World Bank Reprint Series: Number 171

Jacob Meerman

The Definition of Income inStudies of Budget Incidenceand Income Distribution

Do Empirical Studiesof Budget Incidence Make Sense?

Estimating CounterfactualIncomes in Studiesof Budget Incidence(with Parthasarathi Shome)

The Incidence of Salesand Excise Taxes, orWhere Do We Put the Transfers?

Reprinted with permission from Review of Income and Wealth, series 20 (December1974), pp. 515-22; Public Finance, vol. 33, no. 3 (1978), pp. 295-313; Public Finance,

vol. 35, no. 2 (1980), pp. 291-99; and Journal of Political Economy, vol. 88, no. 6

(1980), pp. 1242-48.

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

No 1t S \ND) NILS 1 \NOA

IHF DEFINITI(ON OF INCOME IN STIII)IES OF BUDGETINCII)FNCE ANI) INCOME DISTRIBUTION

13i JAC(OB P. M tiN¶ R sN

International Bank for Recon5l itructioni ani Development

INIR )I)( (l ION

Interest in distribution has recentlv revived. In dIeveloped countries there isincreasing explicit concerni with the size-distributionl of incomes, particularly withrespect to their high and low extremes. InI developing economies, growth of outputas the overriding goal of public policx has been subordinated to conicern with thedistribution of the benefits from growth, and particularly with the persistence ofdeep low end poverty. Since governments directly allocate anywhere f7orm aneighth to a third of total output. increasing concern with income distributioncarries with it, logically. increasing concern with the irncidlenice of pLublic activity inthe distribution of income. As a coniseqtuenlce the nee(d to cstinialtc suclh intidienicecorrectly is also increasing.

Nunmerous researchers have estiMated aspeCts Of buIdtCt incidence througlallocation by income bracket of tax hurden and, occasionally, expenditurebenefit.' The logic of suclh allocation requires that the techniiques usedi havecertain common elements. Thus, to determine tax-burden by inconme level.income per familv before taxes must be e.tindaled. There is no agreement on whatincome would be before gv crniicntl btudget effects orafter all such budget effectshave been accounted for i.e. after reducing inicomec byv tax burdens and increasingthem by benefits received from ipublic e\penditure. Moreover studies whosefocus is income distribution per set frequerntly, if not mis ually, ignore budget effectsand d1efine a concept of income which neither inclu les total taxes as a part ofincome nor in any wav convcernis itself with the benef ts of government spending.Given the magniituide of public btidgets. such cavalier treatment is a serious defectin much empirical work on inicomie (listribution.

In sonie recent work, I appmoached thIs topic from aniother diirectioni I Iow do ihhos s ILI,]% Iig "sIedistribution of intcome handle tisCal Incidence' In 197 1, we recci edi more thaln four dozen empiricaldistributions of income. Of this total, onls 13 considered any aspect of budget incidentce. Nine of these13 countries were in Iatin America, and sexen of these nine were primnarily studies of pubhlic tinanees.The tvpical siie distribution studv inipliciil% assumes neutral budget inciienice. It IS neoteworthiv that inmany of the studies of inicomile distribution, it was impossible to determine how the conrept ofaggregate income was derived.

2The focus is budget icidlenice ralther thani general fiscal incidience. It assumes that the effects ofgowernment budget activity can be sepaiated and analyzedi inidependentlv of all the other effects ofgovernment policies on econormic activity and hence the tlistrihniniion of incomes. 'I'he continuinigcointroversy oni to shat tiegree recenlt empiricwal svirk on bdig(egtt ici(lence and irlncomie dinibuti on.both statistical and analytical, is necessarily invalid becaiuse of a failure to geeneraite a generiliequilibrium approzach is not the subject of this paper. IThe presuimiptioni is that the actual approachescormmonly used are sNuficientlv vaii(d ton make them %k 'rthwhile

5 1 5

TIhis conTfusionl ct ii CCli i tile prop)er delinition of incolme Is Ohs ,Ousl]undesirable. For example. the estimated dlistribLutiol ol tao X tbUtlell deelledS illpart on hiow aggregate inlcomiie is detined. Nexertheless,s it is niot an iire,solvablematter. This paper propolses to rekindle awn inttrest in it. IIopefullV it will hel) leILIto colnsellstis on the be.st defillition olf in<coeli ill uclh x vork. 'vnll il we hlaltd suchConsensus, the 1in ii iit.i i nls problem,,s of using actual t a in eh ii matinc household

antid other inicomiies corresponding to the best dI tifiliiix not to mention tile stillmore intractable probI T-Olem of estilatilng the locus o taX bt den(sTanS et\'vlL il i\ t lI 1.

benefits- would still be with uts. Nevertheless it w&ould be a step forward ifeservoone att least agieed on the basic conlcept to be manipulated.

I uise the phrilase rekkiii(ile interest since. as ustIal, the mattel 15 not wvithouitanteceeieits. lIn tticr lxI 19) hs, thlere wete two basic approaches in emipiricalstudies of budget Incidtince. e'I'Ad -Aijlsteld Naitionail Income Approachr'' 1[21engaged in somiiethlilig ot a debate x Nith the "Net National Product Al)) pilch'',

[II ][9 but they didi not arrive at an agreement. In general howe%ver analystsstudyintg taix or btubldgt tinCidlelnc lhaVe uIeLd thle invalid ap)proach, thlalt is detiedllethie aiggregate as national incomie. or uised an even less cI CLislle conCepll)t.

I lic Illost Sxsternatic uise of the .Akdjusted National Income Approach Is thatitof the Un1ited Nation'. Statistikal ( Commission xhicli itn 11)72 published draftguidelines for emipirical countirtpx Aork In compiling data o, income distributiti[101. In these guidelines, onet foeps is budget incidence 'I a ;iaiN vlc(lilne"' is thte

balsic colcept of acm e1 .i C t inIcomlle betore gox ennimeni eli'ets. It i deltineLl ais totalfactor patimenits betolore S1lt r actinc dlirect taxes. In colitlast, ''consuptliontll' bxhtolsehold is dietinet a totatl factor p-avmile nts less .1is 'ct taxess. plus goveunrilenttransfer pa niments alndl bcnt til', tmitoliL eminlelnle1t expenditures, plus anil estimliateof similar flows within tihe ps .tte economy . T'Ihe guidelines exclude intdirect taxesfrom pre-tax incomeno It is atigued beloNw that this treatment insalidatc, thetapproachi.

Pari t I of this paper discu 1 essts the Adlste d National Incomiie A\pproach, andPairt II thte Net National P ioduct Approaich. Part Ill presients an analysis of thedifferences, and gikses thl euthir 's colu*sions as to tile approplriate conlcept.

l. Iii Aimi sri i) N t1rs xi iNioMi APPROAVi1

In constrtctinig i!er .a tc pryt tax ilionioe (thie n ii t Ii to be dividedaniimi inIcomle brackets). the Ad justed Nationial Inion e Approach uises personalilncolmie as thle pivotal collcipt.' -I his magnitude is thieni iniereaisedi 1v ( I ) thlose taxeswhliich are assUmled to bllLr1dii fact irs of produiction directly (tinsi ifte(d cvplloriate

profits talx, unshifted export taxes, backw%vard shilfted( portion of thtie epllover'ssocial security contribution): and (2) othler intcomiie (1iidlist ir teil profits, Capital

gain1s); atnd(3 decreased bx t3) personial transfer paymentils. the it1111iii ;111ai

melinte(d magnitude us rft-er redi to by dliftiient aulthols aWs ''ali julSted( ilIt111c', Or"broad incomie". Ihe distinguishing feaituire of the concept is e\ctirsion fromaggregate incomrie of all taxes *l ilich atr believed shifted forward to constumers.

I'he tIN guidelines assuic tihar theme aio w, htetivts io hioir'htidti. fr,tti puliti genieralexpenr1itures. suCI a'. diefense adLI adminisiration

'National in,oniet .ontcepit uwtd are ths ft ttie t S t)eMpartmneit ti ( oimiuJc,c

,> I o}

Hence it excluides all indiirect buLsiness taxes. But it also excludes items such as thatpart of social security taxes and corporate income taxes which are assumed to beshifted forward to final constumers. The rationale for this procedure is that weretaxation eliminated, the (money) income of factors, e.g., corporation sharehol-ders, would increase only by the amount of the unshifted burden. This procedureis illustrated in 'I'able 1, which compares components of the "broad income"concept with corresponding national accounts data. It is taken from a very carefulstudv of U.S. budget incidence for the year 1 96() [2]. The major tax componentswhicih the studly assumed shifted forward to consumers are:

(a) forward shifted corporation income taxes (22.3-14. 1) 8.2(b) social security taxes not shifted back to employer(20.7-6.6) 14.1(c) indiirect buwincos taxes 44.3

66.6

IABLE IN NiiN,Ai A(,(o1 Nrs ANi) BROAID IN(oME C)NC(PIS, 19()

(billions of dollars)

Vationa l. Acounts 1-roadtl hltiome Cloncept35 1.8 I)isposable itncomen

n0.4 Personaitl taxes

4(2 .2 Perstonal inicomiie Ixirnilv personal inconme 383- 3t.i IFransfei paymnients Plersonal transfer payments 26 S

N n disodistiibtited tn!pi te profit' tTndistributed corporaite profits 8.X'2o3 ( orporation inmcolmle taxes I. n..hied portion of the

corporation income tax 14.1'20. Social security taxes

Backwardl shifted portion ofthe employer's social securitycontrt ibtition 6.6

( apital gain-s 11.7

41 10 National ineome Broad income 398()44.3 Indirect business taxes

461 .3 Net national product5,04.4 (iross national prodtuct

IDiffers from personal income in excluding income received bv institutional residents. militarvpersonnel overseas, and those not living with their families, and income retained by non-profitnstitutions and private trust, pen!sion and welfare funds.

'Includ,es interest payments.So(urces: National Accounts: [IIl. p 6t,; Broad Income C 'oncept: [21, pp 1 - 175.

Items (a) and (b) are defined as part of factor incomes in national incomeaccounting, Item (c) accounts for most of the difference between National Incomeand Net National Product. Although all three of these components are excludedfrom aggregate income before taxes in the Adjusted National Income Approach,thev are included as reducing income in calculating tax incidence. Attempts aremade to estimate who bears their burden in aiccordanLlce with information on howdiifferernt income brackets or groups use their income to purchase the relevant

517

taxed products. In other wordls, to compare incomiies after estimating tax-incidence, "broad income" or "adjusted national incomne" in the various incomebrackets is reduce(d bv "diriect" taxes, definedl zas all taxes whichi buri-den factors ofproduction directly, plus the amount of the estimated forwar(d shiftedl tilx (items(a). (b) and (c) abovc) "embodied'' in the goo.ds and services conisutmedi by thev-arious income brackets.

We are left wvith the paradLox that tlliiLtninilride, corresponding to certain taxeswhichi are incltu(ded as factor income in national accounts data- that i iincldled innational income, defined as the sum of factor ilncom11es- alre treLted ats n1Ot eXiStinlgin studies of tax inci(dence usinig this approach.

T'h1e Mfc'wimiii of tlit' .Ntitiotil Income ( Concept

TIo this point the argumrient has prroccede(i as thotugh national iTncoUeti doesprovide a measure of total factor incomiies. An implication of the Ad justedlNational Income Approach is that taxes other than indlirect taxes have the sameeconomic effects, i.e., they buriden the consumners of p1r)LdtiitN taxed. To the e.l. utthat this is true, then the distinction between NI and NNI' as conventionallvdefinedi and empirically aFpplied is not meaningful for economic analysis. Iln oth}erwords the "forward shifted' corporation income tax tLnd social securitx t: A arejust as ''indirect' in their inicidence as an excise or sales tax. Note thalt neither' theshifted or unshifted parts of the corporate incomiie tax are ever receise(d as inicomeby the shareholders. 'Ihe important distinictionl is that with respect to th.e unshiftedpart, elimination of the tax presumnably would restilt in shareholder iniconleincreasing, in thtC uII1'shifred amount. But in the case of the part shifted forward tothe consumer, elimination would result in increased real incomes of consumnerrs. Inthis view the whole notion that conventionally defired(i national income is equal tothe sum of factor incomes is seen as misleading. It wotld he useful to redefinle theconcept to excltude all taxes 'shifted forwar(d to conisumtiers 'I The result might bevery close to the adjusted national income concept. Sinice we arec a loIng way from a1consensus on this point, the second best will be to recogni/c fuil this short-coming of national income as currentll detinied.

If we leave to the side the treatment of capital gains, thus far the analysis leadsto the conclusion that the Adjusted National Income Approach can really heregarded as an aittemllpt to ftilne a mearnin1gfUI COnICept of tOtall f;loI( l ilh'1i1Co s. or amore valid measure of '"national income", than the convenitional one. As acornseqIuence it nmight he more meaningful to (describle thIis as the (CorrectedNational Income Approach.'

2. TIni Nj I NA.It )ON- PaRD)1- ( I ArPPRt O Ii

(iiveni a "corrected( NI concept. the queslion reminsizi wlhethler "corrected"NI is the appropriate in'comtle concept in measuring fiscal inlci(lentce. At fir'stapprox imation, this appears to be the case. C'orrected NI woul(d add to factorincomes, which are the total incomes lhouisethold(s have to spetld. Neverthelessmany researchers use NNIP as the relevant concept. 'I'he problem here is again

In 196b4 Nusgrave took the pos9ition that 'net-nationial prmilt aIt [actoit cos%'' was precisely thewrong concept. althoiugh his approach anountts to redelirning thalt coicept in a tolre logicadl aIndmeaningful lashion (['. 1), 54)

5 18

treatment of 'corrected" indirect taxes (defined to include all taxes which burdenconsut)nption). Since they are paid for out of household disposable incomes, howcarn they be imputed on the income side'? As Bishop recognized, this appears to be"double counting" ([ I 1 p. 383). In short, factor incomes do not sum to the marketvalue of output. but to n-tional income, whiclh must be increased by indirect taxesto get outiput valuedz at market prices (NNP). Any size distribution of net finalOlutptut at market prices (NNP) will therefore exceedl factor- payments by the'correCted" Indirect taxes. Henice apparently distributing NNP means distribu-tinig -incomite- whliich f actors would not earn even if taxes and public expendituresdisappeared. Moreover some inidividuals reason that factors do not consume theentire NNP. even after asigning all benefits of public expenditures to them.Assume purchases of good(s an1d services are burdened solely by indirect taxes.'Wlhenl that buLrde n is distributed in the process of defining income after payment oftaxes, the result is after tax ineon e less than national income by the indirect taxburden. Addinig to this government outlays-assumed equial to indirecttaxes gives a magnitude less than NNP. again by the amount of indirect taxes.Making the exercise more realistic by also considering direct taxes and incomes,and the corresponding iniere Ved public expenditures, in no way affects thisoutcome.

