latin america

More than 20 years ago studies on the redistribution of government revenues made their first appearance in Latin America. The state of the art has not advanced very much since those first crude attempts, and while the methodology of these studies has changed, the change reflects more the increased availability of data and the greater emphasis of government policies on redistribution than any improvement in the theoretical underpinning of these studies. In their paper, Bird and De Wulf (1973) compared the approaches and concepts of 26 studies covering 17 countries in the Caribbean, Central America, and South America. For ease of reference, the results of several of the studies are presented in Table 1 (in the Appendix).

Three new studies (McLure, 1973; Mann, 1973; Escuela Superior, 1973) have been added to these summaries. While basically similar in methodology to most of the studies discussed by Bird and De Wulf (1973), the papers by McLure and Mann represent an improvement in that greater account is taken of the possibility of new and alternative formulations. Mann, for example, expressed Puerto Rican taxes as a proportion of two different income concepts (as in Gillespie, 1966), while McLure’s latest study (1973) uses three different assumptions regarding corporate-tax shifting and shows the pattern of tax incidence in Colombia with and without the coffee export tax.

Bird and De Wulf (1973) discussed the many data problems faced by the authors of the studies surveyed and the various methodological approaches used in solving them. They also discussed in somewhat greater detail the various income concepts and incidence assumptions used in those studies, but the implication of these choices for the final incidence estimates was mentioned only briefly.

Finally, in comparing the tax incidence studies for India and other countries with those made for Latin America, remember that most of the latter studies used an income-class breakdown, occasionally combined with an urban/rural distinction, but that the agricultural/nonagricultural breakdown, on which most of the Indian studies focused, is totally absent. The major concern of most Latin American incidence studies thus appears to have been the interclass equity of the tax burden distribution rather than the intersectoral transfer of resources.

india

The first tax incidence study made for India by the Taxation Enquiry Commission (TEC) (India, 1926) attempted to determine what the weight of taxation was on particular social classes, which in this context meant occupational groups. It was soon recognized, however, that occupational groups were very heterogeneous and could not readily be related to their ability to pay. This classification was therefore abandoned, and the TEC of 1953–54 (India, 1955) therefore used an expenditure size classification, which distinguished between the urban and rural sectors of the population.

In a country like India, where living conditions, income-earning patterns, and consumption habits differ greatly between the urban and rural sections of the population, it was expected on a priori grounds that such a classification would shed more light on the tax incidence pattern than would a simple tax incidence estimate for each income class. Theoretical reasoning thus suggested the urban/rural breakdown of the population, and the National Sample Survey (NSS), which published consumer expenditure for urban and rural households, provided some of the necessary data. For once the availability of data and theoretical relevancy pointed in the same direction.

The urban/rural classification used by the TEC (India, 1955) was adopted by several other studies (Raj, 1959; Sahota, 1961; National Council of Applied Economic Research—NCAER, 1972; India, 1960 and 1969), which perhaps illustrates that this breakdown, once brought to the attention of the researchers, was difficult to cast aside in a country where the urban/rural difference embraces all aspects of economic life. Similarly, in designing the special consumer survey used in its tax incidence study for the State of Gujarat, the NCAER also distinguished carefully between urban and rural households (NCAER, 1970). It was only in the later rounds of the NSS that the Ministry of Finance (India, 1969) was provided with tabulations designed especially to facilitate tax incidence studies. The use of the earlier NSS in a number of studies illustrates a situation faced by tax incidence researchers in many other countries, namely, that the only available data are collected for quite different purposes and despite their shortcomings have to be used as inputs for tax incidence estimates. As discussed in Section IV, sometimes the use of these data may have a substantial impact on the estimates obtained.

A development strategy that stressed the necessity of transferring resources from the agricultural to the nonagricultural sector apparently made the estimation of the relative “tax burden” of these sectors particularly relevant (Gulati, 1960; Raj, 1959; Mathew, 1968; Angrish, 1970; Gandhi, 1966 and 1969). To obtain such agricultural/nonagricultural tax incidence estimates, some studies (Raj; Gulati; Bardhan, 1961) adjusted the incidence estimates in earlier studies (India, 1955) for the urban and rural sectors. Others simply assumed that the urban/rural data did not differ significantly from the agricultural/nonagricultural data, so that the first set of data could be substituted for the latter without introducing an undue bias (Angrish; Rao, 1971; Pathak and Patel, 1970).

Gandhi (1969) took a more comprehensive approach; he estimated the burden on the farmer as a producer and compared it to the benefits received by him from government expenditure. Ojha and Bhatt (1962) used a different breakdown in comparing the distribution of disposable income with that of personal income. They differentiated between the incomes of farm households, salary earners, and nonsalary earners, a distinction that perhaps compares most closely with the distribution of incomes by factor shares—a concept discussed at length in economic theory but little used in studies of fiscal incidence. They also grouped the population by deciles.

Several studies also attempted to determine tax incidence patterns by income classes. Since data on incomes for the chosen subcategories of the population were unavailable, the expenditure classes used in the NSS were adopted in the various studies by the Ministry of Finance. As noted in Section IV, the use of expenditure classes in lieu of income classes may in some sense provide a better classification than certain income concepts because expenditure tends to be more stable than income, and thus yields a closer approximation to “permanent income.” However, the availability of data, not theoretical refinement, underlies the use of expenditure classes in some Indian studies. Other approaches were also followed. Gandhi (1966) used national income data, for example; the NCAER (1970) in its studies of the States of Gujarat and Mysore used incomes as reported in an especially designed consumer survey; and Gupta (1972) used taxable income. Ojha and Bhatt (1962), who distinguished only three income classes, considered the personal income concept the most appropriate one. Most studies imply that little attention was given to the particular biases in the resulting incidence estimates introduced by the use of particular income data.

Most tax incidence studies in India attempted to estimate the “formal incidence” of the Indian tax system. This approach can be distinguished from that followed in most studies on Latin American countries, which tries to estimate the “effective incidence” of the tax system (Bird and De Wulf, 1973). The latter approach attempts to pinpoint the final locus or resting place of the tax, taking into consideration all tax-shifting possibilities, while the formal incidence approach is concerned only with estimating the burden of taxation as it is “intended” by the legislator. This means that “direct taxes” (those on income and wealth) are assumed not to be shifted, while indirect taxes are assumed to be fully shifted forward and to burden the incomes of the consumers. The procedure followed in these studies is thus rather simple. All taxes are classified as either direct or indirect and are then allocated to the various income classes or to the sectors considered. The many intricate and unsolved questions about the incidence of some taxes are thus bypassed. Formal incidence estimates are therefore not in reality an attempt to determine the burden distribution of the tax system or the amount of intersectoral resource transfer carried out through the tax system, but a quantification of the presumed intentions of the fiscal authorities. Few authors seem to realize that it is not easy to determine on whom the legislators intend the tax burden to fall and that the direct/indirect tax classification—with its concomitant burden, distribution—may be erroneous.

