a broader definition of this literature. Jaffe 1989 looks at

geographically mediated spillovers and does not differentiate be-tween publicly and privately funded academic research. Ward and

Ž .Dranove 1995 , on the other hand, separate public and privateR&D funding but do not do not differentiate between performersof public R&D.

ing of how overall private R&D budgets respondwhen there are changes in the distribution of publicfunding for R&D performed by different types ofresearch organizations.

Ž .Toole 1999a; b explores the complementaritybetween private R&D investment in the US pharma-ceutical industry, and publicly funded research per-formed outside the industry, in public and privatenon-profit institutes and universities. Since this typeof public research affects the private marginal re-turns to R&D and not the private cost of funds, it isimportant to hold constant other factors that mayshift the private MRR schedule. The analysis at-tempts to control for these other factors in threeways. First, restricting the analysis to a single indus-try serves to eliminates a major source of demandshifts arising from the variation in technologicalopportunities and appropriability conditions. Second,the use of a new and detailed database on publicR&D grants and contracts, allows Toole to separateboth public and private R&D investment expendi-tures into seven medical technology classes. This isimportant since the private marginal returns schedulewill differ by the specific technologies and thechanges in the state of the art in those technologicalareas. Third, public and private R&D investmentdata are matched by technology class over a 15-yearperiod to construct a panel data set. This permits useof an econometric specification that includes bothtechnology class dummies and year dummies. Differ-ences in the marginal returns across classes arepicked up by the class dummy variables while R&Dprice changes and other trends over time are cap-tured by the year dummies.

With additional control variables for regulatorystringency and a correction for the endogeneity of

Ž .expected profitability, Toole 1999a finds that pub-lic basic research stimulates private R&D invest-ment after a lag. For the estimated lag of 6–8 years,the elasticity of private R&D investment with re-spect to the stock of public basic research lies in therange of 0.46–0.53. While Toole’s analysis does notuse an econometric correction for the potential endo-geneity of public R&D funding, that is not likely tobe a serious problem in this particular context. Un-

Ž .like the program-oriented analyses by Busom 1999Ž .and Wallsten 1999 , there is no program selection

bias affecting the for-profit firm’s response to this


publicly supported R&D. Moreover, other sourcesof endogeneity, such as simultaneity bias and omit-ted technological opportunity variables are mini-mized, given the estimated lag in the relationship andtechnology class structure of the model. This muchhaving been said in favor of the use of industry leveltechnology classes, Toole’s approach does assumethat firm effects are less important than technologyclass effects. That well may be the case for the moredynamic high-technology industries, but it is lesslikely to hold for more mature non-science-basedindustries. It would be revealing, therefore, to carryout a similarly detailed analysis at the firm level, but,unfortunately, the distribution of proprietary R&Dinvestments among specific technological areas isnot information that businesses are likely to discloseuntil it ceases to be regarded as having any futurecommercial relevance.

4. Summary and conclusions

The discussion introducing this survey elaborateda standard unifying framework within which to ex-amine R&D investment at the microeconomic level.With its help we sought to identify the distinctivechannels through which the provision of governmentsubsidies would affect the behavior of business firms,associating those effects with shifts either of thefirm’s MRR schedule, or of its MCC schedule, or ofboth.

The implications of the firm-level analysis wascompared with the insights obtained from a heuristicstructural model of relationships that would obtain atthe macro-level, drawing for the latter on the analy-

Ž .sis recently presented by David and Hall 1999 .Recognition of the existence of heterogeneities andasymmetries among firms, quite apart of problemsarising from the interdependence of enterprise behav-ior in imperfectly competitive markets, was seen torender invalid the attempt to pass from the micro- tothe macro-analytic level directly, by separate aggre-gation of the MRR and MCC schedules and solutionof the industry, sectoral or economy-wide equilib-rium. This occasions the need for a separate macro-level framework, in which the effects of fundingupon the prices of R&D inputs, as well as informa-

tional spillovers, can be represented for both theshort-run and long-run cases.