Nc erthele>s in an earlier article addressed to this topic Bishop ([1] p. 388)defended the NNP Appro(ach yi arguilng in effect that the "income base" shouldbe NNP with

imputed items Of income being allocated in proportioil to some index of theassumiled distribution of the henefits of the output involved. This conclusion isdrawn on the assumption that it is a useful procedure to attribute the burdenof all taxes and the benefits of all government expenditures to individuals orfamilies in their individual capacities.

However, as shown above, allocation of all taxes and all benefits from publicoutlays is consistent with the "corrected" national income approach. To do so oneneed niot assumlne a NNP concept of income.

. Iii Ax . xi 'sis %\D) (0N(l.r'Si()NS

'I'he analysis which follows approaches the question of the appropriateincoimi concept in a diitTerenit manner. It focuses on the difference in aggregatetax-burden implie(d by usinig the two alternative aggregate income concepts,"corrected" NI or NNP.T'he :inalNsis uses the following notation.

FI: total factor payrnen tsNNP: net national product (market value)

NI: "corrected national incomeIT: "corrected'" indirect taxes

DT: "corrected'' (lirect taxes(G: governmnent expendittires incluldingg personal transfer paymentsD): capital consuimption allowances

GNP: gross- naztioiizal product

'"Ilere and after indirect taxes means "corrected" indirect taxes.

519

Ass ui nptio)ns:(a) All resour-ces are fully employed.(b) All government revenues are taxes.(c) The budget is always balanf .(d) Indirect taxes are defined as thiose taxes which burden consumption.(e) Fl = NIGiven the above assumnptions. since Fl =NI. if we use thle National Incomle

Approach post-tax factor income will be Fl -(IT + D)T). L)istributing first FH andthen Fl -- (IT + DT) by income brackets gives a measure of the tax burdeni on thedifferent income groups permitting researchers to compute pre- and post-taxincome inequality. This in essence is the Adjusted National Income Approach.Applying this reasoning to those studies which use NNP as the concept of incometo be dlistributed gives the following:

(1) NNP =FI+ IT(2) Fl + ITr - (IT1 + DT) =Fl - DT

Post-tax income is now larger by IT than in the national income case. In countrieswith no direct taxes, we end up with a post-tax distribution summning to NIf; inshort a near reverse of the earlier paradox. In the same countries, the post-taxmagnitude of the National Income Approach would be less than the correspond-ing magnitude of the NNP Approach by the value of inidirect taxes. Since DT aretreated the same in both approaches, the above example is relevantt to the basicquestion of which is better.

A simple wayv to examinie this involves use of a hypothetical example asillustrated in Table 2 below. UJsing NNP as the basic concept of inicomec to be

TABLE 2

Situation I 11 III

Direct Taxes 33 20 0Indirect Taxes 0( 20 50NNP 10 120) 150NI 10 10 0Income after Taxes

(Current Prices) 67 8) 100Income after taxes

(Prices of III) 10I0 100NNII Approachl tITI i- LT) ( I I ± II1) 0.33 0.33 0.33"Corrected" NI Approach (1'1 + D)T) II1 0,33 0.40 0.50

mecasured, a common measure of tax burden (taxes/NNP) gives that burden as athird of NNP for each of the three tax situations. However, use of the national-income measure (taxes/NI) results in increasing tax burden as the tax systembecomes increasingly inclirect.

Assume that the change in tax mix is neutral in moving from Situation Ithrough III, in the sense that the mix and total of output remain unchanged. (The

7Throughout the paper relevant elements in the equations can refer to indlividual households byadding the proper subscripts. Hence NNP =Z.(FP, + IT,).

520

assumFptioni of unchanged output is unrealistic, but it is not crucial. A morerealistic, i.e.. (mnplicated. example would yield similar although less obviousresults.) UJnder these circumstances, the difference in NNP between situations IIIand I would be solely that prices would be S0 percent higher in III than in I.

T he table implies that after tax income in I is 67, or 100 in prices of III. Thisalso equals III's NI. Bv the same token NNP in I is 1)00, equal to 150 in III. If therewere no indirect taxes, the factor incomes in III at III's price level would beequivalenit to 150. In a situation where all taxes are indirect, NI already is ameasure of total incomne after payment of taxes. More generally, the buying powerof Fl will be such that relative to NNP it will always equal total income afterdeducting indirect taxes. At the sinflated prices (relative to I) of III, to get thepre-tax incoome concept we need to add back indirect taxes to Fl to arrive at NNP.Specifically, this consists of allocating IT (all taxes which burden consumption)according to the received canons of tax incidence to the various income brackets.

This becomes even more obvious on considering a variant on the change fromSituation I to III: Assume resources continue fully employed, and the tax-burdenremains one third in terms of NNP. Assume one change, namely that pricesremain stable, i.e., NNP continues at 100. This implies that on removal of DTandimposition of IT in the new equilibrium the tax burden is reflected in decreasedfactor incomes: producers-under the assumptions-will not increase prices, butmust forward the IT' receipts to government. As a conIseqLuence, factor incomesfall in the amount of the tax-burden or tax receipts.

Clearly the resulting national income would equal post-tax income. Toestimate pre-tax income it will be necessairy to distribute and add back IT amongthe various incomie brackets, according to assumptions concerning tax incidence,

All of this leads to the unequivocal conclusion that the NNP Approach is themore appropriate of the two. Use of the National Income Approach implies a totalincome concept before taxes which is already net of indirect taxes. Use of the NNPApproach itn effect includes indirect taxes in the basic pre-tax income. Operation-ally the significance of this distinction is avoidance of an exaggcrated measure ofaverage tax burden.'

From another perspective, the resource claims called indirect taxes representpurchasing power for government. Corresponding to these claims paymentsneither are made nor cani they be imputed to the factors of production. However,since the logic of budget incidence analysis requires that all output be distributedto private claimants, NNP beconies the relevant concept precisely because itexceeds factor p-aymeints or national income by IT, in (ther words, by the amountof such resource clainis.'

Ihis conclCIsiton makes even more difficult the task of generating pre-taxinconmes to estimate tax incidence or for other uses. Most researchers assume thatin(lirect taxes usually reduice incomes of hlouischolds consuiming the taxed items. Inthis new argument, we see that factor incomies are alreadv net of inclirect taxes andneed to be increased to whait they would be, i.e., NNP, were there no indirect

'It is understood that the discussioln is in terms of the basic conceptual approachi. I am not arguingthat one should niccc ... rild use unadjusted NN? as the basic concept of income to he distributed.(ertain alteiiatiinis. e.g., inclusion of capital gains, may be desirable.

lBishop expressed the same idea (11, p. 3.8).

521

taxes. I'his involves thinkinlg about incidence very diffeeretly from the usualapproach. In other w.ords, indirect taxes are niow regarded as overwhelminiglvburdening consumption, at least in empirical work on tax ilicidle ice. 'There is noreason to believe that the pattern of incilienice of such taxes as nowconceived-which is derived from patternis of consumption-would be the samllepattern implicit in "restoring" factor incomes to NNT' to get a conceptually morevalid measure of total pre-tax income. In short, increasing the factor incories ofvarious income brackets by the presumed incidenice of intdirect taxes a1CCording tocurrent calnons of tax incidence would not be a valid procedlure.

RIt [-RI NtIS

IIl Iisthop, (George A.. lI'' ltile Redistiiuition In tlh) Irameork OT tht' li National IntconrcAccounts', Natiottnal Faci Journal. Vol 1'1 Not). 4. De)'cember I ')bt.

121 Gillespie W'. ., "E ffect of Public f penditture on the D)isttibution of Income"', Musgras.Richard A.. ed.. Essavs in ['iscal Federalism t Brookings Institution, I%Sti S

[31 Herriot. Roger A.. and1 Miller. IHernmain P. ''Who Paid the Taxes in I JXfs '. minicogrph.t I S.Bureau of the ('ensus. paper prepared lor the National Industri.al C onferent BoardI M;ticne.New Y'ork. 1(711

141 McILure, Charles F. . 'lhe Incitdence ol Taxation in Colonibia''. in Musgrase and G;iliso. eds .Fiscal Re'foroi for C olomnbia, Final Report and Staif Papier% of tit' ( 'olombihan ( ooti10riis1su OIt l:XReforin IHarsiard t ritix ersit I aiw School. ( anibridge, NlMass., 1 ()7 1)

['l McLure. C'harles F.; I'he Intcidente ot I ixation in West Malinsia. Plapei No I -. Progrlam oifDevelopment Studies. Rice Cnrxersiti. I.ill. 197 I

[0] Meerman. Jicol) . 'P Liscal Incidence In F mpirical Stuidiies oi I nuline Ic)isti 'uiion in Poor('ointries'. Ae nctsn foi International Dewlopmcnt, D)iscussion Paplr -o N *s. 1')'2

| l Musgilae. Rlchid N . '' stilinating the D)istrli'tion of the I ax 13Ihdien Ii I \D1)1 loIut I is

Progrtam. P'rolul'mus of Ittx Admouvroii s tioi iti I at11) Xmt,)i'( t. IP1 op'irs widilI cc , i;s 'I aCon(',renle If eld In MiwIini Aires. ( ), to,hr J 'Jt I ( The .lohnus r lopk in,s 11.1111 inii. I crr-

I SI Snodgrass. D)onald R., 'Ihe Fiscal Svstemin as iatn Itncomtle Redistributor in Viest aa;ia.st.''Economic D)eselopment Report No 2 24. Development iRes.irelr (hiroup, ('enter lor Interna-tional Affairs. Harvard University. 19)2

[9] TaxFoundation. Inc . ''Ta\xBurdensand Bnoenelitsof (ioei tnmentIF sBendituies K Income ('lass1 ')( 1 and 1)1)5,'' New Y'ork. It,i '

[t10l 1_.N1

. Economic and Social Council. A )rOtt Sv%tc m o'f .S'tatstt s (of th I)istriltiion of Income,t.

Consumption anidi Accumtulation. Statistical Comininssion, Scx enteenth session. Itemii 5, of thepros isional agendia. E ( 'N 3425'. IFeb). . I1'02.

1111 tUS. D)epartment of (ornmerce. Auiv of ( urrent Bwsiness. April 1 )t2,

5225

DO EMPIRICAI STUDIES OF BUTDGETINCIDENCE MAKE SENSE? *

by

JACOB MEERMAN**

I.INTRODUCTION

Increasing concern with distribution has led to numerous recent estimates of theincidence of aspects of public activity. In addition to the traditional focus on taxincidence, economists have turned to estimating the incidcnce of specific govern-ment programs, frequently within a cost benefit perspective. More generally, andmost recently, there have been numerous estimates of the incidlence of total publicexpenditures. Usually this has been carried out in a country study (omplementIed byan estimate of overall tax incidence. Researchers have frequently combined the twotypes of incidence into what might be termed budget incidence. 1

This broadening in perspective and analysis has also brought increased con-troversy concerning the meaning of incidence and how to measure it. Measuring taxincidence be it a specific tax or an entire system is increasingly viewed as a generalequilibrium problem which is to say an unsolvable one (as argued below). Never-theless, attempts to estimate incidence of entige tax systems proceed as though taxeshave neither excess burden, nor other general equilibrium consequences. On thespending side, the meaning of expenditure incidence is itself in doubt. And thetechniques for its measurement are regarded as even more dubious. Normally, webelieve that benefits from public expenditures need not equal their resource value,while taxes always equal their resource value. The consequence of this outlook hasbeen failure to recognize the profound symmetry between the incidence of taxesand public spending. Finally, incidence work requires comparisons between statusquo and an imputed state or "counterfactual" such as a world without governmentbefore the tax/expenditure change. This imputation and its implications are fre-quently unclear.

Public Finiance / F'inances PlSlXgiiuc No 319's Vol. XXXIII/XXXI1lline Ann6c

*6¢

296 JACOB MEERMAN

This paper analyzes these and related difficulties and attempts to resolve them.

The method used to do this involves simple algebraic formulation of the household

(HH) budget constraint to permit its decomposition into the various kinds of incidence.

(Such formulation has an advantage in that it readily translates into the accounting

framework actually used in empirical country studies). The attendant analysis puts

the difficulties into a different light making some of them less formidable, or show-

ing them to be non-existent. It also results in a useful clarification of the meaning of

the various measures of incidence.Needless to say, in ground as well worked as incidence theory, very little can be

offered whieh does not have antecedents. Much if not most of what follows has been

discussed, or at least touched upon, in earlier writing.' The paper pulls togetherthese various strands into a basic framework. It is both a summing-up, and hopefully

a slight movement forward. But, unlike much of the earlier work, its point of

departure is the comprehensive empirical country studies of budget-incidence, that is

the attempts to measure the combined incidence of taxes and expenditures for a one

year period.

II. ANALYSIS

A. Clarification of B'asic Concepts

Incidence is defined as the total change in the distribution of HH income -

including publicly provided goods and services - due to a government intervention.

Notice that the change is in income, not welfare. We are, of course, primarily

interested in the latter but cannot measure it. Income, however, is usually measur-

able and presumably correlates highly with welfare. Consequently, we work with

income which is measurable as a proxy for welfare which is niot.

In the paper, the basic perspective is balanced budget incidence defined as the

total change in the distribution of HH incomes due to an increase in taxes used to

finance an exactly equivalent increase in expenditure. We are also initially concerned

with an extreme case, namely the movement from a world without government to

one with a government financed solely by taxation with a balanced budget.3

Thus far, the discussion is straightforward. Hlowever, if we attempt to decom-

pose balanced budget incidence into components to support analysis, we inevitably

run up against a good deal of coinplexity and confusion. Much of this is due to the

lack of interest in this issue until (literally) the last decade. In brief, "the methodo-

logy and theory of estimating benefit and expenditure incidence is largely unde-

veloped" [I1, MeLure, p. 21In (dealing with the same problem, McLure resolves incidence on the spending

side into two comnponents, ('xpenfdit tire incidelnce (how government spending affects

E'STIMATING COUNTERFAC'TtJAI, INCOMES 297

in the counterfactual state I for the decrease (increase) experienced in state 2because of indirect taxes.