Gandhi (1966), however, employed what appears to be an effective incidence approach. Although he explicitly stated the various shifting assumptions used in his study and elaborated on their theoretical justification for India, his assumptions were not much different from those used in other Indian studies. This illustrates the fact that the potentially important distinction between formal and effective tax incidence is blunted through the lack of sophistication of the available tax incidence theory. Export taxes, for example, are assumed to be exported, and income taxes unshifted. Consistent with his own later findings (Gandhi, 1968) and those of Rao and Rao (1971), but conflicting with findings of Laumas (1966) and Lall (1967), Gandhi (1966) also assumed that corporate profits taxes were paid by owners of capital (who were upper-income class and from urban areas). In contrast to the formal incidence assumptions, however, import taxes were assumed in part to be unshifted and to burden the importer-trader.

While the formal incidence approach arbitrarily adopts implicit incidence assumptions, it at least has the merit of exempting the researcher from restudying these assumptions each time a new analysis is made. Most Indian studies thus largely avoided the intractable problem of tax shifting. However, they still had to allocate the tax burden between the urban and the rural, or the agricultural and the nonagricultural, sectors of the economy. The TEC in its first tax incidence studies (1926) estimated the tax content of expenditure for the various expenditure classes of the urban and the rural sectors. These estimates (or slightly modified versions) were then used by various other studies (Raj, 1959; Bardhan, 1961; Gulati, 1960; Mitra, 1963; India, 1960). Gandhi (1966) and Shetty (1971) made more substantial corrections to these studies, and a government study (India, 1969) benefited from the more adequate set of expenditure tabulations provided by the eighteenth round of the NSS in making new estimates of the tax content of the expenditure of both urban and rural households.

The shifting assumptions also influenced the intersectoral allocation of the tax burden. For instance, the assumption that the corporate profits tax was not shifted implied that the rural sector, where no owners of corporate capital were assumed to live, and the lower-income urban sector were unaffected. Agricultural taxes—also taken to be unshifted—were similarly assumed to burden only agricultural incomes. The non-shifting assumptions thus by definition prevent the intersectoral shift of most of the tax burden. The results of a number of studies on intersectoral tax burdens in India are summarized in Table 2.

Until recently, tax incidence studies were made for all of India without paying attention to possible differences that could arise from the existence of various state taxes or from other economic differences between the states. This situation was remedied in 1969 with the publication of the first of several studies that concentrated on particular states. Several other state incidence studies followed, for example, for Mysore (NCAER, 1972), Gujarat (NCAER, 1970; Pathak and Patel, 1970), Rajasthan (Angrish, 1970), and Uttar Pradesh (Rao, 1971). While these studies yielded interesting results, none of the state studies so far has considered the incidence of taxes levied in other jurisdictions, although McLure’s study for the United States (1964) suggests that such interstate transfers of the tax burden may be highly relevant.

other countries

Compared with the numerous incidence studies found for the Latin American countries and India, there are few studies of other developing countries. Since most use methodologies that are comparable to those discussed earlier and by Bird and De Wulf (1973), a short sketch will suffice.

The whole spectrum of possible income concepts has been used to obtain estimates of effective tax rates: expenditure (Azfar, 1972; De Wulf, 1972), income as obtained from a consumer survey (the Philippines, 1964), taxable income (Vlachapapadopoulos, 1971; Haq, 1964; Islam, 1971), personal income fully adjusted for nonshifted taxes and retained corporate earnings (McLure, 1972; Snodgrass, 1974; Azfar; Krzyzaniak and Ozmucur, 1973), and national income (Arsan, 1959; Karageorgas, 1973). Some were partial studies that analyzed the incidence of separate elements of the total revenue structure, for example, income taxes (Haq; Islam; Vlachapapadopoulos), while the remainder covered the whole range of taxes.

The shifting assumptions used are also similar: indirect taxes were generally assumed to be shifted forward, and direct taxes unshifted. McLure (1972), however, assumed export taxes to be shifted backward to the producers of tin and rubber products in Western Malaysia. The possibility of a further backward shifting of these taxes was mentioned but was not incorporated in the estimates. Dealing with the same questions, Snodgrass (1974) generally adopted the reasoning followed by McLure (1972) and Edwards (1970).

Corporate profits taxes were assumed to be shifted forward in various proportions, but no backward shifting was considered. On the whole, these studies reflect a more explicit consideration of the particular business environment of the country concerned than some of the studies surveyed earlier (De Wulf, 1972; the Philippines, 1964; McLure, 1972; Azfar, 1972; Minford, 1970). Perhaps because of McLure’s previous research (1964) on the exportation of the tax burden to families outside the jurisdiction of the areas studied, he (1972) explicitly accounted for corporate profits taxes taken to be exported to nonresident shareholders, as well as for the export taxes that, although unshifted, act to reduce the incomes of nonresident owners of the productive facilities. To the extent that corporate profits taxes are assumed to be shifted forward, and that exports from the incorporated sector are small, the omission of those logical adjustments in these studies has little effect on the results. Where these conditions are not present, however, the absence of those adjustments will result in an overstatement of the effective tax rates of certain subgroups of the domestic population.

Some authors who were aware of the vital importance of particular tax-shifting assumptions and alternative income concepts used different methodologies to estimate effective tax rates (Azfar, 1972; De Wulf, 1972; and Kryzyzaniak and Özmucur, 1973). McLure (1972, p. 83) obtained a U-shaped pattern of effective tax rates but noted that this “can be explained largely in terms of the export duties on tin, rubber, etc.” He added that without these taxes a proportional tax rate pattern would probably have been obtained. A further illustration of the impact on results of the methodology used is the Philippine study in which the lack of expenditure data led to the assumption that luxury products were consumed only by families whose incomes exceeded P 10,000. The result was then that the effective tax rates became progressive at that level.

The sectoral emphasis, which typified the Indian studies reviewed, was also used by Azfar (1972) for Pakistan and Arsan (1959) and Oker (1952) for Turkey, as well as in the analysis of direct taxes by Islam (1971), who indicated that the importance of intersectoral transfers of resources was the main reason for this approach. Owing to data problems, both Islam and Vlachapapadopoulos (1971) used a more descriptive approach in their analyses of the incidence of indirect taxes. Their resultant conclusions are generally as informative as a correct interpretation of incidence estimates that look more precise, and are less likely to be misinterpreted. The data shortcomings that led to the adoption of this descriptive method may well prove to be a blessing in disguise (see Bird and DeWulf, 1973).