In the development of this part of the exposition,simplicity, rather than novelty was the criterion towhich the discussion adhered. Its two immediatepurposes were to highlight the principal econometricissues that would need to be addressed by the ensu-ing critical review, and to provide guidance in inter-preting the empirical findings reported by the quitediverse array of studies that compose the literatureon this topic. Insofar as those goals were met, webelieve that a strong initial case has been made forthe value of paying greater attention to structuralmodeling; certainly, for taking structural modelingfurther than has been the norm for research contribu-tions addressing the issue of whether or not publicfunding of R&D encourages, or simply substitutesfor R&D investments made by business enterprises.

This survey deliberately has eschewed an effort toarrive at any definitive empirical conclusions regard-ing the sign and magnitude of the relationship be-tween public and private R&D. In doing so, we haveacknowledged the multiplicity of the approaches tothis question that appear in the literature, and theconsequent lack of immediate comparability betweenstudies conducted at differing levels of aggregation,and treating a variety of modes and purposes ingovernment funding of R&D.

Quite apart from the difficulties of rendering theresults of those studies in a form that would permitready quantitative comparisons, the heterogeneity ofexperience created by the application of institution-ally different subsidy programs to diverse industriesand areas of technology provides strong grounds fordoubting the usefulness of searching for ‘‘the’’ rightanswer. Beyond our commentary on the individualcontributions, what, then, it is possible to extract, byway of a valid and intelligible overview of thepresent state of empirical knowledge on this ques-tion?

Guided by insights from the analytical frame-work, our examination of the literature proceeded bycomparing and contrasting empirical studies that firstwere grouped according to the level of aggregationat which the relationship between publicly and pri-vately funded R&D was examined. In addition aneffort was made to distinguish between findingspertaining to the impact of government contracts,


Ž .and those concerned with other grant provisions offunding for R&D. Further, the discussion of thepublications arrayed in our comparison, Tables 1–4noted those contributions that rest exclusively ondata about US experience. The latter comprise abouttwo thirds of the 33 sets of results assembled forexamination, and keeping in mind this dimension ofheterogeneity in the sources of ‘‘the evidence’’ — inaddition to that arising from variations in ‘‘level ofaggregation’’ dimension — would seem to be perti-nent for a number of reasons. Among the moreobvious and straight-forward of these should be men-tioned the fact that observations drawn from the thirdquarter of the century carry considerably greaterweight in the body of US evidence than is the casefor the studies of experience elsewhere. The compar-

Žatively greater importance of defense and aeronautic.and space contracts for R&D in the total of US

government funding for industrially performed R&D,also is a differentiating characteristic that may besignificant.

This suggests we should reconsider the shape ofthe literature explicitly from those two taxonomicperspectives. And, indeed, several striking featuresof the distribution of overall findings are exposed bythe simple tabulation in Table 5, in which our sampleof studies are arrayed according to the level ofaggregation, and national source of data. As may be

Table 5Summary distribution of econometric studies of the relationshipbetween public and private R&D investment

Studies reporting Total number‘‘net’’ substitution of studies

aLeÕel of aggregation: firm and lowerNumber of studies surveyed 9 19Based on US data only 7 12Based on other countries’ data 2 7

bLeÕel of aggregation: industry and higherNumber of studies surveyed 2 14Based on US data only 2 9Based on other countries’ data 0 5

cAll levels of aggregation 11 33

a Ž .The findings in Toivanen and Niininen 1998 for large firmsand small firms each are counted here as a separate study.

b Ž . Ž .Adams 1998 and Toole 1999a; b are included here, andare assigned following the text discussion in Section 3.5.

c Ž .The assignment of Von Tunzelmann and Martin 1998 fol-lows Table 4, although these results are preliminary.

seen from the table, exactly one-third of the casesreport that public R&D funding behaves as a substi-tute for private R&D investment. This result is farmore prevalent among the studies conducted at theline-of-business and firm level, than among thosecarried out at the industry and higher aggregationlevels — where the relative frequency approaches

Ž .one-half 9r19 .A second pattern that stands out from the table is

that whereas five-sixths of the studies based on datafrom countries other than the US report overallcomplementarity, the corresponding proportionamong those based purely on US data is only four-

w Ž .xsevenths 1– 9r21 . That has some bearing upon athird feature of interest in Table 5: the regionalcontrast in the findings that emerges within thegroup of studies conducted at and below the level ofthe firm. Here one sees a marked difference betweenthe distribution of the US-based findings and themuch higher relative frequency with which comple-mentarity is reported by analysts working exclu-sively from US evidence.