This discussion indicates how important it is to explicitly consider the effects oftaxes on relative factor and output prices in generating the state 1 that is thecounterfactual incomes. The inability of the CSS to adequately deal with suchchanges in relativ-e prices limits their usefulness even as short run studiies, in whichtechnology, factor endowments, aggregate employment, and saving andinvestment are assumed independent of budget activity.

Even if the studies are circumscribed in being explicitly short term and limited tothe tax side for which the counterfactual is a comprehensive proportional personalincome tax. the necessarily unconsidered relative price effects may implysubstantiallv different results from those hitherto reported.7

NOTES

* The authors are Senior Yconomist. the World Bank, and Associate Professor, the AmericanU7niversity. Washington. D.C., respectivelv.

Of course changes in :eLbnolog%, labor supply, and so forth will also affect relative prices.Meerman's categories interact orn each other.

2 MlClrure and T'hirsk l:'l. Ballentine and fnris [21, Va.ndendorpe mnd Friedlaender [141, Ratti andShome [12, 131. .ind others.

Thle case of mnonopoly elements in the corporate sector has also been considered by fHarberger ['I.'rhe effect of introducing monopolv is that the tax will fall on monopoly profits as well as on the ordinarvreturn to capitail. The part which f'alls on monopolv profits will be borne by monopolists. That partwhich falls on the ordinary return to capital will be distributed bv a mechanism similar to the competitivecase (pp. 160-162).

4 And are not shifted to labor or toi consumers.lFior example household incomes are treated as independent of incidence assumnptions in Andic [1 1,

Dodge [4], Fran7en et al, [S5. Gillespie [61.6 [)odge [4, p. 71 for example points out that government budget policies "... affect personal income

indirectly bv aff'ecting the composition of output and hence changng bo !h the relative and the absoluteprices ot' final goods and services, and of factors of production". Theoretically a study should take theseeffects into consideration. To do so "...it would be necessarv to calculate the level and distribution ofpersonal incomes that would hIave existed in the absence of the activities of the public sector. Thiscalculation is not feasible as the behavioral relationships on which such a calculation could be based arenot available, nor, in practice, estimable".

Browning [31 discusses this isstuC at some length in an a priori manner, and concludes that relativeprice effects on the product side, when using the proportional income tax as a counterfactual, areprobablv offsetting.

298 JACOB MEERMAN

as the value to the recipients of the resource transfer made through the publicexpenditure. We use the definition of tax incidence as the resource transfer awayfrom those who pay the tax. It is defined as equal to the amount of the tax.Associated with this rechanneling of resources -in our balanced budget context -

we have what Musgrave refers to as Ricardian output effects, namely changes inoutput "due to resulting changes in technique, voluntary changes in labor supply,changes in saving and capital formation or in the efficiency of resource use" [ 14,Musgrave, p. 2081. In ad(dition, there may be change in the level of unemployment.Equally importanit may be the changes in relative prices of both products and factorswhich also affect factor incomes. The net effect of all these we call RPTO incidence,the abbieviation for relativp price, techniques and output incidence. Defining in-comes broadly, RPTO incidence can be defined as a residual: Thus for a given HH, agiven balanced budget change eventually settles down to a new equilibrium with anassociated net change in after tax income, broadly defined. The income change less thesum of tax incidence, with negative sign, and benefit incidence must be identicallyequal to RPTO incidence. In another perspective, it appears accurate to think ofbenefit incidence and tax incidence as the direct benefit and burden on the house-holds ultimately paying the resource transfer or enjoying it and the RPTO incidenceas the associated general equilibrium effects. Conceptually, and occasionall: inpractice, the JRP7T incidence of a given tax or expenditure can be analyzed inde-pendently of all other taxes andior expenditures. This possibility is also brieflyconsidered in the paper. These basic concepts are defined and used in the modelbelow.

B. The Miodel and the No G(crirnnent Counterfa. tallBasically, the model consists of dvfinitioim and manipulation of the household

budget constraint in a way which permits its decomposition into the various kinds ofincidence. Many of the assumptions in the model are unrealistic. However, thegenerality of the conclusions does not depend on the realism of the assumptions:Increasing the realism of the model would only reinforce the conclusions.

In the model, all income in the economy is received by households and consistssolely of returns to factors, physical or human. The factor labor is defined as thecapitalized value of labor income. 5 Hience, only factors and their returns -

dividends, interest, wages, et cetera -- appear in the budget constraint. All factors areassumed privately owned, but they have no exchange value after their initial sale orallocation. (Were this not the case, the measurement of income would be greatlycomplicated.)

The analysis uses coniparat in4 statics. There are three equilibrium. states, de-noted by the first subscript of each variable. The first state refers to an economy in

DO EMPIRICAL STUDIES OF BUDGET INCIDENCE MAKE SENSE? 299

equilibrium without government. The second and third states involve government.The model concerns solely these equilibrium states. Problems resulting from themovement from equilibrium to equilibrium are not considered.

There are Q households each with its income Yj(j = 1,2 ... 9) based on the earn-ings of its factor endowments. Factors of the jth household are indicated by Ak/(k= 1, 2...m). Each type of factor (including capitalized labor) earns a one periodreturn equal to Rk(l, 2...m).

There are n private goods and services Qfi = 1, 2... n) and n prices correspond-ing to the various Qi designated as Pi (i = 1, 2... n). Returns per unit of factor aswell as prices of goods are identical for all households due to perfect competition infactor as well as goods markets. There are no externalities or economies of scale.Thus, Pi equals marginal cost.

Initially, the world is without government. 6 All outputs are completely private.There are no inter-household transfers. Thus, in this first equilibrium state, there is abudget constraint for each household j:

(1) Ylj mP i Q i n A

Qli is the amount of the ith good purchased by the jth household whether forconsumption or investment. (Note that asset accumulation through investment doesnot affect the supply of factors available since this is only a one period model.)

Now introduce government which is funded solely from taxes, has a balancedbudget and in turn finances either completely private consumption or investmentgoods which are furnished directly to households or a completely public good ortransfers.7 Thus, in the Second State, assuming equilibrium is established, we havefor household j:

nn m(2) j 2i + A= BP Q2 i + B2 G2 ;k R2kA2k

where:h r - marginal value to recipient household of Qi financed by government,

(i = 1, 2 ... n). (In terms of the usual utility analysis,

B/P = MUQi/MUy.J

G = quantity or total output of public good

Bg marginal value per unit of G to household j.(In utility analysis Bg = MUG/MUy).

A few implications of relation (2) are worth emphasizing. Consistent with ourfocus on income, benefits are measured in terms of value analogous to the measure-ment of value in markets: Just as P1Qi purchased by a household is taken as the total

300 JACOB Ml,ERMAN

measure of the value of Qi to the household, so BPQ or BgG equals what thehousehold would be willing to pay for government supplied Qi or G given that theirmonetary income equAlled Y1. Hence, we are not attempting to integrate consumersurplus into the analysis. This is analogous to what is done in measuring tax incid-ence where tax burden is inevitablv defined in value terms, not in terms of the totalreduction in utility (consumer surplus) associated with levying the tax. In otherwords, the estimate of benefits from such public services "should be based uponmarginal rates of substitution between the services and other (private) goods andservices. This approach ... involves ignoring consumer surplus provided by infra-marginal units of the ptublic service, just as national accounting ignores the consumersurpluses inherent in consuimption of private goods".8

The analysis assumes, realistically, that government provides publicly financed(private goods) at zero cost to the households, typically rationing their use throughnon-market mechanisms. Consequently, the marginal value to the household of pub-licly financed (private good) Qi and its public cost of production or its private costof production are not necessarily equal. This can occur whenever household con-sumption of Qi is solely through public provision. If, in addition to governmentallotment, a household purchases part of its consumption of Qi privately, BP! willequal Pi. However, even in the latter case, P- need not equal the marginal cost ofpublicly financing Qi (defined as be) because Ct < Pi, depending upon the relativeefficiency of government finane c.

Note that housy hold income (Y 2 ) is measured after all government effects havebeen considered, and that each household consumes all of G2. The existence ofgovernment means that household consumption plus saving is no longer identicallyequal to factor income.

Spending per household j from factor income in state 2 must equal

n at w" "2i Q2 i - Ž R2k 42k hi=1 k=1 h=l

where: Th tax or negative tax, namely transfer, (h = 1, 2 ... w).

Consequently, we can rewrite relation (2) as

(2') Y2 R iAi+ B T22' *- 2k A2k -- =T2h + 2 t' MQ 2 t + B22] k=l h 1 i-] 1 2

In the typical empirical country studies of budget incidence, benefit incidence isn w

defined as B2 Qi I 9G; tax incidence as Ž2 Th; and RP'O incidence as

zero,9 One crucial consequence of the latter assumption is that factor incomes in

DO EMPIRICAL STIJDIES OF BUDGET INCIDENCE MAKE SENSE? 301

mState 2 (E RkAk) are defined as equal to factor incomes in State 1. Since factor

k=incomes in State 1 are defined as equaling household income (Ylj), this leger demain in turn permits comparing the change in incomes between the two states: Y

mis simply assumed equal to k1 R2kA2k.

But of course RPTO incidence is by no means zero and it is precisely theassumption - all too often implicit - that it is which results in some of the highlevel of controversy concerning the empirical country studies. What is clear is thatbecause of RPTO effects, we can not have a very good idea of Y1 j. It follows thatempirical country studies of budget incidence as well as tax incidence alone, areinvalid if they claim to be stuidies comparing our two states of the world. Thisconclusion is elaborated below, at first as concerns the effects of changing relativeprices in the context of the Lindahl Solution; and then very briefly in terms ofchanges in techniques, output, and its composition.

C. The Lindahl SolutionIf in State 2 we also have solely efficient benefit taxation, such that marginal

value of benefit received = tax paid (a Lindalil Solution), then Relation (2) above isan equation for bousehold j and for the economy as a whole:1 0

n(3 i Z P2ij Q2ij + I. Ž BiiQ2u + Bj G2 k 2kjA2kj

And for each household j.

w n(3') h T2h= BPiQ2 i + B2jG2h=1 Th ij2

The belief is widespread that a Lindahl Solution to the problem of financingpublic expenditure is necessarily Pareto superior,1 1 that is the household utilitycorresponding to incoine in the second State (Y2 j) equals or exceeds that of incomein the first State (Y11). However, this belief is false because government taxing andexpenditure affect relative prices, returns to assets, and/or asset endowments.

It nmay be useful to belabor this point. Given a constant level of prices, inempirical work, we normally assume that the larger of two incomes is more desirablebecause household utility is assumed to be a positive function of income. However,if reiative prices associated with the two incomes are radically different - say as theconsequence of substantial changes in government taxes and expenditures - a givenhousehold may prefer the lesser income with the more favorable relative prices - ashas been amply demonstrated by Hicks, Slutsky and others. For this reason even asuccessful comparison of measured incomes in two different states of government

302 JACOB MEERMAN

activity does not necessarily give unambiguous results. In terms of our LindahlSolution, the fact that all first order conditions in both private and public sectors aremet for an optimum, which is to say inter alia that for all households marginal valueof public output equals the tax price [Equation (3'T) does not mean that all house-holds are necessarily as well off as in State 1. The Lindahl Solution is Paretooptimal, not Pareto superior.

We can make this point symbolically by going behind income to householdutility. That is:

(4) IT2 1 - U 0,

where U1T - utility correspondinig to Yj and subscripts 1 and 2 refer to States 1and 2.

Pareto superiority implies I2 - T ;. But as noted, this is not a necessary con-sequence of (3). UT,j [-;j < 0 is also possible.

I'he non-Pareto superiority of the Lindahl Solution by implication also rein-forces our earlier con! :usion that meaningful comparison of incomes between States1 and 2 is impossible: Even were it possible to determine the Ylj in some sort ofdeus ex machina fashion, we would not know what meaning to attach to it asconcerns welfare (or "real income") becauseof theprobablyvery substarf'- 1 changein relative prices.1 2

D. Subsidies in roduic iionWe can also develop equation (2') by assuming that the newly introduced

government also provides subsidies in production (including completely free inputs)such as the cost of agricultural extension services or the subsidy element in loans atsubmarket interest rates. Thus, we revise equation (2') for household j as follows:

(5) Y2 = k1 A -412k + k1 '5 2f -i= T2h + § BPiQ2i + B§ G2ki=l 2i1Qh=1+ B1 G2

where we assume that there are r subsidies, Sf (f = 1, 2 ... r).Subsidies in production are analytically identical to taxes with opposite signs. Recallthat in the case of perfect competition, indirect taxes are passed forward to burdenfinal output. In our model, which assumes perfect competition, so that all produc-tive subsidies are available to all firms, productive subsidies are therefore passedforward to final output as well. (It may also be well to recall here that the conceptof factor returns (R) is net of indirect taxes and production subsidies. Thus, outputprice always equals total factor returns plus indirect taxes less subsidies.)

In contrast to government financed private goods (BOQN) such as social services,note that the benefit incidence of production subsidies - as in the case of taxes - is

D)O EMPIRICAL STUDIES OF BUDGET INCIDENCE MAKE SENSE? 303

defined as equal to the subsidy: A subsidy is assumed to always reduce the costs ofproduction by the cost of the subsidy, just as an indirect tax has exactly theopposite effect. Consequently, the problem of value to consumers being differentfrom costs, which is freqently so pressing with respect to publicly provided specificgoods, does not arise in the case of production subsidies. All that has gone beforeconcerning RPTO effects of taxes and expenditures is equally applicable to produe-tion subsidies.