A few authors (Haq, 1964; Islam, 1971; Vlachapapadopoulos, 1971) estimated the incidence of income taxes on taxable income. This approach is, by reason of its limited framework, of relatively little interest. Income taxes are shown to reduce income inequality; yet, no other result could be expected. Such analysis, found also in Gupta’s (1972) study for India and Boelaert and de Vliegher’s (1973) study for Belgium, leaves tax evasion and nontaxed incomes out of account and yields conclusions that could have been reached by a simple inspection of the formal rate structure.

Finally, Minford’s (1970) ambitious study of Malawian tax incidence goes beyond estimating what can be called the “first-order effects” of the taxes, and as such differs from all other studies thus far surveyed. His “paper sets out a theory of how different taxes affect the Malawi economy” and attempts to estimate “the effects [of the different taxes] on the economic variables whose behaviour is relevant to the efficient management of the economy” (p. 67). He analyzes the long-run effects of the tax system on the following: (1) the supply of labor; (2) the efficiency of domestic production; (3) the balance of payments; (4) incidence between domestic sectors; (5) incidence between domestic income groups; and (6) effect on investment. In setting up the model used for tracing the second-order and the higher-order effects of taxation, Minford reviews the tax-shifting possibilities. He concludes that currently withheld income taxes (pay-as-you-earn), income taxes levied on higher-salary earners (mostly expatriates), corporate profits taxes, and license fees are, within the Malawian context, fully shifted. Successful tax shifting influences the price level, consumption, and investment, which in turn gives rise to compensating increases in certain salaries and to changes in smallholders’ marketing of cash crops. These mutually interdependent reactions then continue until their final effects work out.

The approach taken by Minford is to analyze first the effects of the existing tax mix of Malawi, and then the effects of the taxes assuming that each of the individual taxes provides all tax revenues. The author claims that these simulations “bring out qualitative differences” (p. 77). However, the assumption that all tax revenues are obtained through one tax is rather extreme and will completely disturb the sectoral income distribution. The results, which are based on the existing income distribution, are thus unlikely to be meaningful. In all fairness to Minford’s approach, however, his assumption that all revenues are collected through one specific tax does not differ in substance from the basic assumption used in differential tax incidence estimates, where a proportional income tax is usually assumed to replace all taxes.

Despite the many problems with Minford’s approach, however, the questions he attempts to answer in this paper are extremely interesting, and the paper, though unsophisticated, ranks as one of the most stimulating tax incidence studies reviewed here.

summary of results

Most of the authors mentioned here realized the tentative nature of their estimates, but nevertheless apparently believed that the general trends discovered were close enough to reality to permit some policy recommendations. While this section summarizes the results, Section IV reviews the methodological and statistical problems that cast doubt on the value of most of these studies.

Excluding some of the less frequently used approaches, the studies reviewed here have been divided into two major groups—those that concentrate on an income or expenditure size group classification, and those that focus on a sectoral classification. Several studies used both these approaches and as such gave a more complete picture of tax incidence.

Although there is no clear-cut and generally accepted method for measuring precisely the progressivity or regressivity of an effective tax rate schedule, a brief inspection of the effective rate patterns suggested by the various studies surveyed here may assist us in summarizing the results. Some of these rate patterns are depicted in Table 1.

Of the 44 studies surveyed here covering 22 countries other than India, most used a population breakdown by income class and discussed and/or estimated effective tax rates for these income classes. (The studies on India by and large emphasize the difference between the urban and rural sectors and, occasionally, the differences among income classes within these two sectors: their results are discussed later.) For 12 of these studies, owing to their specific approach or their limited coverage, no general impression of suggested rate progression could be obtained; of the remaining 32 studies, 22 suggested some progressivity in the effective tax rate schedule. This progressivity pattern was often an uneven one and frequently did not extend up to the highest income or expenditure brackets or started only from the second or third income class. The degree of progressivity of the tax rates also varied from steep (however defined) to moderate. The results of 8 other studies could be characterized as suggesting some wandering proportionality in the rate pattern. These countries are mainly in Latin America, which explains the somewhat different conclusion of the Bird and De Wulf (1973) paper. The taxes of only two countries (Greece and the Philippines) could be characterized as regressive. While results of different studies covering the same country differ, for no country did the results of different authors contradict each other significantly. The 1966 effective tax rate progression obtained by McLure (1971) for Colombia, for instance, exceeds the rate of progression suggested by Taylor and others (1965) for 1962 and Bird (1970 a) for 1963. While in part this results from differences in methodology and basic data, it also follows from the income-class breakdown that was used. The quartile distribution, used by both Taylor and others and Bird, uses wide income brackets, thus averaging out the potential rate progression that could exist for this group of families. Breaking up this fourth quartile into five or six separate income groups, as in McLure (1971), would differentiate more sharply between the families of that group and would illustrate rate progression more clearly.

The general impression left by the studies surveyed here is that the tax system in developing countries tends to burden the incomes of rich families relatively more than the incomes of the poor. While a few studies suggest proportionality of the tax structure, there are only two exceptions, which indicates that the tax system is basically regressive. This finding is interesting in light of the prevailing view that tax structures are regressive in developing countries, since they rely heavily on indirect taxes. Some of the authors mentioned earlier seem pleasantly surprised with results that indicate that the tax system either improves the income distribution or, at least, does not impose a higher tax burden on the poor than on the rich.

This conclusion contrasts with the view of others who have studied the impact of tax policies on income distribution. Meerman (1973), for example, after comparing the results of ten tax incidence studies made for developing countries, was skeptical about the potential redistributional effects of taxes in developing countries. Similarly, a paper presented at the Third Interamerican Conference on Taxation (held under the auspices of the Organization of American States, Inter-American Development Bank, and United Nations Economic Commission for Latin America) concluded that neither through progressive taxation nor through expanded public spending for social services could the distribution of income be improved significantly (OAS, 1973, p. 354). The impression obtained from analyzing the results of the tax incidence studies surveyed here does not coincide with this assessment of the tax system; the impact of public spending on income distribution is discussed in Section II. These conclusions may result in part from the unfamiliarity of the authors with the results of the full range of tax incidence studies surveyed here, or from a (possibly justified) suspicion that these results are not convincing enough to supersede conclusions arrived at through deductive reasoning or through less ambitious interpretations of the piecemeal data that are available on the subject.

Studies that estimate the average effective tax rate progressivity for a specific country are investigating the vertical equity of taxes: do rich taxpayers pay a larger share of their income in taxes than the poor? These estimates are presumably to be appraised within a more normative framework of how the tax burden ought to be distributed over the population. In reality no such framework has been generally agreed upon. In addition, many researchers have addressed themselves to questions related to horizontal equity: are families with equal incomes paying the same taxes? This question is highly relevant, since the average tax burden calculated for each income class hides great differences in tax burdens for individual families. If it is true that the dispersion around the average of the effective tax rate for individual families increases as the average income of the income classes rises, then these average effective tax rate estimates have even less equity significance than suggested by the inadequate estimating procedures used (Tanzi, 1974). Those who want more informative results are thus led toward disaggregation of these totals.