It may well be that this latter contrast is in partreflecting underlying differences between the charac-ter of the US federal R&D contracts and awards,and the purposes and terms of the more recentEuropean government programs of funding for in-dustrial R&D. It should be noted, however, that thefrequency with which ‘‘complementarity’’ appearsamong US-based studies pertaining to experience atthe line-of-business and firm levels cannot be re-garded as being anomalously low; not at least whenit is viewed within the overall context of the distribu-tion of findings summarized by Table 5. 31

Our analysis directed particular attention to thedifferences one should expect to find in the results ofstudies that are conducted at different levels of ag-

31 Quite the contrary, this can be seen by taking the analysis ofthe figure in Table 5 one step further. Consider that the proportionof ‘‘complementarity’’ results among all the US-based studies isŽ .4r7 , and the fraction of all lower aggregation level studies that

Ž .report complementarity is 11r19 . Under independence of theŽ .two effects, therefore, we might expect 4r12 of the US studies

based on line-of-business and firm level data to have reportedŽ .‘‘complementarity’’; whereas the observed frequency is 5r12 .

This is not a significant deviation from the theoretical expectation,and, in addition, it is in the direction opposite that which superfi-cial examination of the table might suggest.


gregation. It was pointed out that the effects ofunobserved inter-industry differences in the techno-logical opportunity set are likely to induce positivecovariation in the public and private components oftotal industry-level R&D expenditures. Further, atthe aggregate level, the likely positive effect onR&D input prices of expanded government fundingalso contributes to the appearance of complementarymovements in the private and public components ofnominal R&D expenditures. One may note, then,that the summary tabulation of results that we have

Ž .reviewed in Table 5 conform well with these theo-retical expectations. Complementarity appears moreprevalent, and substitution effects all but vanishamong the subgroup of studies that have investigatedthis relationship at the industry and national econ-omy levels.

Is this to be read as telling us something about thestrength of the positive impacts that inter-firm andinter-industry spillovers, through knowledge and tan-gible R&D input markets, have upon the expectedprivate rates of return on company-funded R&D? Oris it reflecting some combination of the endogenousresponses of both government and business alloca-tion decisions to opportunities being open by funda-mental scientific and technological advances, and the‘‘R&D price effects’’ of the competition generatedby private and public funders for limited scientificand engineering resources?

At present, these questions remain open, and noless important than they were when Blank and StiglerŽ .1957 launched the search for answers. Progresstowards resolving them will require further, micro-level studies that make a serious effort to control forthe effects of cross-section and temporal variationsin technological opportunities, along with othersources of variation affecting expected private ratesof return. To the extent that government policiesaffecting public R&D funding are correlated withinitiatives intended to enhance appropriability of re-search benefits by investing firms in areas of newtechnological opportunity, identification of the for-mer effects from single country analyses will remaindifficult. Further utilization of international paneldata seems a promising avenue for further work inthis as in other connections.

ŽResearch using use quasi-experimental propen-.sity score or sample selection corrections to com-

paring ‘‘treated’’ firms and ‘‘controls’’ offers ano-ther line of future advance, especially if it couldbe coupled with the design of actual policyexperiments. 32 This suggests the concluding thoughtthat really important gains in the informational basisfor economic policy in this area are only likely tocome as a by-product of coupling better policy de-sign with the application of the sophisticated econo-metric techniques that have now become available.

References

Acs, Z.J., Audretsch, D.B., Feldman, M.P., 1991. Real effects ofacademic research: comment. American Economic Review 82,363–367.

Adams, J.D., 1990. Fundamental stocks of knowledge and produc-tivity growth. Journal of Political Economy 98, 673–702.

Adams, J.D., 1998. Endogenous R&D spillovers, ‘invisible’ R&Dand industrial productivity, presentation at the American Eco-nomics Association Meetings. Chicago, January.

Antonelli, C., 1989. A failure-inducement model of research anddevelopment expenditure. Journal of Economic Behavior andOrganization 12, 159–180.

Arrow, K.J., 1962. Economic welfare and the allocation of re-Ž .sources to invention. In: Nelson, R. Ed. , The Rate and

Direction of Inventive Activity. Princeton Univ. Press, pp.609–625.