E. Effects on Techniques, Output, and Its CompositionThus far, the focus has been goods produced in both periods. In reality, taxes,

public spending and other public activity affect saving decisions, occupational de-cisions, and the development of fundamental institutions. Consequently, not onlydo relative prices and outputs change, but also factor endowments, and techno-logical relations that is the "very nature" of production itself. Thus, once govern-ment exists, there may be production of things like battleships, public parks, andsocial security systems. On the other hand, certain outputs may be repressed such asunlicensed practice of medicine, or unlimited publication of newspapers. Whichoutputs are repressed and which produced of course depend on the type of govern-ment.1 3 In short, to compare the distribution of income in an economy withgovernment to what it would be without governiment is a bit like comparing thelocomotion of the adult frog with the progenitor tadpole. Making the equationsmore realistic by adding externalities in both private and public goods in no wayalters this simile.

The only empirical comparisons which it may be possible to make rigorously are(i) those involving small changes or (ii) those in which R, A, P, and Q are independ-ent of government budget activity.

F. Small Changes: Differential Balanced Budget Incidence1 4

Define a Third State occuring after a new government specific expenditure (Qc)is financed by an equivalent increase in a tax (TC). To make things less complicated,define remaining tax incidence and benefits from other government financed out-puts as unchanged and assume that there is no effect on techniques. As a conse-quence, we generate two or more identities. The focus is total incidence, that is onthe change in income in State 3 compared to State 2. For household j, this is st4atedas in (6).

m m(6) A A Rh A 2k + R Rk AA - ATc + B3 Q3c

Or focusing on the spending side in the Third State gives

304 JACOB MEERMAN

(7 Ai n n Q.B Q3(7) Y A Pi Q2i + XP Q3 B c

Concrete illustration is useful. Suppose the expenditure newly financed (QC) ishospital care for the poor. The short term effects may be:(a) ARi, an increase in the wages of medical personnel and in the returns to other

hospital factors, plus other possible changes in R.(b) AA i, an increase in employed medical personnel, plus other changes in employ-

ed assets.(c) AT., an increase in tax c equal in the aggregate to the total costs of Q3c'(d) Xc 3c' consumption of "free" hospital care (hitherto Qc, publicly financed,

was non-existent.)(e) APi, an increase? in price of medical outputs, plus other changes in relative

prices.

(f) AQi, an increase in the quantity of medical outputs (in addition to Q3c) plusother changes in outputs.

Obviously, not all effects apply to all households.Although we assume an unchanged average price level, relative prices have

changed. As a consequence for household j, although (6) and (7) measure the changein income at a constant price level, we cannot be certain that even where Yj ispositive, that total utility has increased. The change in relative prices may make itpossible for one household to attain a higher level of utility per dollar of privatespending; while in spite of substitution another household perforce moves to a lowerindifference curve for a given private budget constraint. Hence, unambiguous com-parison even in the case of differential balanced budget incidence is impossible.

Many will argue that for small government programs, changes in factor endow-ments and relative prices are likely to be small, so that calculation of the change inmoney income is a good estimate of the change in total utility. I would take thesame position. Nevertheless, if one wants to be rigorous, then the earlier conclusionholds.

Note, however, that conceptually it is possible to measure not only total incid-ence AY1, but also to break out our three components. In identities (6) and (7), AY-is total incidence of the new tax/expenditure. The final term in both identities isbenefit incidence. In identity (6), the third term on the right side is tax incidence.Finally, RPTO is most simply defined as a residual equal to AY 1 - (IpcQ3c-ATc).

G. Tax and Benefit Incidence ComparedUsing our definitions, the first and second terms on the right side of equation

(6) encompass RPTO incidence. In other words, they measure the combined RPTO

DO EMPIRICAL STUtDIES OF BUI)GET JNCII)ENCE MAKE SENSE? 305

effects of both the increased tax (ATc) and the new publicly financed expenditure(Qd) In our example, we cannot separate out the RPTO effects of the tax fromthose of the expenditure. But one can devise examples in which the sole change is acompensating change in either expenditure or tax; e.g., replacing one tax with anequally-yielding alternative. Such cases would permit calculation of either the RPTOincidence of the change in taxes or in expenditure separately.

In identity (6) the change in taxation affects relative prices of outputs andreturns to factors and may have long-run effects on asset formation. None of thesechanges is assumed in studies of total tax incidence. See for example the recentstudy of Pechman and Okner 1181 or Gillespie [6].15 Such tax studies measuresolely the analogue of benefit incidence with respect to taxes. Yet this unmeasuredRPTO incidence on the tax side is important. Obviously, effects of decreased and/orre-channeled private spending on kelative prices, factor endowments, and even tech-nological change are significant. The failure to deal with these in tax studiesprobably means errors as grave as those on the spending side where similar effectsare ignored. 1 6

An implication of this reasoning is that, conceptually at least, we can also de-compose total incidence into two symmetrical pieces of two parts each: Tax inci(dence per se (defined as equal to the total tax) and the resulting RPTO incidence.Benefit incidence per se and the RPTO incidence resulting from the expenditure. Acertain conceptual symmetry results. We define tax and benefit incidences as reduc-ing and increasing incomes directly. Their RPTO incidences are more veiled, operat-ing through changes in relative prices in factor and product markets, as well asthrough changing quantity and composition of output.

One additional conclusion is suggested by the above material. An attempt toassess even the differential incidence of an existing program is apt to be extremelycomplex, demanding research inputs which may not be at all commensurate with theexpected research benefits.

Empirical country studies of tax incidence although frequently criticized be-cause they ignore RPTO effects, are nevertheless widely accepted. Similar studies ofexpenditures, however, have been more seriously criticized, and have less widespreadacceptance than those of taxation. 17 Yet, one conclusion suggested by the sym-metry of incidence with respect to taxes and expenditures is that if it makes sense tostudy tax incidence ignoring RPTO effects, as is commonly the case, it should makeequal sense to study benefit incidence in the same fashion, as is equally commonlythe case. It is possible that the greater resistence to the expenditure studies may bedue to confusion of I¢PTO and benefit incidences. Just as it is possible to separatetax-incidence per se from the associated RPTO effects, so it is possible - in fact,thus far, inevitable - that study of benefit incidence oceurs independently of RPTOanalysis with respect to expenditure.

306 JACOB MEERMAN

H. RPTO Incidence in the Short RunOur earlier conclusion was that the empirical country studies of budget in-

cidence were seriously invalid if they claimed to compare the effects of budgetactivity between a state of the world with government and a state of the worldwithout it. (Just what is claimed to be compared in sueh studies is frequently notclear.) Can we make an alternative and less defective interpretation? The answer isyes if we move from a general equilibrium to a partial equilibrium framework.

Most of the concern with incidence questions by both citizenry and decision-makers focuses on the effects of budget activity on the returns to the existingdistribution of factors of production. The vector of interest is therefore (R, P). Andthe relevant qLuesti(rn is, to what degree does government budget activity so radicallyaffect relative factor and product prices that for a given household endowment offactors, household utility corresponding to pre-tax factor returns in the two periodsdiffer substantially.

This question is unstudied. Presumably, it is an important component of incid-ence in many situations. For example, when government is the major employer ofcertain groups, it may also increase the returns to those groups substantially abovewhat they would be if government did not exist. Additional examples come to mind.Clearly of importanee is the size of government relative to the economy, and ofgovernment demandl for highly differentiated labor, or capital. By the same token,the incidence due to changes in relative prices would decline as an economy in-creases the diversity of production, as labor and capital increase their mobility, asproduction functions become increasingly variable oIn both the input and outputsides. Whiat these considerations suggest is that such incidence is apt to be higher in ahighly dual, developing country with a large government than in an advanced eco-nomy.

Returning now to our question, the empirical country studies of budgetincidence can be most usefully viewed as "short-run backwards", in which factors ofproduction and technology are constant. Their error lies in the assumption that Rand P are independent of budget activity. We do not know when this error is seriousand when it is trivial.

L Measuring Benefit IncidenceAnother major source of discontept with country studies of benefit incidence is

the assumption that total costs equal total value. As concerns public goods thisfrequently is interpreted as follows:

(8) BRG = a 7T('9I J

DO EMPIRICAL STUDIES OF BUDGET INCII)EN 1l MAKE SENSF"? 307

where: T.g = total costs of producing Ga1 = proportion of total income received by household j..

This of course gives the same relative results as the assumption that no one benefitsdirectly from general goods - an assumption of some attraction given the characterof most such expenditures as general overhead costs.1 8 These allocation techniqueshave no scientific basis in fact. They can be described as pseudo quantification ofphilosophical positions or a priori assumptions as to the arguments in utility func-tions.1 9 Yet the problems of allocating benefits from such goods may not be insur-mountable. For example, if one's interest is improving the long-term welfare ofdisadvantaged groups, it makes sense to disregard such expenditures since it is un-likely that even substantial changes in their magnitude and form will have muchimpact on absolute welfare of the poor.2 0

More problematic is the frequent assumption in such studies of allocative effic-iency for specific expenditure. In other words, mode of production and allocation ofpublic outputs is such that for each household, assumed value received for anyoutput equals corresponding cost of production. In terms of the model, this meansfor household j that

(9) B; =-b

where = average and marginal cost of publicly financing i.It is obvious that the political process does not give such fine-tuned results. It is alsousually impossible to empirically measure Bfi: Even if households wanted to, itprobably would be virtually impossible for them to consistently value - i.e., decidewhat they would be willing to pay for - a school year, or park-sojourn, or clinicvisit. Moreover to the extent that expenditures redistribute from wealthy to say apoor family, it may be that B2. < C2i because of the meager level of income of thepoor relative to the magnitude of government programs. Nevertheless, this lattersituation may be compatible with a net increase in utility for the community as awhole, to the extent that redistribution is income equalizing and marginal utility is adeclining function of income. Under such circumstances, the likely associated pro-gressive tax could reduce utility less than the spending increases it.2 1

Such thinking leads some to assert that aggregate government spending onaverage and on the margin produces at least as m uch "welfare" or utility as theprivate spending for which it substitutes. In other words, the utility of the publiclyprovided Qi to the recipient household is seen as equal on average to the utility ofthe taxes to the household which pays for it. One conclusion from this line ofreasoning is that charging the public costs of providing a benefit to a recipient, maygive better results in terms of the underlying distribution of welfare across the

308 JACOB MEERMAN

community than attempting to estimate the value to the recipient, as thus farassumed in the model.

But one can make a third and still weaker assumption:

(10) ub = f(Qd) f'.' °

where =b total utility to the beneficiary of publicly financed good Qi.

Since average costs per unit of output are fairly stable, it follows that utility andtotal costs are positively associated. If this is true, obviously, it continues funda-mental to know who beiiofits, and to identify beneficiaries by the amount spent ontheir behalf. This leads to the conclusion that study of benefit incidence should beredefined. It should not attempt the impossible, namely to estimate the value of allbenefits to recipients, but rather to estimate the distribution of publicly financedoutputs and corresponding public costs by beneficiary. In other words, the aimshould not be to measure the value of benefits received, but to min -sure the distribu-tion of the costs to the community of providing those benefits.

Incidentally, this interpretation of "benefit incidence" although frequentlyeclipsed is not new. In 1941, Stauffacher's concern with the discrepancy betweenpublic costs and subjective evaluation of benefits led to an outcomne similar to thatsuggested above. In his words: "The benefit approach is ... primarily concerned with... charging the cost of certain services provided by the government to the group(s)which they are intended to benefit". 2'

Because of the way this paper is organized, the importance of measuring benefitincidence of existing programs - independently of the question of budget incidenceor tax incidence - may not have been brought to the fore. More concretely, if thepoor are to escape poverty through public expenditure then, measuring benefitincidence becomes a basic policy input: Intelligent activity to assist the poor te-quires knowledge of how well existing programs are functioning in terms of whichreach the poor and which do not. Information concerning the distribution of publiccosts by beneficiary is a necessary first step in acquiring such knowledge. Conse-quently, whether or not we compare the actual with an ex ante hypothetical dis-tribution, knowledge concerning benefit incidence is valuable per se.

Throughout this discussion we have assumed, unrealistically, either completelyprivate or public goods. Usually it is argued that many publicly financed specificgoods have positive externalities not capturable by the private producer. Hence,public finance is needed to increase output to the optimal level. Introduction ofsuch considerations into the above argument would result in reinforcing the con-clusion that the assumption of costs of public production highly associated withbenefits - which now distribute more diffusely - is usually valid. But such introduc-tion would also suggest the difficulties in treating even specific expenditures as

DO EMPIRICAL STUDIES OF BUDGET INCIDENCE MAKE SENSE? 309

completely private goods. Yet, at the same time, it would reinforce the desirabilityof estimating on whose behalf and in what magnitude public spending proceeds.More specifically, if utility functions are interdependent, so that on average house-hold k derives utility from the publicly financed "merit goods" consumed byhousehold j, it is of value to be able to state clearly what the cost to the communityis of providing the goods toj. Such information is a necessary condition 'or rationalplanning of public expenditures.

III. MAIN POINTS

(1) Estimating budget incidence in terms of an economy with government andthat same economy without government or minimum government is impossible.

(2) Scientific measurement of differential balanced budget incidence is im-possible. Even approximate measurement is extremely difficult if RPTO incidence issubstantial.

(3) Although it is impossible to measure benefit incidence, in general, benefitincidence and community costs will be closely associated. Consequently, we takecosts, which we can measure, as a proxy for benefits. This procedure brings a usefulby-product, identification of the costs incurred by the community in providingbenefits to different groups. This is a necessary first step in many types of policyanalysis.

(4) It is possible and worthwhile to study both tax and benefit incidence inde-pendently of RPTOincidence.

(5) Study of RPTO incidence per se is desirable. It would be useful to gauge itsimportance relative to that of benefit incidence.

(6) Country studies of general tax and/or public spending incidence would beimproved if they discussed explicitly two necessary limitations: (i) the usual assump-tion that relative prices, asset endowments, and production functions are independ-ent of budget activity; and (ii) their usual failure to consider RPTO incidence. Itmay be best to regard such studies as being of the short run "backwards" with assetendowments, production functions, and relative prices fixed.

(7) Thinking about the incidence of both taxes and public expenditures wouldbe improved if the symmetry of the two were recognized: tax incidence, in con-ventional definition is analogous to benefit incidence; RPTO incidence with respectto taxes (largely ignored in country studies) is analogous to RPTO incidence withrespect to expenditure (largely ignored in country studies).