The disaggregation most frequently used distinguishes between rural and urban families, or between agricultural and nonagricultural families. Table 2 summarizes the results of several Indian studies that have estimated these sectoral tax burdens. The results for a few other countries that have analyzed the tax burden along sectoral lines are given in Table 1 (Brazil, Pakistan, Tanzania). Although these estimates vary, especially for India, these studies generally agree that taxes represent a smaller share of income for the agricultural sector than for the nonagricultural sector. Most authors also conclude that, in fact, the agricultural sector was undertaxed. Periodic exhortations of increased agricultural taxation seemed to constitute the only logical policy conclusion to be drawn from this (e.g., India, 1972). The relative importance of the agricultural sector in the Indian economy and the financial needs of the successive development plans only added fuel to the arguments for a greater tax effort in the agricultural sector.

Several authors, however, noted that the existence of undertaxation of the agricultural sector could not be shown simply by effective tax rates. A lower tax rate for the agricultural sector might, for example, reflect the fact that the average incomes of that sector are lower than those in the nonagricultural sector. Several approaches were followed in dealing with this question more adequately. Some authors estimated effective tax rates on a per capita basis for each sector (Bardhan, 1961; Mitra, 1963; Gandhi, 1966; Mathew, 1968; Arsan, 1959; Oker, 1952), while others concentrated on the effective tax rates by income or expenditure classes for each sector (India, 1955, 1960, 1969; Gandhi, 1966; Mathew, 1968). The latter approach was also followed in the studies on Brazil (Sahota, 1968; Aaron, 1968), Pakistan (Azfar, 1972), and Colombia (McLure, 1973).

The estimates of the effective tax rate on a per capita basis did not challenge the theory of undertaxation of the agricultural sector despite the fact that families in the agricultural sector tend to be larger than families in the nonagricultural sector (Gulati, 1960). However, effective tax rates on a per capita basis did not appear to Gandhi (1966) and Shetty (1971) to contain sufficient information. These authors therefore computed the taxable income for each sector by deducting subsistence income from total sectoral income. Shetty, for example, found that the taxable capacity of the nonagricultural sector over the period 1961–66 was 10.8 times that of the agricultural sector using data unadjusted for subsistence income, but only 4.4 when adjusted, and that the tax burden of the nonagricultural sector exceeds the burden of the agricultural sector by a factor of four.

Without introducing an explicit standard of equity, however, no general conclusion about the overtaxation or undertaxation of either sector is possible. The normal procedure of comparing tax “burdens”—expressing taxes paid as a ratio of income—in effect takes proportionality as the appropriate equity norm. Following a lead given by Frank (1959) and Bird (1964), both Gandhi (1966) and Shetty (1971) introduced alternative progressivity indices. Gandhi concluded that for “a mild degree of ‘desirable’ progression” (e = 1.5 in an equation where taxable capacity equals income raised to the power e), “the inequity was greatly in favor of the agricultural sector” (p. 63). The few years that constituted exceptions were explained by particularly bad harvests. In an extension of this approach, Gandhi estimated the respective taxable capacities of both the agricultural and the nonagricultural sectors by considering, in addition to per capita income and subsistence income, data on income distribution, wealth distribution, and wealth ownership. Although tentative, these extensions did not lead Gandhi to revise his conclusions. Shetty, on the other hand, found that contrary to the “conventional wisdom,” the agricultural sector was relatively overtaxed during the years covered by the three Five Year Plans (1951/52–1965/66). Undertaxation of the agricultural sector, according to him, occurred only after 1966. Neither author considers the proposition that the appropriate level of taxation cannot be resolved in terms of tax equity alone but depends also on other economic and social objectives (Bird, 1974).

Those Indian studies that concentrated on income or expenditure size classes found a regressive effective tax rate structure in the agricultural sector (mainly because of the absence of taxes on agricultural income) and a progressive tax rate structure in the nonagricultural sector (Table 1). These findings not only appeared to confirm the earlier findings of the relative undertaxation of the agricultural sector but also suggested that this undertaxation held for all income groups and increased with the level of income.

In contrast to the situation in India, the effective tax rates of both the rural and the urban sectors in Pakistan (Azfar, 1972) and Colombia (McLure, 1973, in his estimate where the coffee export tax is included and the corporate profits tax is borne by shareholders) appeared to be progressive, although more so in the urban than in the rural sector. In Brazil, on the other hand, the tax burden in the rural sector (Sahota, 1968; Aaron, 1968) was found to be proportional over the greater part of the income range, and then slightly progressive for high-income families. For the Brazilian urban sector, the effective tax rate pattern was found to be U-shaped. Although these results have not received the same publicity as those for India, they too suggest that preferential tax treatment is received by many families in the rural sector. To determine whether rural undertaxation really favors that sector it would be necessary to obtain estimates of how the government allocates its expenditure between the two sectors. Few authors extended their analysis that far, and those who did so (Mathew, 1968; Gandhi, 1966) faced great problems, as noted in Section II.

That national averages are not always representative of individual states has been clearly illustrated by the work of the few authors who have studied tax incidence for individual states. In India, for example, Rao (1971) found that in 1960/61 the effective tax rates on both the rural and urban sectors of Uttar Pradesh were basically identical, while for 1965/66 the effective tax rate of the nonagricultural sector was 7 per cent, compared with 5 per cent for the agricultural sector. Since average nonagricultural incomes are estimated to be 3.5 times as large as incomes in the agricultural sector, this result would suggest, on average, overtaxation of the agricultural sector even if only a mild degree of progressiveness were assumed to be desirable (e.g., exemption of subsistence income). A similar result was found for Rajasthan (Angrish, 1970). The studies on Mysore (NCAER, 1972) and Gujarat (NCAER, 1970), however, confirmed, for those states, the all-India conclusions.

Two estimates of the effective tax rate relating to the urban sector of Gujarat differ so widely—Pathak and Patel (1970) estimated the effective tax rate on per capita income in the early 1960s to be about 20 per cent, while the NCAER (1970) suggested that these taxes averaged 8.5 per cent—as to call into question the validity of the methodology used. This is especially true of the NCAER study of Gujarat, since the income taxes used in estimating progressivity of the effective tax rate structure in urban areas are estimated by applying statutory tax rates to the income data provided by a consumer survey. This procedure is bound to yield a progressive effective tax rate pattern that may bear no relation to reality.