Bartelsman, E.J., 1990. Federally Sponsored R&D and Productiv-ity Growth, Federal Reserve Economics Discussion Paper No.121. Federal Reserve Board of Governors, Washington, DC.

Blank, D.M., Stigler, G.J., 1957. The Demand and Supply ofScientific Personnel. National Bureau of Economic Research,New York.

Branscomb, L.M., Parker, G., 1993. Funding civilian and dual-useŽ .industrial technology. In: Branscomb, L.M. Ed. , Empower-

ing Technology: Implementing a National Strategy, Chap. 3.MIT Press, Cambridge, MA.

Busom, I., 1999. An Empirical Evaluation of the Effects of R&DSubsidies. Working Paper No. B99-05, Universitat Autonomade Barcelona.

Buxton, A.J., 1975. The process of technical change in UKmanufacturing. Applied Economics 7, 53–71.

Carmichael, J., 1981. The effects of mission-oriented public R&Dspending on private industry. Journal of Finance 36, 617–627.

Cockburn, I., Henderson, R., 1998. Absorptive capacity, co-authoring behavior and the organization of research in drugdiscovery. The Journal of Industrial Economics 46, 157–182.

Cohen, L.R., Noll, R.G., 1991. The Technology Pork Barrel. TheBrookings Institution Press, Washington, DC.

32 Ž .See Klette et al. 2000 , for further discussion of this point.


Cohen, L.R., Noll, R.G., 1997. Research and development afterŽ .the Cold War. In: Sedaitis, J.B. Ed. , Commercializing High

Technology: East and West. Rowman and Littlefield, Lanham,MD.

Dasgupta, P., David, P.A., 1994. Towards a new economics ofscience. Research Policy 23, 47–521.

David, P.A., 1998a. Common agency contracting and the emer-gence of ‘open science’ institutions. American Economic Re-

Ž .view 88 2 , 15–21.David, P.A., 1998b. The Political Economy of Public Science. In:

Ž .Smith, H.L. Ed. , The Regulation of Science and Technology.Macmillan, London.

David, P.A., Hall, B.H., 1999. Heart of darkness, public–privateinteractions inside the R&D black box. Economics DiscussionPaper No. 1999-W-16, March, Nuffield College, Oxford,forthcoming with revisions in Research Policy.

David, P.A., Mowery, D., Steinmueller, E.W., 1992. Analyzingthe payoffs from basic research. Economics of Innovation andNew Technology 2, 73–90.

Diamond, A.M., 1998. Does federal funding crowd out privatefunding of science?, Presentation at the American EconomicsAssociation meetings, Chicago, January.

Edgerton, D., 1996. Science, Technology and the British Indus-trial ‘Decline’, 1870–1970. Cambridge Univ. Press, Cam-bridge.

Feldman, M.P., Lichtenberg, F.R., 1998. The impact and organiza-tion of publicly-funded research and development in the Euro-pean community. Annales d’Economie et de Statistique 49r50,199–222.

Globerman, S., 1973. Market structure and R&D in Canadianmanufacturing industries. Quarterly Review of Economics andBusiness 13, 59–68.

Goldberg, L., 1979. The Influence of Federal R&D Funding OnThe Demand For And Returns To Industrial R&D, WorkingPaper CRC-388. The Public Research Institute.

Goolsbee, A., 1998. Does government R&D policy mainly benefitscientists and engineers? National Bureau Of Economic Re-search Working Paper No. 6532.

Griliches, Z., 1980. Returns to research and development in theŽ .private sector. In: Kendrick, J.W., Beatrice, V. Eds. , New

Developments in Productivity Measurement and Analysis.Univ. of Chicago Press, Chicago.

Griliches, Z., 1995. R&D and productivity: econometric resultsŽ .and measurement issues. In: Paul, S. Ed. , The Handbook of

the Economics of Innovation and Technological Change.Blackwell, Oxford.

Griliches, Z., Lichtenberg, F., 1984. R&D and productivity growthat the industry level: is there still a relationship? In: Griliches,

Ž .Z. Ed. , R&D, Patents and Productivity. Univ. of ChicagoPress, Chicago.

Hall, B.H., 1996. The private and social returns to research andŽ .development. In: Smith, B., Barfield, C. Eds. , Technology,

R&D and the Economy. AEI-Brookings Institution, Washing-ton, DC.