(8) Because marginal utility declines with income, in many situations chargingcosts to recipients may be a better measure of the overall impact of budget activityon welfare than attempts to measure benefits in value terms: Even if a dollar of tax

310 JACOB MEERMAN

produces a quarter of benefits, total welfare may be increased if the tax-payer iswealthy and the beneficiary poor.

NOTES

* In this paper I speak solely for myself, and not for my employer.** The author is an employee of the World Bank, Washington, I).C.1 This is still a very narrow perspective. Government affects income distribution fundamentally

by supporting basic institutions and attendant legislation, e.g., a constitution, and less fundament-ally by affecting numerous variables which in turn affect income distribution; e.g., foreign ex-change rates, interest rates, treatment of business, wage and migration policy, and rate of inflation.Budget incidence, therefore, can only measure a small part of government imnpact on the distribu-tion of income.

2 For a review and critique of country studies of budget incidence see Bird and de Wulf [29,McLur+? [11], Meerman [12], de Wulf [25]. An early critique of United Kingdom measures ofbudget incidence was led by Peacock and Shannon [17]. Alan Prest [201 continued this attack onthe IJ.K. measures in part by stressing the contradictory nature of the "traditional" assumptionsconcerning tax incidence as also discussed in his 1955 paper. The frustrating question of theincidence of public goods has been discussed at length in Aaron and McGuire [1] and Brennan'scomment plus rejoinders [-l ].

3 The focus is solely effects from taxing and spending. Concomitant changes in the institution-al framework and their impact on distribution are not considered, e.g., licensing requirements.

4 1 , p.7]. McLure states that the concepts and terms expenditure and benefit incidence arefrom Musgrave f 14, pp. 213--15]. McLure is correct in this, although it is interesting that nowhereelse (including the index) does Musgrave use such language, nor does he make a distinctionbetween expenditure and benefit incidence as such in hlis 1973 text. Rather he defines expenditureincidence as encompassing both benefit and MIcClurian expenditure incidence. See Musgrave andMusgrave [ 1i, pp. 360-61 ].

6 Slavery presents a real world example of this assumption.

6 A world without government is an idealization. More realistically, define it as minimalgovernment spending compatible with maintenance of sufficient order and security to permitmodern economic life. Depending on circumstances, this would probably lie between 1 and 3percent of GNP for most countries. This is close enough to zero to permit such idealizationwithout seriously biasing analytical results.

7 The restriction to private consumption or investment goods directly to households is unreali-stic. It is eliminated below where the methodology is developed for considering subsidies in theform of production inputs. Public good is defined as a good "which all enjoy in common in thesense that each individual's consumption of such good leads to no subtraction from any otherindividual's consumption of that good" [21, Samuelson, p. 387].

8 [11, McLure, p. 52]. See also Maital [p. 562], Aaron and McGuire [1, p. 909] all of whomaccept this approach.

F For two good recent examples, see Gillespie [71 and Musgrave and Musgrave [15, pp.365-77 3

10 Equation (3) could have been written in national accounts terms; assuming that total finaloutput equals total factor-incomes is consistent with a Lindahl equilibrium.

11 Pareto superiority exists when in moving from equilibrium State 1 to equilibrium State 2,no household's total utility diminishes and at least one increases.

DO EMPIRICAL STUDIES OF BUDGET INCIDENCE MAKE SENSE? 311

12 In other perspective, we have here a version of the Index Number Problem in which solelyrelative prices change. Use of a Laspeyre index defines the maximum amount "real" income couldhave fallen for any household in State 2 relative to State 1. Note that the comparison is betweenan actual and a hypothetical given income distribution. Whether State 2 could be Pareto superior,given the range of possible income distributions in State 2 is another issue. But even an actual orpossible Pareto superior sitation for State 2 does not mean that State 2 is inambiguously better. Itmay have been possible to achieve an income distribution in State 1 preferred by all to the actualor possible distribution in State 2. See Scitovsky [22]. The debate between Aaron and McGuire [1]and Brennan [4] revolved around whether, on introducing public goods, a Lindahl solution is"distributionally rieutral". The outcome came close to the position presented here, namely thenecessary ambiguity in comparing incomes in two states with different relative prices, even thoughincomes in terms of the standard numeraire remain constant.

13 In terrrms of the model only a subset of the n possible Qi will be produced in any givenstate.

14 Defined as the combined incidence of a tax atid the expenditure it finances.

15 The Pechman/Okner study is also of interest in avoiding many of the problems discussedabove. They did this by eschewing comparison of the status quo with a hypothetical zero govern-ment but used as their "counterfactual" a hypothetical proportional income tax equal to totaltaxes.

16 Although RP'PO effects are always neglected in the empirical country studies, in a widevariety of other wcrk they are the focus of attention. See Break [3]. But the results of such workhave not been of a kind to permit their application in the country studies.

17 See the references to the general critiques in note 2.18 For an analyis of some of the problems in allocating benefits from general goods see

Shoup [23, pp. 66ff] Aaron and McGuire [:1], Brennan [4], Meerman [13].

19 See Aaron and McGuire [1i.20 Another alternative is to regard the! services from outlays for internal and external security

as a peculiar general intermediate good whose benefits cannot be aliocated by households. SeeKuznets [9, p. 156] where he argues that such goods are "the cost of membershiip in our businesscivilization".

21 As illustrated by the case in which the poor family prefers a block grant equal to the costof the government service received, e.g., suxbsidized housing. Such subsidized housing need not beinferior to the ex-ante situation. To give a new twist to the Kaldor compensation criterion, in theabove example if the poor could transfer utility, they might be more than willing to fullycompensate the loss of utility to those who finance the housing subsidy.

22 See also the discussion in Cartter [5, pp. 11-13], Musgrave et al [16, pp. 282-84] and thecryptic remark of Gillespie [6, p. 176].

REFERENCES

[1] Aaron, Henry, and McGuire, Martin, "Public Goods and Income Distribution",Econometrica,Vol. 38 (1970), pp. 907-2Q.

[2] Bird, Richard M. and de Wulf, Luc H., "Taxation and Income Distribution in LatinAmerica, A Critical Review of Empirical Studies", International Monetary Fund StaffPapers, Vol.20, No. 4/1973, pp. 639-82.

[3] Break, George F., "The Incidence and Economic Effects of Ta -ation", in The Economicsof Public Finance (Washington, D.C., The Brookings Institution, 1974), pp. 119-240.

312 JACOB MEERMAN

[41 Brennan, Geoffrey, "The Distributional Implications of Public Goods", Econometrica,Vol. 44, No. 2 (March 1976).

[5] Cartter, AUlan M., The Redistribution of Income in Postwar Britain (New Haven, YaleUniversity Press, 1955).

[6] Gillespie, W.I., "Effects of Public Expenditure on the Distribution of Income", inMusgrave, Richard A , ed., Essays in Fiscal Federalisrn (Washington, D.C., Brookings Institution,1965).

[71 Gillespie, W.I., "On the Redistribution of Income in Canada", C'anadian Tax Journal, Vol.X.IV, No. 4/ 1976, pp. 419--50.

[8] Hotelling, Harold, "The General Welfare in Relation to Problems of Taxation and ofRailway and UJtility Rates", Econometrica, Vol. 6 (July 1938), pp. 242-69.

[9] Kuznets, Simon, "National Income, a New Version", Review of Economics and Statistics,Vol. 30 (1948), pp. 151 79.

[10] McLure, Charles, E., Jr., "'rhe Theory of Expenditure Incidence". Finanzarchi', Vol.30,No. 3/1972, pp. 432-53.

[11] McLure, Charles, E., Jr., "On the Theory and Methodology of Estimating Benefit andExpenditure Incidence", Paper presented at the Workshop on Income Distribution and its Role inDevelopment (Houston, Texas, William Marsh Rice University, 1974).

[121 Meerman, Jacob, "Fiscal Incidence in Empirical Studies in Income Distribution in PoorCountries", U.S. Agency for International Development, Discussion Paper No. 25 (Washington,June 1972).

[131 Meerman, Jacob, "Income Redistribution in West Malaysia and Recent Empirical Workon Budget Incidence: A Comment", Public Finance/Finances Publiques, Vol. XXX, No. 1 1975,pp. 131-35.

[141 Musgrave, Richard A., The Theory of Public Finance (New York, McGraw HIill, 1959).[1i] Musgrave, Richard A., and Musgrave, Peggy B., Public Finance in Theory and Practice

(New York, McGraw Hill, 1973).[161 Musgrave, Richard A,, et al., "The Distribution of Fiscal Burdens and Benefits", Public

Finance Quarterly, Vol. 2, No. 3 (July 1974).[17] Peacock, Alan and Shannon, Robin, "The Welfare State and the Redistribution of

Income", Westminster Bank Review Quarterly (August 1968).[18] Pechman, Joseph A. and Okner, Benjamin A., Who Bears the Tax Burden? (Washington,

D.C., Brookings Institution, 1974).[19] Prest, Alan R., "Statistical Calculations of Tax Burdens", Economica, Vol. XXII, No. 87

(August 1955).[20] Prest, Alan R., "The Budget and Interpersonal Distribution", Public Finance/Finances

Publiques, Vol. XXIII, No. 1-2/1968, pp. 80-98.[211 Samuelson, Paul, "The Pure Theory of Public Expenditures", Review of Economics and

Staitistics, Vol. 36 (1954), pp. 387-89.[22] Scitovsky, Tibor, "A Note on Welfare Propositions in Economics", The Review of

Economic Studies, Vol. 9 (1941), pp. 77-88.[23] Shoup, (Carl S., Public Finance (Chicago, Aldine Publishing Company, 1969).[241 Stauffacher, Charles, "The Effect of Governmental Expenditures and Tax Withdrawals

upon Income Distribution, 1930-1939", Public Policy, No. 2 (1941), pp. 232-61.[25] de Wulf, Luc H., "Fiscal Incidence Studies in Developing Countries: Survey and

Critique", International Monetary Fund StaffPapers, Vol. 22, No.; i 975.

Summary: Do Emnpirical Studies of Budget Incidence Make Sense? - This paper attempts to reduce thc controversyconcerning statistical studies of the incidencc of taxes and public expenditurc. The analysis is based on an algebraicformulation of the household budget constraint permitting comparisons under various states of the world. Incidence isso defined as to include general cquilibrium effects of taxes and expenditures, namely changes in relative prices,technological change, and the outputs of various goods and services, (RPTO incidence). The paper shows that the

DO EMPIRICAL STUDIES OF BUDGET INCIDENCE MAKE SENSE? 313

empirical country studies wrongly assume that RPTO incidence is zero, Economists have been more willing to ignoreRPTO incidence in aggregate tax studies than in the analysis of incidence of public expenditures notwithstanding thesymmetry between the two: The statistical studies exclude RPTO considerations, in analyzing both tax and benefitincidence but should include them in both cases. In fact, such studies are really "short run backwards-" in that RPTOincidence is implicitly assumed constant in the backwards comparison of the actual distribution of tax burdens andexpenditure benefits among households with the hypothetical situation of zero government. The paper also shows that aLindahl solution to production of public goods although Pareto optimal may not be Pareto superior. The paper alsoconsiders the assumptions used to distribute benefits of public expenditures across households. 'The studies inevitablyassume that the recipient's valuation of such benefits equals their public costs. This is improbable, for example, wherethe cost is high and the recipient's income is low as may be the case with poor children in secondary schools. However wecan make a virtue of necessity. The researcher should drop all pretense of measuring benefits and allocate costs tobeneficiaries. This will be useful to policy makers who need to know where the spending goes. It may also give a bettermeasure of the overall impact of budget activity on welfare than attempts to measure the value of benefits.

R6sumr: Les etudes ernpiriques de l'incidence budge'iaire ont-elles uin sens? - Cet article tente d'apaiser la controverse surles 6tudes statistiques relatives A l'incidence des impots et des depenses publiques. L'analyse est fondee sur uneformulation alg6brique de la contrainte relative a la gestion budg6taire permettant des comparaisons en diff6rents paysdans le monde. L'incidence est d6finie de facon A inclure les effets d' 6quilibre g6neral des impots et des depenses, Asavoir modifications dans les prix relatifs, evolution technologique et productions de divers biens et services (incidenceRPTO). L'article montre que Ies recherches empiriques par pays posent A tort comine hypothese que l'incidence RPTOest nulle. Les economistes ont eu davantage tendance a ignorer l'incidence RPTO dans les recherches en matiered'imp6ts que dans l'analyse de l'incidence des depenses publiques, malgre la sym6trie entre les deux: Ies etudesstatistiques negligent le RPTO en analysant l'incidcnce A la fois des impots et des avantages, alors qu'elles devraient lesinclure dans Ies deux cas. En fait, de telles etudes sont r6ellement "en arriere A court terme", en ce sens que l'incidcnceRPTO est implicitement supposee constante dans la comparaison "enarriere" de la distribution reelle des chargesfiscales et des avantages des d6penses entre les m6nages, avecunesituation hypothetiqued'un gouvernement inexistant.L'etude montre aussi qu'une solution de Lindahl potur la production de biens publics, bien que respectant l'optimumparetien, n'est pas superieure A celle de Pareto. L'article traite egalement les hypotheses relatives A la distribution desavantages des d6penses publiques entre les menages. Les 6tudes posent inevitablement en hypothlsse que la valeur de telsavantages pour qui les recoit equivaut A leurs couts publics. Or, ceci est improbable, par exemple quand le cofit est 6leveet que le revenu du beneficiaire est bas, comme c'est le cas pour les enfants pauvres dans les ecoles secondaires.Cependant, on peut faire d'une necessite une vertu. Le chercheur devrait renoncer A la pretention de mesurer Itsavantages et de r6partir les cofits entre les ben6ficiaires. Ce serait pourtant utile pour les d6cideurs politiques qui ontbesoin de savoir la destination des depenses. Cela peut aussi donner une meilleure evaluation de l'effet global del'activite budgetaire sur le bien-etre que les tentatives faites pour mesurer la valeur des avantages.