Studies covering Argentina (Consejo Federal, 1963), Brazil (Sahota, 1968), and West Malaysia (Snodgrass, 1974) also estimated the tax burden for different regions of the country. Sahota’s results for Brazil (Table 1) are averages of the estimates obtained for a sample of urban and rural areas. His full results show that effective tax rates found for these small jurisdictions are widely dispersed around these averages. While this dispersion may be partly explained by particular shortcomings of the data used, it reduced the meaning of the national average. Snodgrass, on the other hand, found that the average effective tax rates for three subregions of West Malaysia were close to that estimated for the country as a whole. In Argentina (Consejo Federal, 1963), however, no definite relation was detected between the levels of income in the different jurisdictions and their effective tax rates.

On the whole, it appears that tax incidence studies of separate regions or states that attempt to isolate the economic variables responsible for the variation in tax incidence estimates for families with the same incomes, within the same country (and thus under the same national tax system), can add useful information to the standard national tax study. Like studies of individual families, state studies give one a better idea of the degree of reliability of national averages as a guide to the equity of a tax system.

In conclusion, it should be emphasized that the impression of precise-ness left by the studies surveyed here is definitely questionable; the estimates obtained in these studies are at best approximations. In any study, the overall effective tax rate pattern or the effective tax rates of those income classes that, from a political point of view, deserve more attention—the wealthy and the poor—can be changed considerably by altering the shifting assumptions used or by using different consumption and income data.

approaches

The distributional impact of governmental expenditure can be analyzed from three points of view. First, there is the “money flow” approach, which concentrates on the recipients of direct payments made by the government (Snodgrass, 1974, p. 61, calls these “indirect beneficiaries”) without considering who ultimately benefits from the services supplied through the public expenditure. For instance, school teachers are considered the recipients of educational expenditure independently of any benefits accruing to their students or others. Second, there is the “benefit” approach, which emphasizes the services rendered by the public expenditure and those who in the final analysis benefit from those services. To continue the illustration, students or their families, rather than teachers, are then considered to benefit from expenditure on education. The third approach would consider how the transfer of purchasing power from the private to the public sector modifies the expenditure pattern in the economy, and affect factors and output prices. The income distributional implications of these changes have not been carefully studied in any incidence study. Yet, they are familiar to cost-benefit analysis as pecuniary externalities, but have been given little attention until recently (McLure, 1974 b).

The government raises revenues to finance its expenditure presumably in order to provide certain services to the community and not to maintain a certain number of civil servants on the payroll or to be able to pay certain contractors. The pattern of payments is an intermediate, not a final, product. Although this picture may be reversed in certain countries or with respect to certain specific expenditure, generally it seems a reasonable assumption. Distributional analysis therefore ought to focus on the distribution of the services rendered and not on the flow of money (Bhatia, 1960, pp. 25–27), although the latter is not without interest.

There is, for instance, specific expenditure that, even within the benefit approach, is best analyzed by focusing on the flow of money. An obvious example would be transfer payments where the recipient and the beneficiary is generally the same person. A somewhat similar situation is encountered when the government purchases goods and services at prices that exceed the fair market price, or when subsidized loans are given to some families or firms, although the benefits of some of these subsidies may, like taxes, be shifted (Shoup, 1969, pp. 153–58).

The benefit approach alone is also inadequate when government expenditure does not provide the services that it is supposed to provide, or provides substandard services. In this case, it would be naive to insist on distributing nonexistent benefits to those for whom this expenditure was allegedly made. Such a situation may arise when the government assumes the role of employer of last resort for certain categories of the labor force, or when the educational system continues adding graduates with certain qualifications to a labor market that may already be saturated with similarly trained people. Even if the additional disbursements made to these employees yield no services, at least those who receive the payments from the government are better off with these payments than without them. The elimination of this inefficient public expenditure might result in reduced taxes, which in turn would give taxpayers a higher disposable income from which a higher private demand for output, and hence for employment, would ensue. Only under the very restrictive assumption that the private demand thus generated would have the same distributional impact as the government expenditure that it replaces would this wasteful government expenditure be distributionally neutral. In fact, however, the government production function is generally labor intensive relative to that in the private sector (Gandhi, 1972).

In a situation of unemployment or underemployment of factors of production, it may thus be highly relevant to determine who (regionally, racially, or by income group). receives government appointments or contracts. In addition to the service or benefit produced for the ultimate beneficiaries, the public servant or contractor realizes benefits from the employment opportunity that are often highly valued. The significance of who carries out the government’s work is not lost on any group in countries in which the ethnic background of civil servants and contractors is an important selection criterion. One might also refer to the substantial increase in the demand for middle-class social workers and “poverty” experts that accompanied the U.S. attempt in the 1960s to wage “war” on poverty.

Finally, some studies that concentrated on the urban/rural classification for most of the expenditure considered in fact used the money flow approach as much as the benefit approach in that they allowed for very little shifting of benefits between the sectors. Money spent in a certain sector or region was assumed ipso facto to benefit only that sector or region.

On the whole, however, most studies adopted the benefit approach. Their principal task was therefore to allocate to specific subgroups of the population the benefits assumed to be generated by public expenditure. Hence, a method of benefit valuation was required.

The absence of a market system for most of the goods supplied by government and the resultant lack of a process of output valuation that, under certain circumstances, would have given an adequate index of their benefit content led all the studies reviewed to equate the value of publicly produced goods with their cost of production. Although all studies acknowledged this assumption to be unsatisfactory, it seemed to be the only available method, given the state of the art. This procedure, however, assumes that resources are allocated efficiently between the public and the private sectors. Under certain circumstances, the political process of decision making can conceivably yield this result, but it seems a rather implausible generalization on the whole. To say that one cannot think of a better assumption is not to say that this one is any good. The welfare interpretation of incidence results obtained by valuing government output at the cost of the inputs used is, at best, uncertain and, at worst, quite useless.

Only Sahota (1972) discussed this problem at some length. He conceded that not much was known about the optimality of the resource allocations between the private and public sectors or within the public sector. However, he suggested that a cross-sectional international comparison of government expenditure would yield a kind of yardstick to measure the optimality of Panama’s government expenditure. The comparison turned out favorably for Panama, which had not done badly in allocating its public resources and also was on the right side of the regression line in public administration. While Sahota stressed the desirability of output measures, he believed that when direct output measures were impracticable, using input as a surrogate for output would be less unrepresentative than it would be if Panama’s allocation of public resources were not so good. However, Panama’s specific situation (e.g., hardly any defense expenditure) makes such comparisons uninformative with respect to the purpose at hand. On the other hand, the finding that Panama compares well with an international sample in its use of public funds can only tell us that Panama seems to follow a general pattern that is similar to the other countries in the sample. It does not provide information about the efficiency or optimality of resource allocation in Panama.