Hall, B.H., van Reenen, J., 1999r2000. How effective are fiscalincentives for R&D? a review of the evidence. ResearchPolicy, this volume.

Hamberg, D., 1966. R&D: Essays on the Economics of Researchand Development. Random House, New York.

Hertzfeld, H.R., 1985. Measuring the economic impact of federalresearch and development investment in civilian space activi-ties. Paper prepared for the Workshop on the Federal Role inResearch and Development, National Academies of Scienceand Engineering, Washington, DC.

Higgins, R.S., Link, A.N., 1981. Federal support of technologicalgrowth in industry: some evidence of crowding out. IEEETransactions on Engineering Management EM-28, 86–88.

Holemans, B., Sleuwaegen, L., 1988. Innovation expenditures andthe role of government in Belgium. Research Policy 17,375–379.

Howe, J.D., McFetridge, D.G., 1976. The determinants of R&Dexpenditures. Canadian Journal of Economics 9, 57–71.

Irwin, D.A., Klenow, P.J., 1996. High-tech R&D subsidies: esti-mating the effects of Sematech. Journal of International Eco-nomics 40, 323–344.

Jaffe, A.B., 1989. Real effects of academic research. AmericanEconomic Review 79, 957–970.

Jaffe, A.B., Trajtenberg, M., 1996. Flows of knowledge fromuniversities and federal labs: modeling the flow of patentcitations over time and across institutional and geographicboundaries. National Bureau of Economic Research WorkingPaper No. 5712.

Jaffe, A.B., Fogarty, M.S., Banks, B.A., 1998. Evidence frompatents and patent citations on the impact of NASA and otherfederal labs on commercial innovation. Journal of IndustrialEconomics 46, 183–206.

Klette, T.J., Moen, J., 1998. R&D Investment responses to R&Dsubsidies: a theoretical analysis and econometric evidenceŽ .presentation to the NBER Summer Institute, July .

Klette, T.J., Moen, J., Griliches, Z., 2000. Do subsidies to com-mercial R&D reduce market failures? Microenomic evaluationstudies. Research Policy.

Lenoir, T., 1998. Revolution from above: the role of the state increating the German research system, 1810–1910. American

Ž .Economic Review 88 2 , 22–27.Lerner, J., 1999. The government as venture capitalist: the long-run

Ž .impact of the SBIR program. Journal of Business 72 3 ,285–318.

Levin, R.C., Reiss, P., 1984. Tests of a Schumpeterian model ofŽ .R&D and market structure. In: Griliches, Z. Ed. , R&D,

Patents and Productivity. Univ. of Chicago Press, Chicago.Levin, R.C., Klevorick, A.K., Nelson, R.R., Winter, S.G., 1987.

Appropriating returns from industrial research and develop-ment. Brookings Papers on Economic Activity 3, 783–831.

Levy, D.M., 1990. Estimating the impact of government R&D.Economic Letters 32, 169–173.

Levy, D.M., Terleckyj, N.E., 1983. Effects of government R&Don private R&D investment and productivity: a macroeco-nomic analysis. Bell Journal of Economics 14, 551–561.

Leyden, D.P., Link, A.N., 1991. Why are government and privateR&D complements?. Applied Economics 23, 1673–1681.

Leyden, D.P., Link, A.N., Bozeman, B., 1989. The effects ofgovernmental financing on firms’ R&D activities, a theoreticaland empirical investigation. Technovation 9, 561–575.


Lichtenberg, F.R., 1984. The relationship between federal contractR&D and company R&D. American Economic Review Papersand Proceedings 74, 73–78.

Lichtenberg, F.R., 1987. The effect of government funding onprivate industrial research and development: a re-assessment.The Journal of Industrial Economics 36, 97–104.

Lichtenberg, F.R., 1988. The private R&D investment response tofederal design and technical competitions. American Eco-nomic Review 78, 550–559.

Lichtenberg, F.R., Siegel, D., 1991. The impact of R&D invest-ment on productivity — new evidence using linked R&D-LRDdata. Economic Inquiry 29, 203–228.

Link, A.N., 1982. An analysis of the composition of R&D spend-ing. Southern Journal of Economics 49, 342–349.

Link, A.N., Scott, J.T., 1998. Public Accountability: EvaluatingTechnology-Based Institutions. Kluwer Academic Publishers,Norwell, MA.