Zusammenfassung: Sind empirisehe Studien fiber Budgelinzidenz sinnvoll? - Dieser Artikel versucht, die Kontroverse umstatistische Untersuchungen der Inzidenz von Steuern und dffentlichen Ausgaben abzubauen. Die Analyse basiert aufeiner algebraischen Formulierung der Budgetnebenbedingung, die Vergleiche zwischen verschiedenien Staaten der WeltzulaBt. Die Inzidenz ist so definiert, daB sie allgemeine Gleichgewichtseffekte von Steuern und Ausgaben, insbesondereXnderungen der relativen Preise, technologische Veranderungen und die Produktion verschiedencr Gilter und Dienst-leistungen (RPTO-Inzidenz) einschlieBt. Der Artikel zeigt, daB die empirischen Lainderstudien falschlicherweise eineRPTO-Inzidenz von Null annehmen. Wirtschaftswissenschaftler sind eher bereit, die RPTO-lnzidenz in Steucrunter-suchungen unberilcksichtigt zu lassen als bei Analysen der Inzidenz offentlicher Ausgaben ungeachtet der Symmetriezwischen den beiden: Statistische Untersuchungen schlieBen RPTO bei der Analyse der Steuer- und Nutzeninzidenzaus, sollten sic jedoch in beiden Fallen beruicksichtigen. Tatsachlich sind solche Untersuchungen wirklich "kurzfristigruickwarts gewandt", insofern als die RPTO-Inzidenz implizit als konstant in dem "ruckwarts" orientierten Vergleichder gegenwartigen Distribution von Steuerlasten und Ausgabeinutzen zwischen Haushalten unter der hypothetischenSituation einer Null-Regierung. Der Artikel zeigt ebenfalls, daB eine Lindahl-L,sung fur die Produktion offentlicherGilter, auch wenn sie Pareto-optimal ist, nicht unbedingt der Paretos-Ldsung tiberlegen ist.Darfl6crhinaus werden dieAnnahmen untersucht, unter denen die Nutzen dffentlicher Ausgaben auf die Haushalte verteilt werden. Die Unter-suchungen gehen zwangslaufig davon aus, daB die Bewertung des Empfdngers solcher Nutzen mit den diffentlichenKosten gleichzusetzen ist. Dies ist z.B. dort unwahrscheinlich, wo die Kosten hoch und die Empftingereinkommen nied-rig sind, wie es bei armen Kindern in hoheren Schulen der Fall sein kann. Aus der Notwendigkeit laBt sich jedoch cineTugend machen. Der Wissenschaftler sollte auf eine Messung der Nutzen verzichten und die Kosten unter den Bc-gtinstigten aufteilen. Dies ware fulr die Politiker von Nutzen, die uiber die Verwendung der Gelder informiert seinmussen. Dies lieBe auch eine bessere Bewertung der Gesamtwirkung budgetiirer MaBnahmen auf die Wohlfahrt zu alsVersuche, Nutzen zu bewerten.

ESTIMATING COUNTERFACTUAL INCOMESIN STUDIES OF BUDGET INCIDENCE

by

JACOB MEERMAN AND PARTHASARATHI SHOME*

I. INTRODUCTION

In a recent issue of this journal Meerman asks, "Do Empirical Studies of BudgetIncidence Make Sense?" He answers that although sensible, their usefulness is muchconstrained because they are short run "backwards" in that they implicitly assumeceteris paribus conditions for relative prices, technology, labor supply, level ofunemployment, savings, and investment in making the incidence estimates [9, p.298]. Meerman sums these together under the heading of RPTO incidence (relativeprices, technology, output).

Perhaps the most serious and most vitiating of these ceterisparibus conditions isthat for relative prices.' In addition, the failure of these studies to, in any way,indicate the effects of government budget activity on savings and investment is alsoa very serious limitation. It is noteworthy that these problems are to some degreedealt with in the Harberger Model [7] and in its development. 2 In what follows wewish first to show how results from the Harberger Model can be used in conjunctionwith the Meerman approach to suggest some of the error in the comprehensivestatistical studies because of assuming ficed relative prices. In other words,Meerman somewhat overstates his case. To some degree relative price changes forfactors of production - principally capital and labor - can be taken intoconsideration in the comprehensive statistical studies. Second, by correcting andelaborating on Meerman's model, we wish to point out two additional errors in thecomprehensive statistical studies because of the fact that if taxes are in part (or intotal) indirect, total factor incomes (or in Meerman's model the household budgetconstraints) necessarily sum to less than total output.

II. IMPLICATIONS OF THE HARBERGER MODEL

In Section 11(B) of his paper, Meerman points out that in the comprehensivePublic Finance / Finances Publiques No. 2i'1980 Vol. XXXV/XXXVbme Ann6e

292 JACOB MEERMAN AND PARTHASARATHI SHOME

statistical studies of incidence, a "counterfactual" or a hypothetical, before-taxposition (state 1) is defined and then compared with the observed post-tax position(state 2), to obtain a measure of incidence. He goes on to show how RPTO incidenceis implicitly assumed to be zero in these short run studies and how thecounterfactual household income, Yl, is simply taken to be equal to the observed

mvalue of the total factor incomes, X R2k A2k, where Rk is the per unit return to

k=1

the k factor, and Ak is its quantity.Ir. his orginal article Harberger used as the counterfactual - Meerman's state I

- a perfectly competive US economy with Cobb-Douglas production funtions,fixed technology, fixed supply of factors and no corporate income tax. He alsoassumed expenditure neutrality: Government expenditure from the tax was toexactly mimic the former expenditure of the private resources now taxed away.Harberger compared labor and capital incomes in counterfactual state 1 with whatthey were in actual state 2, principally by permitting relative factor prices to vary. Asufficiently long run analysis allowing for intersectoral factor movements to becompleted, together with the assumption of perfect competition, implies thatcapital earns the same net return in all uses. Consequently Harberger's state 2 had ahigher gross rate oi return per unit of capital in the more highly taxed corporatesector than in the non-corporate sector.

We illustrate Harberger's case by the following example using Meerman'snotation. Let us focus only on the returns (RK:) a.nd c uantities (AK) of capital in thecorpora e and non-corporate sectors. Then let RgKAgK be the net incoe of capitalin the 1 sectcr (i = corporate, C; and non-corporate, NC) for the g state (g =before-tax, 1; and after-tax, 2). Then

C C NC NC(1) RlKAIK +RI K Al K =RK AIK

After the tax is imposed and equilibrium has been achieved in state 2, the grossincome of total capital can be defined as

+S NC NC C C(2) R2K A2K +S=R2 K A K + R2 K A2 K + S

NC NC C Cwhere R2 K A2 K is the post-tax income of non-corporate capital, and (R2 K A2 K +S) is the gross income of corporate capital, of which S is the tax on corporatecapital. In the unsophisticated approach to corporate tax incidence it is assumedthat

(3) R1K AIK - R2K A2K = S

that is, the entire tax appears to be "borne" by corporate capital. But, as explained

ESTIMATING COtTNTERFACTUAL INCOMES 293

below in the simple Harbergerian example elaborated in Table 1, in reality the netper unit return to capiial in both sectors is depressed.

It is in the nature of Cobb-Douglas production functions that factor shares oftotal income remain constant at all input prices. In our example, constantcomposition and price of final output is implicitly assumed. Consequently - asillustrated in Table I - aggregate gross returns to non-corporate and corporatecapital remain constant at 400 and 200 respectively. In Table 1, the tax on incomefrom corporate capital is 25%. Since the gross return remains 400, the net returnmust fall to 300, with the proceeds from the tax equaling 100. We now have clearlyin mind what is needed to calculate the post-tax positions shown in Table i. (Table Irefers to capital income only since labor incomes in both sectors will remain thesame in pro-tax and post-taxcpositions.) C

In the post-tax situation, R2 K is taken as the net price of capital and A2 K is thequantity of capital in the corporate sector. Then net income in the corporate sectormust be equal to 300, or

(4) RX2K A2K = 300

Given that the totai amount of capital is 600, we can write (600 - Ag ) as theCamount of capital in the non-corporate sector. For equilibrium, R2 K has to be the

net price of capital in both sectors. Thus, since the income of capital in the non-corporate sector is 200, we have

C C(5) R2 K(600 - A2 K) = 200

C CSolving equations (4) and (5) we have R2 K = S/6 and A2 K = 360. The post-taxquantity of capital in the non-corporate sector is then (600-360) = 240. And the

CR2)K

gross return per unit of capital in the corporate sector ,where t is the tax rate

(25%7(). Thus it is equal to 1 10/9.

It is now obvious that in economies similar to that assumed here, in order to getto the counterfactual capital income, RIKAIK, that is (400+200), one should simplyadd the gross income of capital in the corporate sector (400) to the income of capitalin the non-corporate sector (200). Labor income in state 2 can be simply added onto thig since it if, not affected by the tax, and thus the total income forcounterfactual, RI Al, is obtained.3

rhis outcome has impficauthr'is for the comprehensive statistical studies (CSS).Consider two houLseholds in state 2 (from the world of Table 1). Household A hasincome from corporate capital only. Household B has income from non-corporate

294 JACOB MEERMAN AND PARTHASARATHI SHOME

TABLE I

Capital Income, Quantity And PriceIn Pre-Tax And Post-Tax Positions

State I (Before Tax)Income Quantity Price

RIK AIK = 400 AIK =400 AIK I

NC NC NC NCRIK AIK =200 AIK =200 RIK I

State 2 (After Tax)

(Assuming Cobb-Douglas Production Functions, and a 25% Tax on Corporate CapitalIncome, Equal to 100)

Income Quantity Price

R2K A2K +S=400 (gross) A2K = 360 R2K = 10/9 (gross)

R2K A2K = 300 (net) R2K = 5/6 (net)

NC NC NC NCR2KA2K = 200 A2K = 240 R2K 5/6

capital and no other source. Both households own 40 units of capital. They live inour simple Harbergerian Cobb-Douglas world, where legally "only" corporateincome is taxed and at 25 percent. We indicate their state I and state 2 incomes inTable 2 below.

';ABLE 2

Incomes of Two Households in State 1 and State 2

Source ofHousehold Capital Quantity Income

Gross Net State 2 State I

A Corporate capital 40 (10/9)(40) (5/6)(40) 33.33 40= 44.44 = 33.33

B Non-corporate capital 40 (5/6)(40) (5/6)(40) 33.33 40= 33.33 = 33.33

ESTIMATING COUNTERFACTI JAL INCOMES 295

The Harberger Cobb-Douglas approach makes the proper measure of income in

state 1 and state 2 obvious -under competitive assumptions. Yet in many of the

C(SS, the approach used appears to have been defective. Such studies frequently

answer the question, what would incomes be in state l, that is were there no

government, by simplv taking gross factor incomes as they are in state 2, even

though the studies assume that taxes on part of capital actually burden all capital in

part or in total. In terms of the example above, this means taking total income for

househiold A in state 1 as 44.44, since at first glance this is what factor income would

be were there noz government. Consequently, this approach exaggerates

cotuiiterfacttual inconme of corporate capital owners, if corporate taxes are above the

average of capital taxation.4 In addition relative tax burden of the corporate

capitalists is estimated as somewhat too low, counterfactual incomes are too low

for capitalist lhuseholds with low corporate income, while their estimated burdens

from capital taxes are too high.

Pechm21an and Okner [11] - we believe - avoid this error, although it is not

ct%mpltele,. clear that this is the case, since they do not provide detail on just how

they "restore" corporate incomiie taxes to households. Musgrave et al [10] may have

added the entire corporate income tax back to corporate capital shareholders in

building their coLnterfactual for a US study of budget incidence for 1968. To wit:

"...proper treatment of the cmrpor;it iol1 tax calls for imputation of total corporate

source income to Thareholder,- " (NMwgra%ve et al [ 10, p. 30 1 ]). If they did do this it

A, mid not be cosistent with their six tax shifting assumptions for the corporate

inc 'me tax, none of which assuimes that the tax burdens solely owners of corporate

shares. In general in most studies no attempt is made to adjust counterfactual

income to be consistent with incidence assumptions.5

We should also emphasize that some investigators are very much aware of the

effects of government tax and expenditures on relative prices.6 But it has hitherto

been widely believed that it is impossible to in any waY come to grips with the

problem. This is not completely the case, as suggested by the work on the Harberger

model as well as the discussion on indirect taxation below. We do have a beginning

in dealing with the more general issues of the impact of taxes on counterfactual

incomes. But the spending side is, alas, largely terra incognita.

IllII. 'II'LIFFECTS OF INDIRECT TAXFS

In Sectioni II(C) of his paper, Meerman makes an implicit and unwarranted

a-I- 'l.upt W-n to the effect that all taxes are direct. He states the Lindalhl solution,t his

equlati(on (3), a9 total ouitput equals total factor income, or

n I n I / in I

(6) Ei Q2fj +i 4 v J V ?t Q:r,l + X I*?j G2 (2 > R2ki A_kjI1 H- i -i jl k- jiI

296 JACOB MEERMAN AND PARTHASARATHI SHOME

where P,ij and Q27y are the price and quantity respectively, in state 2, of the X

commodity purchased by the 2 household, Byij is the marginal value to recipienthousehold j of the private commodity i, in state 2, provided through governmentprovision; B2g is the corresponding marginal value of the public good (G); G2 is the

m Iamount of the public good, and XV -' R2kj A2kj is total factor income.

k=l j=1

This equality will hold only if all taxes are direct. For if taxes are in part or in totalindirect, the left-hand-side of the equation (total output) must be greater than theright-hand-side (total factor income) by the amount of indirect taxes. Thus theabove statement of the Lindahl solution is valid if and only if all taxes are direct sothat there is no wedge between total output and total factor incomes. Of coursewhether or not indirect taxes exist, in no way affects Meerman's conclusion that aLindahl solution is not necessarily Pareto superior.