The method of equating the value of government output to the cost of government inputs may be called an accounting approach. When $1 billion is spent on education, exactly $1 billion worth of benefits is therefore allocated either to those who receive educational services or to those who receive payments from the government. In contrast to this method, the “behavioral” approach to allocating these services seems highly relevant, although extremely difficult and largely unexplored: the services provided by the public sector are valued according to their appraisal by the alleged beneficiary. One procedure for allocating specific expenditure benefits would be to allocate the benefits (valued at production cost) to different subgroups of the population. This allocation might then be used as a first approximation of the value of the services provided, since they were produced with the use of factors of production for which the market has established prices. This first approximation then would have to be adjusted for the “subjective valuation” of these services by the presumed beneficiaries. Failure to make this adjustment implicitly assumes that each recipient of the public services would buy this particular package of services at these prices if he were given the money (corresponding to the estimated benefit) to do so. An interesting approach along these lines, although one with many problems, is presented in the study by Neenan (1970) and the contribution by Greene (1973). One problem is that only in certain circumstances, and in a roundabout way, do the beneficiaries of government services have a say in the composition of services provided, so that to a large extent a package of services is forced upon them. The situation is similar to that of Kleinwächter’s Flügeladjutant (mentioned in Simons, 1938), who benefits from free opera and hunting while performing his duty but hates both of these activities. The conflict between “consumer sovereignty” and “taxpayer sovereignty” referred to by Weisbrod (1965) and the discussion centering around income distribution in money or in kind in current antipoverty programs in the United States refer to the same general problem.

There is no systematic explanation of the discrepancy between the perceived costs and benefits of government expenditure, so it is not surprising that no study has considered the problem. All expenditure incidence analyses are subject to severe criticism for this and other reasons discussed later in this paper, where further attention is also given to the possibilities of the behavioral approach. The following discussion of technical allocation procedures and of the results of the studies surveyed temporarily suspends the inherent disbelief with which one ought to regard these exercises.

expenditure on general benefits

As Table 3 indicates, there appears to be no great variation between the allocation bases used in the recent studies of developing countries and those used in Gillespie’s (1965 and 1966) well-known studies of the United States and Canada.

One might think that the rather uniform treatment of government expenditure in these studies results from a consensus arising from the advanced state of theorizing and empirical testing on the matter. Nothing would be further from the truth. All authors acknowledge that the allocation formulas used are arbitrary and oversimplify the allocation pattern (Bonnen, 1969). What has happened is that certain basic allocation formulas were first attempted (Adler and others, 1952; Wallich and Adler, 1951) and then further elaborated upon by others (e.g., Gillespie, 1966) who seem to have set a trend. Despite this uniform pattern of allocation formulas, there is in fact general agreement on their fundamental unacceptability. The overriding interest in redistributional effects of government activities explains why these formulas came to be used.

Expenditure on general benefits is incurred in producing social goods, or those goods for which private allocation of resources is inefficient because they are consumed jointly by all. Defense, justice, diplomacy, and general administration are usually considered to be in this category. The allocation of the benefits of such expenditure among the subgroups of the population obviously poses unusual difficulties. Various procedures have been followed with vastly different results.

First, several authors have distributed this expenditure in proportion to income. One argument for this procedure is that people benefit from governmental protection against the violation of the body of existing laws in proportion to what they have to protect, that is, income-generating activities and/or wealth (e.g., Bonnen, 1969, p. 423). Adler and others (1952) and Bird (1970 a) use this as their only allocation basis for this expenditure. Urrutia Montoya and Sandoval (1971) use a variation on this procedure that leads them to allocate all expenditure on general benefits to the wealthiest 10 per cent of the people of the country, who have the most to protect. Mann (1973) also uses this basis as an alternative. Other authors provided estimates using this proportionality rule but clearly preferred another allocation formula (Wallich and Adler, 1951; Aaron, 1968). Sahota (1972), judging the various items of expenditure according to their particular merits, allocated only part of them proportionally by income groups.

A second widely used method of allocation is based on the assumption that each family benefits equally from general government expenditure. The benefits from this expenditure can then be distributed on an equal per capita basis. This formula was often used as an alternative (Mann, 1973; Wallich and Adler, 1951; Andic, 1964; Urrutia Montoya and Sandoval, 1971; Aaron, 1968), or as appropriate for certain expenditure only (Sahota, 1972). Bhatia (1960), on the other hand, distributed 50 per cent of expenditure for protection on an equal per capita basis and 50 per cent in proportion to income. Without giving any rationale, he then proceeded to distribute the remaining government expenditure on general benefits in proportion to the distribution of the expenditure on specific benefits.

Although individual authors thus expressed their preferences, on the whole it is clear that most felt that no a priori reason exists for favoring one allocation formula over the other. This problem would not be serious if the results were insensitive to the alternative formulas used. Wallich and Adler (1951), Mann (1973), Aaron (1968), and Urrutia Montoya and Sandoval (1971), however, illustrate the considerable sensitivity of their results to alternative allocation formulas by estimating the distribution of the benefits of general government expenditure under different assumptions.

Aaron (1968) and Aaron and McGuire (1970), followed by Maital (1972 and 1973), were unhappy with the lack of conceptual basis for the allocation formulas mentioned earlier. They therefore introduced the concept that the willingness to spend money on public goods may be assumed to rise more than proportionally with income. But it quickly became apparent that altering the specification of the utility of money function would considerably change the distribution of the benefits from the public expenditure over the different income classes. The introduction of this factor—although an improvement over the older and cruder allocation methods, if only because it is more explicit—thus does not remove the arbitrariness of the allocation procedures. Green (1973), analyzing the decision-making models implicit in the distribution of the benefits arising from government expenditure, further points out that Aaron and McGuire’s results are only plausible within one such model. As several other decision-making models are plausible, and as no agreement exists on a specific utility of money function, there is no unique, agreed way to allocate the benefits from general public expenditure.

A final point on expenditure on general benefits that has been made by Gandhi (1972) is that they may have additional redistributional effects owing to their high labor intensity (e.g. administration), compared with other public or private expenditure. A marginal employment effect might thus result from the increase in general government expenditure. Furthermore, the fact that in certain countries defense personnel are used for socioeconomic purposes, such as road construction, rehabilitation, education, health, and farming, can also have a significant distributional impact, particularly benefiting the rural sector. The explicit treatment of these specific benefit elements of expenditure on general benefits would appear to yield a more regressive (pro-poor) expenditure incidence pattern.

expenditure on specific benefits

In addition to expenditure on general benefits, the government also provides specific services, the beneficiaries of which are, in theory at least, identifiable. Such services include roads, education, and agricultural services. This section surveys the allocation methods most frequently used for this expenditure in the studies surveyed here, prefaced by four comments of a more general nature.