Mansfield, E., Switzer, L., 1984. Effects of federal support oncompany-financed R&D: the case of energy. ManagementScience 30, 562–571.

Mowery, D.C., 1985. Federal Funding of R&D in Transportation,The Case of Aviation. Paper prepared for the Workshop on theFederal Role in Research and Development, NationalAcademies of Science and Engineering, Washington, DC.

Mowery, D.C., Rosenberg, N., 1989. Technology and the Pursuitof Economic Growth. Cambridge Univ. Press, New York.

Narin, F., Hamilton, K.S., Olivastro, D., 1997. The increasinglinkage between U.S. technology and public science. ResearchPolicy 26, 317–330.

National Research Council, 1999. Funding a Revolution: Govern-ment Support for Computing Research, Report of the NRCComputer Science and Telecommunications Board Committeeon Innovations in Computing: Lessons from History. NationalAcademy Press, Washington DC.

National Science Board, 1998. Science and Engineering Indicators— 1998, NSB 98-1. National Science Foundation, Arlington,VA.

Nelson, R.R., 1959. The simple economics of basic scientificresearch. The Journal of Political Economy 67, 297–306.

Nelson, R.R., 1982. Government and Technical Progress: ACross-Industry Analysis. Pergamon, New York.

Ž .Nelson, R.R. Ed. , 1993. National Innovation Systems: A Com-parative Analysis. Oxford Univ. Press, New York.

Odagiri, H., Goto, A., 1993. The Japanese system of innovation:Ž .past, present and future. In: Nelson, R.R. Ed. , National

Innovation Systems, Chap. 3. Oxford Univ. Press, New York,pp. 76–114.

Pavitt, K., 1991. What makes basic research economically useful?Research Policy 20, 109–119.

Robson, M., 1993. Federal funding and the level of privateexpenditure on basic research. Southern Economic Journal 60,63–71.

Scott, J.T., 1984. Firm versus industry variability in R&D inten-Ž .sity. In: Griliches, Z. Ed. , R&D, Patents and Productivity.

Univ. of Chicago Press, Chicago.Shrieves, R.E., 1978. Market structure and innovation: a new

perspective. The Journal of Industrial Economics 26, 329–347.Ž .Soligen, E. Ed. , 1994. Scientists and the State: Domestic Struc-

tures and the International Context. Univ. of Michigan Press,Ann Arbor, MI.

Stoneman, P., 1999. Government spending on research and devel-opment in the UK. Institute for Fiscal Studies WorkshopPaper, London, 6th July, forthcoming in Fiscal Studies.

Terleckyj, N.E., 1985. Measuring economic effects of federalresearch and development expenditures, recent history withspecial emphasis on federal R&D performed in industry. Paperprepared for the Workshop on the Federal Role in Researchand Development, National Academies of Science and Engi-neering, Washington, DC.

Toivanen, O., Niininen, P., 1998. Investment, R&D, subsidies andcredit constraints. Working Paper, Department of EconomicsMIT and Helsinki School of Economics.

Toole, A.A., 1999a. Public Research, Public Regulation and Ex-pected Profitability: The Determinants of Pharmaceutical Re-search and Development Investment. Stanford Institute forEconomic Policy Research Working Paper, Stanford Univer-sity.

Toole, A.A., 1999b. The contribution of public science to indus-trial innovation: an application to the pharmaceutical industry.Stanford Institute for Economic Policy Research WorkingPaper, Stanford University.

Trajtenberg, M., 1989. The welfare analysis of product innova-tions, with an application to computed tomography scanners.Journal of Political Economy 97, 444–479.

Von Tunzelmann, N., Martin, B., 1998. Public vs. private fundingof R&D and rates of growth: 1963–1995. Working Paper,Science Policy Research Unit, University of Sussex.

Wallsten, S.J., 1999. Do government-industry R&D programsincrease private R&D?: The Case of the Small BusinessInnovation Research Program. Department of EconomicsWorking Paper, Stanford University.

Ward, M., Dranove, D., 1995. The vertical chain of R&D in thepharmaceutical industry. Economic Inquiry 33, 1–18.

Documents

Is public R&D a complement or substitute for private R&D ...bhhall/papers/DavidHallToole RP00.pdf · into the provision of government-provided services, both the defense-related and