This failure to be explicit about the directness or indirectness of taxation suggestsanother aspect of the problem of what is the proper concept of total householdincome for state 1. As noted, if in state 2 all taxes are indirect, then, even ignoringRPTO incidence, for all households combined, the sum of factor incomes is lessthan total output oi NNP by the amount of indirect taxes, In other words if Y2i isthe income of the. household in state 2, A is the average indirect tax per privategood, and the variance in X is independent of the distribution of incomes, then if wehave a Lindahl solution or we take a national accounts perspective we have thefollowing for each household j:

m n(7) y2j = s R2k Ark + X( E Q2)

k=1 i=1

Under these circumstances to get a proper measure of Yl we need to impute -

somehow - to each household its change in income on the assumption of completeremission of indirect taxes. (The resources corresponding to the sum of all such

n Ichanges will necessarily be (X) Y2 I Q2y). Many researchers have ignored

i=1 j=1

this problem and taken as their counterfactual simple factor incomes in economieswith considerable indirect taxation. Even the more sophisticated researchers, whoput these resources into the indirect taxes would be distributed in state 1 inproportion to the distribution of factor incomes and that changes in relativeproduct prices are neutral (Pechman and Okner [ I 1]; Musgrave [10], and Browning[3]). In brief, the CSS ignore RPTO incidence, and they also fail to consider themost obvious aspect of RPTO incidence, namely how to adjust household incomes

D)O EMPIRICAL SIr'uIIES OF BUDGET INCID)ENCE MAKE SFNSE? 297

private incomes) and benefit incidence (who receives the benefits of governmentservices)" [11, Mclure, p.2 1. In McLure's thinking, total incidence can therefore bedecomposed into: "The burden (and benefits) of taxes used to finance public activ-ity, the benefits of public services, and the redistribution of income resulting fromchanges in relative factor rewards and product prices induced by the shift ofpurchasing power from the private to the public sector. For convenience, we canrefer to these three effects as tax, benefit and expenditure incidence,respectively":

Earlier versions of this paper attempt to use both this basic tripartite decompos-ition of incidence as well as the terms benefit and expenditure incidence. Althoughlogically there was no problem, the result was continuing confusion of readers not tomention incandiary red herrings. Three difficulties were basic:

(1) The distinction is made between expenditure incidence as consisting ofimpacts on private incomes - in Musgrave's language "changes in the distribution ofincome disposable for private use" 1 14, Musgrave, p. 2141 - and benefit incidence as"the benefits derived from public services". Clearly, however, the benefits frompublic services have private income equivalents. In a common definition of income,transfer payments, rent allowances, and public medical care received free or atsubsidy can be regarded as private income by their recipients. Musgrave and McLurereally define income rather rigorouslv as solely returns to factors of production andon the spending side solely the goods and services which those returns can purchase.

(2) It is very common, among economists, to use "expenditure incidence" torefer to benefit incidence alone or to benefit incidence plus McLure / iMusgravianexpenditure incidence. And it is difficult to fault people for using the terms in suchan apparently straightforward and common sense fashion. But it does lead to un-necessary confusion, and a great deal of explanation.

(3) Finally, because of the common definition of tax incidence as equal to theamount of the tax, the phrase "expenditure incidence" as used by Musgrave andMcLure may carry something of an implication that only an expenditure has effectson relative prices, on techniques, and on the volume of output. We realize, of course,that taxes have similar effects. More generally, we realize that by and large increasesin taxes and expenditures go hand in hand, i.e., over the long run budgets have to be(nearly) balanced. Consequently, it is probably useful to think of incremental taxesand their associated incremental expenditures together making explicit the fact thatthe resulting effects on tax/expenditure on relative product and factor prices as wellas the volume and composition of output are closely associated.

Where this leads us for the analysis at hand is to (i) developinent, of a model interms of balanced budget incidence, as the situation which most corresponds to thereal world; (ii) to keeping the McLure/Musgrave tripartite decomposition of in-cidence but to putting it into new bottles. Specifically, we define benefit incidence

298 JACOB MEERMAN AND PARTHASARATHI SHOME

REFFERENCES

[1] Andic, Fuat M., "Poverty and Tax Incidence in West Malaysia", Public Finance Quarterly, Vol. 5,No. 3/1977, pp. 329-370.

[2] Ballentine, J.G. and I. Eris, "On the General Equilibrium Analysis of Tax Incidence", Joturnal ofPolitical Economy, Vol. 83 (1975), pp. 633-644.

[3] Browning, E.K., "The Burden of Taxation", Journal ofPoliticalEconotn y, Vol. 86 (August 1978),pp. 649-67 1.

[4] Dodge, David A., "Impact of Tax, Transfer, and Expenditure Policies of Government of theDistribution of Personal Incorne in Canada", Reviewv of Income and Wealth, Series 21, No. I (Mvlarch1975), pp. 1-52.

[5] Franzen, Thomas, Lovgren, Kerstin and Rosenberg, Irma, "Redistributional Effects of Taxes andPublic Expenditures in Sweden", Swedish Journal of Economics, Vol. 77, No. 1/1975, pp. 31-55.

[6] Gillespie, Irwin W., "Effect of Public Expenditures on the Distribution of Income", in Essays inFiscal Federalism (Washington, D.C., Brookings Institution, 1965) pp. 122-186.

[7] Harberger, A.C., "The Incidence of the Corporation Income Tax", Journal of PoliticalEconomy,Vol. 70 (1962), pp. 215-240, reprinted in Taxation and Welfare (Boston, Little, Brown and Company,1974).

[8] McLure, C.E., Jr., and W.R. Thirsk, "A Simplified Exposition of the Harberger Model, II:Expenditure Incidence", National Tax Journal, Vol. 28 (1975), pp. 195-207.

[9] Meerman, J., "Do Empirical Studies of Budget Incidence Make Sense," Public Fiinance/FinancesPubliques, Vol. 33, No. 3/1978, pp. 295-312.

[10] Musgrave, R.A. et al., "The Distribution of Fiscal Burdens and Benefits", Public FinanceQuarterly, Vol. 2, No. 3/1974.

[11] Pechman, J. and B.A. Okner, Who Bears the Tax Burden (Washington, D.C., BrookingsInstitution, 1974).

[12] Ratti, R.A. and P. Shome, "The Incidence of the Corporation Income Tax: A Long Run, SpecificFactor Model", Southern Econonmic Journal, Vol. 44, No. 1/1977, pp. 85-98.

[13] Ratti, R.A. and P. Shome, "On the Separability of the Incidence and Efficiency Effects of theCorporation Income Tax", Public Finance/Finances Publiques, Vol. 32, No. 3/1977, pp. 348-353.

[14] Vandendorpe, A.L. and A.F. Friedlaender, "Differential Incidence in the Presence of InitialDistorting Taxes", Journal of Public Economics, Vol. 6 (1976), pp. 205-229.

Summary: Estimating C'ounterlactual Incomnes in Studies of Buidget Incidence. - In his article "DoEmpirical Studies of Budget Incidence Make Sense?" [9] Jacob Meerman argued that one of the mostserious shortcomings in such studies was the failure to in any way consider the effects of governmentactivity on relative factor and output prices. But Meerman overstated his case. This paper shows how theeffect of the corporate income tax on relative returns to factor incomes is dealt with in the HarbergerModel. In the process the paper shows how use of the Harberger Model could deal with some commonerrors in the empirical statistical studies. The paper also deals with a minor error in Meerman's article:His demonstration that a Lindahl-solution to taxes and public production is Pareto optimal but notnecessarily Pareto superior, implicitly involved the notion that all taxation is direct. It is obvious,however, that the conclusion also holds in economies with indirect taxation. The paper then takes ofifrom this point to discuss some rarely considered problems arising from the fact that in economies withindirect taxes, total factor incomes are less than total incomes. Many studies simply err in generating the

ESTIMATING COUNTERFACTUAL INCOMES 299

counterfactual incomes of a before-government world without including indirect taxes. Those few

studies that do incorporate the wedge of indirect taxes, hack into factor incomes, do so in an

unsatisfactory manner.

Resume: vstintlon Itic revvenu "counlerltnlual" tdanxs les etudcs d'incidence hudg(Itaire. - Dans son

article "D)o Empirical Stujdies of Budget Incidence Make Sense?"[91- .Jacob Meerman affirme que l'ulle

des dCficiences les pitis uinportantes de ces etudes est l'absence de prise en consid6ration des efiets de

l'activite gouvernementale stir les prix relatifs des facteurs et de l'output. Mais tMIcerman a exagere son

eas. Cet article montre comment l'effet de l'imp6t sur les societes sur les rendements relatifs de revenLus

des facteurs est traite dans le modele de Harberger. Au cours de cet article, nous montrons comment

l'utilisation du modile de Harherger pourrait eviter certaines erreurs communes dans les Ctudes

statistiquies empiriques. L'article present traite -gallment d'une erreur mineure dans l'article de

Meerman: sa demonstration qu'une solution du type l1indahl aux problemes des taxes et de la

production publique est optimale au sens de Pareto, mais pas necessairement sup6ricure au sens de

Pareto, fait appzl implicitement a la notion que toute taxation est directe. 11 est cependant evident que

cette conclusion vaut egalement pour des economies avec taxation indirecte. L'article present part

ensuite `_ cette conclusion pour discuter certains problemes, rarement consideres, reultant du fait quc

dans des economies avec taxes indirectes, les revenus des facteurs totaux sonit moindres que les revenus

totaux. De nombreises Etudes %'egarent en engendrant les revenus "counterfactual" d'uni monde

d'avant-gouvernenient sans inclure les imp6ts indirects. 1,es quclques etudes qui incorporent la part des

taxes indirectes dans les revenus de facteurs, le font de maniere peu satisfaisante.

Zusammenfassung: Eine Schdtzung der ,,c ounzerfactiual"-Einkonmiten in UIntersuchlungen uher die Budget-

inzidenz. - In seinem Artikel "Do Empirical Studies of Budget Incidence Mlake Sense?" [9] argumen-

tierte Jacob Meerman, dal3 einer der schwerwiegendsten M angel solcher I lintersuchungen darin bestehe.

dal3 die Effekte von StajtNaKtlis itinen auf relative EFaktor- und Outputpreise unberucksichtigt bleiben.

Aber Meerman tberbewertete seine Argumentation. Dieser Artikel zeigt, wie die Effekte einer Kdrper-

schaftschaftssteuer auf relative F`aktorertr5ge im Harberger-Modell abgehandelt werden. Im weiteren

Verlauf wird gezeigt, wie man bei Anwendung des Harherger-Modells mit einigen allgemeinen Fehlern

bei empirischen statistischen Untersuchungen fertig werden k6iinte. D)iese tUntersuchung befal3t sich

ebenfalls mit einem geringfdgigen I.ehler in Meermans Artikel: Scin Nachweis, dal3 eine Lindahl-Lbsung

fur Steuern und offentliche Produktion pareto-optimal. aber nicht unbedingt pareto-superior sei,

impliziert die Bedingung, daB jede Besteuerung direkt sei. D)iese SchluBfolgerung ist offenkundig in

Wirtschaften mit indirekter Besteuerunlg gtltig. Der Artikel nimmt dies zum AnlaB, einige kaum

beachtete Probleme zu diskutieren, die sich daraus ergeben, daB in Wirtschliften mit indirekten Steuern

die gesamten Faktoreinkommen geringer sind als die Gesamteinkommen. Viele Untersuchungen gehen

einfach fehl, wenn sie "counterfactual"-Einkommen in einer Volkswirtsclhaft ohne Staatssektor berech-

nen, ohne indirekte Steuiern einzubeziehen. In den wenigen IJntetsuchungen, in denen die Wirkungen

indirekter Sr uerur a uf Faktoreinkommen einbezogen werden, geschieht dies aufunbefriedigende Weise.

Confirmations and Contradictions

The Incidence of Sales and Excise Taxes, orWhere Do We Put the Transfers?Jacob Mteerman

W'orld Bank

Whether sales and excise taxes are regressive or progressive dependson howv the question is asked. If tax incidence is measured in thecontext of a comprechenisive study of both taxes and benefits, they arelikelv to be regressive. If onie uises differenitial incidence analysis inthe manner of Edagr Browninig (1978), they are likely to be progres-sive. As discussed below, of crucial iIpo)0Itance in such an analysisis the implicit or explicit (lefinitioni of what incomes would be werethere no taxes. So-calle(d counterfactual income is the crux of theproblem.

The measturemiienit of tax intcidenice involves an estimate of wvhlatincomes wvould be were there no taxes to compare against actualincomes, after taxes have been paid. Such estimationi inmmediatelyruns up againist the difficulty that taxes finance public spellnding wihichalso affects the distributioni of incomes in maniv ways. Various tech}-ni(lues have been developed to deal with this (liffictulty. One of thesehas beeni the neutralitv assumption, which in its earliest formY1 simplyprovi(le(I a wav to ignore public Npending by assuming that the gov-ernment spent in the same wav as those who paid the taxes woul(dhave spent (Harberger 19(32. p. 224). TI'he assumption has unilergoniea certaini developnicint, but it remains ssc jtiallv as it was in theb)eginning (Pcclmian:ui and Okner 1974, p). 29).

In cmitlast, sttu(lies of budget incidence wviich tteuniptl lo estimateb)oth tax illi(It l (c andi the (listribution of' I)enefits froM 1)lbli(spell(ting (lo lnot lliave this problem. In such stud(ies the attempt is, or

In this paper I speeak solelN for niNself and tiot fI' tm emplover, the World Batik, Iwiould( like to thank (Chatiles MI(oure tor helpful couinwinlts oTI a previous (draft.

[IJnman I ,/ P eli E wwi E .inmny. (8)), \ot 88, 11o. 6ifc 198') bv I he UvitetmtItv oI ( hi ago. 0022 iX8S8. 80 SHO0()0 I01301 '5t0

1242

CONFIRMATIONS ANI) CON I RAOI U(IONS 1243

shotild be, to estimate what household in(onies woull( l be w,ere thereneither taxes nor public spendinig. Such counterhfactual primlarv in-comes are theni reduce(d bv taxes and mcinrease(l l) benefits to arrive ata postgovernment distrihlltion.'