First, the studies analyzed attempted to estimate the distribution of expenditure benefits in a given period of time. This procedure, however, ignores an important aspect, since some public expenditure is for current consumption, while some is for investment. Expenditure on consumption is correctly allocated during the time period in which it is made. Expenditure on investment, on the other hand, requires a different approach because it generates benefits in future years. The allocation of all investment expenditure in the year in which the outlays are made is incorrect. The assumption that the benefits that a family receives from past investment expenditure are reasonably well approximated by allocating this year’s capital expenditure avoids this problem. But this assumption will not do when investment outlays are growing over time, or when the composition of investment is altering sharply. The accounting approach in which the benefits generated are assumed to equal the expenditure is also questionable, since good investments should yield a positive return and bad investments yield negative returns.

Few studies gave much attention to this problem—Mathew (1968) circumvented the problem by allocating only current expenditure, while Snodgrass (1974) provided separate incidence estimates for investment outlays and for public consumption outlays. In future work on these problems, it would appear most useful to recognize explicitly the time dimension and to attempt to correct the assessment of the distributional effects of public investment.

Second, government expenditure on some specific benefits—for example, public health, education, and transportation (Weisbrod, 1965)—not only provides benefits to some more or less identifiable individuals or subgroups but at the same time increases the overall efficiency of the economy. Few studies seem to have taken this into consideration, since most of them allocated all the expenditure on specific benefits to selected subgroups of the population. While Adler and others (1952) recognized this possibility but did not incorporate it in their estimate (p. 224), Mann (1973) and Sahota (1972) took it into account in allocating benefits from educational expenditure. A U.S. study (Tax Foundation, 1967) allowed for this factor by providing some estimates in which part of the expenditure on specific benefits was allocated on the same principle as general government expenditure. This procedure seems commendable in some ways, since it represents a recognition of the problem. Yet, given the basic unreliability of the allocation of general government expenditure, the overall operational value of the estimates is not greatly improved by this procedure. An additional reason for not including this factor in incidence studies is that the “general benefits” are presumed to arise in addition to the specific benefits, which most often have been valued at cost and therefore cannot be included by studies that rely on the accounting procedure.

Third, the urban/rural classification was explicitly taken into account only by Sahota (1972), Manrique (1972), and Jallade (1974). As in tax incidence studies, this sectoral differentiation was found to be of crucial importance, considerably reducing the apparent informative value of more aggregative estimates. Aaron (1968) indirectly accounted for the differences between sectors by using allocation bases such as business income and wages and salaries, which discriminate somewhat between different sectors of the economy, but he presented his results for national income groups only. Studies using an agricultural/nonagricultural classification (Gandhi, 1966; Mathew, 1968) take this sectoral approach into account in their basic methodology.

Fourth, an analysis of the intentions of the government that lead to the initiation and/or expansion of certain programs is not a good basis for the design of allocation formulas. While little research on this topic has been carried out for developing countries, the experience gained in the United States, with some of the programs that were initiated for distributional purposes, is informative. Farm programs (e.g., Bonnen, 1968; Schultze, 1972), programs of investment in higher education (Hansen and Weisbrod, 1969), and water resource development (Bonnen, 1969) are only three examples of programs that, once initiated, have developed a dynamism of their own and have turned out to have distributional consequences quite different from those apparently intended. In developing countries, few programs have been initiated with the explicit purpose of redistributing income, but several programs are commonly assumed to have this effect. A critical review of their distributional effects is surely called for in light of the U.S. evidence.

summary of results

It should be clear that research on the distributional effects of government expenditure faces tremendous conceptual difficulties. So many allocation and valuation choices are left to the discretion of the individual researcher that the meaning and value of the estimates produced are inevitably questionable, probably more so than the tax incidence estimates. No single allocation base for general benefit expenditure, for example, stands out on a priori grounds. Nor are the allocation schemes for specific benefit expenditure devoid of conceptual and empirical problems. All this illustrates the fact that “we do not really know today what we are doing to ourselves in equity. Our state of knowledge is frankly pitiful” (Bonnen, 1969, p. 427). As with tax incidence estimates, the political and ethical importance of the issue has led authors to try the impossible in the hope of gaining some valuable information in the process.

The results of a selected number of these estimates are given in Table 4. The general picture that emerges from an inspection of this table is that upper-income groups receive benefits from government expenditure that are a smaller share of their incomes than do their lower-income counterparts. Benefit incidence rates are thus regressive (or pro-poor). (The meaning of the terms “progressive” and “regressive” is reversed in Sahota’s (1972), Manrique’s (1972), and Snodgrass’s (1974) contributions, which does not make the reading of their papers any easier.) Alternative income concepts and allocation formulas would influence these results just as they influence the results of tax incidence studies. Only Mann’s (1973) estimates have been made using alternative income concepts. Urrutia Montoya and Sandoval (1971), Mann, and Aaron (1968) used alternative allocation formulas for their expenditure on general benefits, and the results seemed rather sensitive to the alternative formulas used. The final results are also sensitive to the percentage of total expenditure that was classified as expenditure on general benefits and to the method used in allocating these benefits. The share of total expenditure considered to be general benefit expenditure varies considerably from one study to another.

Given that the benefit incidence pattern is highly sensitive to the particular arbitrary allocation formula used for the general benefit expenditure, we may want to concentrate on the incidence of specific benefit expenditure. Five such estimates are available (Aaron, 1968; Urrutia Montoya and Sandoval, 1971; Sahota, 1972; Bhatia, 1960; Mann, 1960). The rate pattern is unambiguously regressive. The results of a detailed study of the educational benefit distribution in Colombia (Jallade, 1974) also indicates that a regressive pattern of benefit distribution exists. Although many problems remain with these estimates, these results suggest that specific benefit expenditure may constitute the most important fiscal method of redistributing income. While taxes, however fairly levied, cannot make the poor any better off, Table 4 suggests that expenditure on specific benefits may improve their relative real income position.

A review of a smaller sample of estimates similarly led Meerman (1973) to conclude that “on average increasing specific expenditures has some effect in decreasing income-concentration” (p. 31). An uncritical inspection of Table 4 appears to support this conclusion.

When the benefits received as a share of income decrease as income rises, the pattern is said to be regressive (pro-poor), and the “favorable” impression is left that the public expenditure is redistributional. This impression, however, ignores the important fact that the absolute amount of benefits obtained by upper-income families may still be much larger than that obtained by lower-income families. Urrutia Montoya and Sandoval’s (1971) results may be used to illustrate this point. For specific benefit expenditure, they find that the effective benefit incidence rates are regressive. For the highest-income class families, the rate is 4.5 per cent, while it is 57.6 per cent for the lowest-income class. In reality, however, the rich group of families on average was estimated to receive Col$910 in benefits, while the poor group received only Col$302.

The example of tax incidence studies, which customarily present their results as effective rates, has apparently established the format used in presenting the results of the expenditure incidence studies. With tax incidence studies, a progressive rate always refers to larger absolute amounts of tax payments with an increase of income and thus unambiguously reflects the fact that richer families pay more taxes than the poor. This is not true with expenditure incidence studies, and it can be argued that any attempt to evaluate the redistributive performance of the government expenditure should be concerned primarily with the actual amounts of benefits provided to different subgroups of the population.