Aniothet solutioni to theS problem presented by public Spending is todefinie inicomies as they would l)e were government Npeiding uinl-

changed but financed tlhrough a pr polrtional incomle t.ax, in otlierwords, a comnprehlensive diflf'renltial incidence approa h Pechmanand Okner (1974, p. 29) used this .aprii)ach, which t1ey (des(ribedthus: "The (luestioni it [tile studiv] attemInlpts to anlswerl is-fHow dloesthe distribution of disposable ilc'OIllmS of houIseholds ulniler tilepresent tax systeiim differ from n , what it would1 be if' thie . . . taxes levpav were collecte(d tlhrouglh a pIr)portional income tax witlh thesamile yield?" lIhis proportiontal ui to iln tax approach in tax incidencestudies deals wvith the problem of the inc iid e it t of pul)lic sp)endling in asuperior way, because it assumes ceteris pa1ribtVs coldlitions for every-thing on the spendinig side rnd assumies that a "neutral" cmtniterf'ac-tual tax svs,seni is substituted for the actuial one.2

In using this approach, Pechimian and Okikner include in their in-come base certain cash trallsfers to houselholdis. Therefore, these arenecessarily iInclud(le( in the counterf actual whichll 1ev use, imp)licitly,as the basis for coIm1parinig household inicones. In brief, certain cashtransfers al-e inc'ludled as part of htltouchobld incotmles b)oth1 itl tilebefore-governnellnci-exist situationl (tlie counlterfactual) aiid theafter-taxes-are-paid situationi (actual). It is inlportant to niote that incomprehensive country statistical studies of budget incidence-whichestimate both tax and b)enefit incidence-such ani inc luoinii of certaincash transfers as part of counterfactual incotiie wouli not be correct.3

In such studies cash traiif'fers are a benefit and, therefore, are in-cluded solely as part of income after government benefits are addedto it. Pechman and Ok(ner are cutting ilito the inl('(ome streauls to taketheir measurement arlI develop a counterfactual at a poinit that gives

One of the major ad justmnts to actual inicomes is to increase them bv the amountof indirect taxes. Were governments eliminated, their 'bleedlig" of' the income streamthrough indirect taxes wuould cease, and real inconmes woul(d increase from the totalnational income to net national product.

2 Each tax is examined and contrasted with the equivalent proportional ineorne tax.The results are sumnmed to give the i, indent e of all taxes ( oml)ined. The counterfactualcomes closest to nieutralitv in the short runi in which factor sup)plies are inelastic. Buteven short-run neutralit% is not attained insofar as the demiiand(I for leisure is a functionof wages. Mtusgrave and others first used this tehniique in 1 95 1. a(or-dinig to Pechmanand Okner (W17.1, p. 24, n. 14).

3 This point is discussed below. Fol a review and critique of sto(h studies see M(IL.ure(1972), de Wulf (1975), and(i Meerman t1978). A hbodinlg commnon to such stu(lies, andof relevance to this paper, is that usuallk government expenditure-includingtransfers-redistributes f'ir more in favor of' the poor thani tax systems.

1244 JOURNAL, OF POI,ITICAI. ECONONIY

neither incomes as tthev would be were there no government norincomes as they are after all governnment effects are taken into con-sideration. But there is nothing wrong with this approach given theirbasic point of departure and their approach based on differentialincidence.

In a recent article in the Journial f' Political Ecornomn, Browning(1978) took the Pechnian and (kAter approach and( data to dc evelop ananallysis which reverses the convemtional wisdomi concerning the inci-dence of indlirect taxes. Browning was able to coniclude that excise andsales taxes are progressive tunder the very reasoniable assumption thattransfer payments are lii(lep)eleilt of such taxes,4 He reasoned that,since excise anid sales taxes force a wedIge bettveen finial outpUt pricesand factor costs, they reduce fartor payiientit by the amount of thewvedlge which becomes governmernt resouirces. Real transfer income ineffect escapes such taxes, with or without inflation.5 Since transferincome is a negative funictionr of factor incomes and accounts for asubstantial part of total incoomes, the outcoimle is highly progressiveexcise ancl sales taxes. This conclusioni is warranted as long as thediscussiorn is in the context of' (lif'feirential tax inicilence and incomesare as defined )by Pechnizrl and Okner (1974) in their studiy (antdBroNwning's). Undier such ternms Browninig's assertiotn that tlhe entireU.S. tax system is more progressive thlan Peclimiiii and(1 Okiier con-clucde can be cmnsidlered valil.

But Brownin,g's outcome is very depenldent otn the patterni of'gov-ernmenit spending. If transfer payments distribute in proportion towhat pretax income woulkl be, the conventional wisdom of regressiveexcise and sales taxes holcs even if transfers are included in theincome base. More generally, if we use the Browning approalch,the regressivitv or progressivity of indirect taxes dependls on hovthe governmenlt spends the resouirces it colletis.

'I'his conclusion is illustrated in table 1. Assumine a worl(d with onerich household an(l onie poor onie. In situation A, there are no taxesani(d NNP (nlet national product) e(iuals NI (national incorne). Insituation B, excise ano(l sales taxes re(clice factoI incomes by a niinth.(Thiere is nio inflatiorn.) TIli taxes finiance solely tranisfers wvhich distrib-ute in proportion to factor inicome. 'I'hus the in(cidence of sales taxesan( ex('ises is proportional. Sie-c thte inicome base is factor incomesplIs ui;aiisfcrs, the tax burdlen for1 all is 10 peIcent. Not until situationC, do we get the Browninrg otac mulc, in which the taxes are stron glvplrogressivc. In C( all the taxes collected arie tralnsferred to the p)oor.

4E.g., if simiar taxes were ir(ticatsed t N,crO i.ttl. say a value-a(dded tax im-plernented, Congress wotill ixi tease %elfille pna rnents to off.set their effects.

5 In cotntrast, Pechinan and(i (koier (1974) had tlie cash tranisfets bear a proportionalshare of thie burden ol texcise and sales i.lxes.

TABLE 1

COMPARATIVE TAX BURDENS USING BROWNING'S APPROACHIN A TWO-HOUSEHOLD WORLD

RICH HOUSEHOLD POOR HOUSEHOLDBROWNING'S

Factor Tax Factor Tax INCOMEn SITUATION Income Transfers Taxes Ratio* Income Transfers Taxes Ratios NNP NI AGGREGATE*

A 90 ... 18 ... ... 108 108 108B 80 10 10 .10 16 2 2 .10 108 96 120C 80 ... 10 .111 16 12 2 .067 108 96 120

* Income base is pregovernment faclor income plus transfers.

124(6 JOUTRNAI. OF POLITICAL ECONOMY

TABLE 2FIs(.At. INCIDENCE USING COMPREHENSIVE APPROACHF

IN A rwo-HOUUSF)LD WORLD

Income AggregateRich Poor NI or NNP

Income. pregoveronlmIlt M() 18 108 108Income af'ter tax paymenit 8( t .Tax burden 11F( I%(, ... ...Incotiie after receipt 80 28 96 108

of benefits (transfers)Net fiscal incidence -11% 56 % ... ...Net benefit inciderwe 0' 67-% ... ...

Since a la Brownlirng these transfers are includ(ledI in the incomiie base,while the indirect taxes burden solelv fac(or income, the outtcomiie is atax ratio for the rich of 11. 1 percent and 6.7 percent for the poor.

In the comprehensive stu(lies ok b)udget inci(lence, transfers wouldbe move(l to the postgovernmnent distribution, and inicomes woul(1total to NNP. ConIsequentlI the o il(ome Nvoul(d be as slhowvn in table 2.

In table 1, ihere is Io prob)lem in finding oUt wlhat inicotnes wvouildbe in the pregovernment ( mo iiterf at tial (situation A). In actuial inci-denice work, wlhat inlcotlues wvoul(d settle dowvn to-that is, counter-factual incomes-were sales andl excises elimiiinated is uiiknioNsi.Browniing's solutioni wvas to increase all factor incomes in e(qual pro-portion so that the aggregate increase would equal the total valueof excise and sales taxes. TI his is the approach used in table 1. To g,etto the couriterfactual situationi A from situatiorn C, it was niecessaryto increase all factor incomies bv orne-eiglhth, that is, bv total taxes,to arrive at the aggregate equal to NNP. This illustrates the factthat couniterfactual or pregovernment income itself depends on theincidence assumptions.!

A basic implication of the comiiprehenisive approaclh is that theinicideuL e of excise and sales taxes may or may not be regressive. Theybur(leIn factor incon es, which are made to sumIl to total incomes ingenerating the counterfiactual pregmoerti nicirt income used in thecopl c p lehnic e statistical stu(lies. Btut tlhey also burdell incotmes ac-cordinig to the (legree of puirchase of itemis taxed and the change inrelative product prices of those items dtiie to the taxes. 'I'he coIliIcd

6 In Pechrmarla and 0)ner (1974), iiS well as the wvor-k 0o1 the cort)oratioin incotne tiXinspired bn the HaIbet)eIger iodel. pregovert-nnicx t c11oticoes (1d oindeed varv a(cording toinci(len1ce assiniptioris. %lost statistical sit(dies, lhiovever, err on1 th;is point (see Meer-marn 1974, pP). 521-22).

CONFIRMATIONS ANI) CONTRADICTIONS 1247

effect of production and contsumption burdenis is llot clear. Browvn-

inlg, for examlple, imakes the uniustual argument that the "liet effects on

ireal illnOcne resultin)g fromII chaniges in relative p'ices" ar'e close to zel'o

(1978, pp. 660-65). In the comprehensive studTs (if the Browning

approach is uised, in which all factor incomes are inlcretase(d inl equal

pr-opoirtioin by an1 atnoilIlt e(lllal to the total valtue of excise alld sales

taxes), this outcome would meanl that sales and excise taxes are

proportiotial.A fundamental consequenc e of the Bi owning approa Ii is all ill-

come aggregate which mar del)art substantially fr-onii NNI). (Ini table

1, thte Brownin-g income aggregate excee(le(d NNIP by I .1 peircent.)

Pechlman anid Okiler included onyi]N cash tralisft l-s in incomes. TIhlese

accountedl for less thani 5 percent of' thle aggregate. Browniing

broadens the concept of transfers to tratnsfers in kind, stclh as beniefits

fr-onm ediucatioTn ai(I ndie(lical care. Ihese accotnit foir 20 per-cenIt of hlisinlcomlle aggregate. 7 ConlSe(IuelutlyN he ell(nS wN-itll verCy pogreSsie sales

anid excise taxes because, as noted earlier, transfer income is a niega-

tive fuinction of factor incomne. Browinig also enids withi all income

total whliiclh necessarily far excee(is total otitptit (N NP), In tetmIs of tlhe

comlprehensive btu(det. studies, his illcome concept is misdefilled:

Beniefits are part of postgovernmenll illcoile, nlot the p-'g - I1iiii ictl

coilintenractual ."OTne could, therefore, criticize \ownillg b)v asser'tinig that the inl-

come concept is invalid. 'J'he logic of natiolnal inlcome ,I( moiltlug

requii-i-e that hloscmhold incomiies sumll to aggregate outl)ut. Hithierto

iniconmes equial to N NP (or national inicoIIe) have always beell assumed

in tax iincidlenice w;ork. In moving out of the naltion'al accounts

framnework, Browniing has made a radical departure fionm earlier

work. Yet it is also tirtie that if' one is iw cIt''( vw1 in (fel ciit ial ilnci-

denice analvsis, as is ofteni the case ill )oliix (liscussion, Browning's

approach is useftul. Beniefits are part of houselhol(d incomes, and cash

benefits, at least, are taxed wUhtei spent. 'Io conclu(le, the iticidlti'. cc of

excise and(i sales taixes (depen1d on how von ask the qIL'Slion0. 9

7 Browning implicitdy assomnies thit household recipients Naloe benefits in kindi at

their resource (osts. Of course, thev (it) not. He is, therefore. igglcmm ing re source costs

anid incomes. His aggregate woul(d he lettei destrlibed a.s (qiuasi it oTile.

I Browning has all the elemenits needed ftor a comprehensive hudget incidenceanalssis of the Utnited States for 1977. 1 used his da(Ita to (it) juxst ithis and enidiled with the

lowest income (uintile increasing its income bN 53 pert(ent as the conmbinedl result of

tixes and public spend(ing. TIhe highest quintiie's in(o rne was rc(luced hNw 15 per( et.

)etails available on re(luest.9 Charles Mceure (1979, p. 1141) expressed a similar cont lusion in hiis (onnient on

Btonviing's paper: " . . perhaps we shioli cotonsider his [Bro-wrning'sl estitnates to be

reasotnab)le alternatives to thie conventional estimates r-atiher thian uniqlue replacements."

1248 JOURNAL OF POLITICAL ECONOMY

References

Browvning, Edgar K. "The Burden of Taxation." J.P.E. 86, no. 4 (August1978): 649-71.

de Wulf, Luc H. "Fiscal Incidence Studies in Developing Countries: Surveyand Critique." Internal. MonetarT Fund Staff Papers 22, no. 1 (March 1 975):61-131.

Harberger, Arnold C. "The Incidence of the Corporation Income Tax."J.P.E. 70, no. 3 (June 1962): 215-40.

McLure, Charles E., Jr. "On the Theory and NMethodology of EstimatingBenefit anti Expenditure Incidence." Paper presented at the Workshop onIncome Distributioni and Its Role in Development, William Marslh RiceUniv., Hotuston, 1972.

"Commentar v" (on Browvning's paper). In Income Inequalitv, Trenru andInternational Comiparisons, edited by John R. Moroney. Lexington, Mass.:Heath, 1979.

Meermanj, acob P. "The Detinition of Incc)mnie in Studies of Budget Incidenceand Income Distribuxtion." Rev. Income and WVealth 20, no. 4 (December1974): 515-22.

- . "Do Empirical Studies of Budget Incidence Make Sense?" PuiblicFinan ce 33. no. 3 (1978): 295-313.

Mtusgrave, Richard A.; Carroll, John J.; Cook, Lorrne D.; and Frane, Lenore."Distribution of Tax Payments by Income Gr-oups: A Case Study for 1948."ANat. TaxJ. 4 (March 1951): 1-53.

Pechmiiaii, Joseph A., and Okner, Benjamninl A. WTho Bears the 7ax Burdetn?WNashiingtoin: Brookings Inst., 1974.

THE WORLD BANKHeadquarters:1818 H Street, N.W.Washingtore, D.C. 20433, U.S.A.

European Office:66, avenue d'1ena75116 Paris, France

Tokyo Office:Kokusai Building,1-1 Marunouchi 3-chomeChiyoda-ku, Tokyo 100, Japan

The full range of World Bank publications, both free and for sale, isdescribed in the World Bank Catalog of Publications, and of the continuingresearch program of the World Bank, in World Bank Research Program: Ab-stracts of Current Studies. The most recent edition of each is available with-out charge from:

PUBLICATIONS UNIT

THE WVORLD BANK

1818 H STREET, N.W.

WASHINGTON, D.C. 20433

U.S.A.