Contrary to the results summarized earlier, Gandhi (1972) analyzed the benefit incidence of public expenditure in somewhat more qualitative terms and concluded that, although he thought the poor benefited in absolute terms from the major elements of public expenditure in developing countries, “it is perhaps the ‘rich’ who benefit more than the ‘poor’ in relative terms” (p. 12), partly because the allocation of scarce public services is likely to be biased in favor of the socially privileged groups. At the Third Interamerican Conference on Taxation, a paper on fiscal policy and income distribution was presented that treats with skepticism the redistributive effects of public expenditure and also contrasts with the results suggested by Table 4 (OAS, 1973, p. 354). So does the conclusion drawn by McNamara at the Annual Meeting of the Fund and the World Bank in 1973 when he stated that “one can conclude that policies aimed primarily at accelerating economic growth, in most developing countries, have benefited mainly the upper 40 per cent of the population and the allocation of public services and investment funds has tended to strengthen rather than to offset this trend” (McNamara, 1973, p. 20). Given the inherent weaknesses of all existing expenditure incidence studies, the present survey clearly does not support such pessimism but cannot be said to definitely refute it either.

conceptual limitations

Interest in the effect of the government budget on income distribution has stimulated many researchers to undertake tax and benefit incidence studies. Allocating the burden of taxes and the benefits of expenditure over different subgroups of the population has been taken to indicate what the budget does with respect to income distribution. The knowledge thus gained can then be compared with what the public desires the government to do in this respect. The assumption of the whole analysis is that the estimates arrived at are relevant and close enough approximations to reality so that they can be used. As a previous paper (Bird and De Wulf, 1973) has already questioned this premise, a short reminder of the major problems with this approach will suffice here.

The tax and/or benefit incidence estimates shown in Tables 1 and 4 were arrived at, in essence, by aggregating the results of several partial incidence estimates. The purpose of this exercise is to compare two income distributions, one of which is the actual one. If comparison is made with a situation in which there are assumed to be no taxes and/or expenditure, the analysis is an estimate of absolute incidence. If, on the contrary, comparison is made with a situation where a specific tax (or expenditure) is replaced by a different tax (or expenditure) of the same real amount, the result is a differential incidence estimate.

This distinction, to which few of the surveyed studies draw explicit attention, was first made by Musgrave and others (1951) in their study on the tax incidence in the United States. Musgrave subsequently (e.g., 1959, 1965) repeatedly stressed this approach and used it in his own incidence estimates. This approach recognizes the fact that if one wants to analyze the incidence of tax or expenditure measures separately from the inflationary or deflationary side effects that would follow their removal, the tax or expenditure analyzed must be replaced by an equal-yield tax or an equal-cost expenditure. Proportional income taxes, value-added taxes (Shoup, 1969), and income transfers seem to be the most adequate substitutes, as they are often assumed to be neutral, although they are not, as leisure is untaxed. The existing income distribution is then compared with that which, it is assumed, would prevail if the substitute taxes or transfers were to be in effect.

The differences between these two distributions then indicate how each subgroup of the population is affected by the fiscal instrument analyzed. (See the presentation of results in an explicitly differential incidence format by Bobrowski and Goldberg, 1970, and Sahota, 1968.) Absolute incidence estimates, on the other hand, assume away the tax or expenditure to be analyzed and consider that the income positions that would have existed in the absence of the tax or expenditure can be adequately approximated by adjusting the observed income distribution for the incidence estimates. Since both approaches give the same numerical results, the comment has been justly made that the differential incidence approach is really a way to escape allegations that the concepts are invalid (Azfar, 1972).

The real importance of the distinction between the differential and absolute incidence approaches comes only when the results are interpreted. While absolute incidence estimates give the impression that they indicate to what extent each subgroup can be made better off by the removal of a tax (or all taxes) or would suffer from the removal of a specific (or all) expenditure, the differential incidence approach really emphasizes that this interpretation is not conceptually valid.

With respect to net-fiscal incidence estimates, the distinction between the absolute and the differential approach loses its importance. A combination of proportional income transfers and proportional taxes would cancel each other out and leave everyone exactly as if no government tax or expenditure exists—which is precisely what the absolute incidence approach takes as its standard of reference. Urrutia Montoya and Sandoval (1971), following a lead from Gillespie (1966), seem to realize this when they claim that they are assessing the redistributional role of the 1966 budget of the Colombian Government compared with a situation where no government exists at all. Many other authors certainly interpreted their results in the same way.

Whether the absolute incidence or the differential incidence approach is used, the basic assumption is thus that the existing income distribution is the same as that which would exist under an alternative tax/expenditure system or in the absence of a revenue or expenditure system. In light of the many studies of individual taxes that have amply illustrated that no tax is really neutral and that each tax/expenditure system affects resource allocation in various ways, this fundamental assumption is invalid. In an analysis of the incidence of one small tax or expenditure, it may be assumed without great danger of biasing the results that the presence or absence of that tax would not noticeably alter the income distribution. But this assumption becomes untenable when the incidence of large and important taxes and expenditure is estimated. This is so a fortiori when the total budget is the subject of analysis, since all that is known for certain is the actual income distribution, which is itself partially determined by the present pattern of taxation and expenditure. The elimination, for instance, of public expenditure on education would yield a different income distribution, that is, the relative importance of the various income size classes would change and the position of individuals and families would be altered.

A full-fledged general equilibrium model—able to trace the full effects of all the taxes and expenditure—could, ideally, estimate differential incidence. Minford’s (1970) interesting attempt to analyze the tax incidence in Malawi illustrates how difficult it is to incorporate presently existing incidence theory into a general equilibrium framework, owing largely to the fundamentally partial equilibrium framework within which this theory is cast and to its loosely defined time dimensions.

A moment’s reflection on the forces leading to budget decisions in any country indicates the extent to which they are determined by the prevailing income distribution and, in turn, determine it. The mutual interdependence of these elements makes it impossible to substantially modify one without altering the other. Budget incidence studies ignore this impossibility and radically alter a part or the whole of the budget without acknowledging the inevitable concomitant changes in income distribution.

Even apart from these problems, aggregating the results of various tax and expenditure incidence estimates, each obtained within its own time dimension, also leads to conceptual difficulties (Prest, 1955 and 1968). The full forward shifting of indirect taxes, for instance, implies a long-term analysis (perfectly elastic supply curve), while the nonshifting assumption of the personal income taxes (inelastic supply curve of labor) calls for a short-term analysis. Each of these conceptual frameworks may well be valid for certain purposes, but the results so obtained cannot logically be aggregated. Most of the other conceptual problems with tax analysis, noted in Bird and De Wulf (1973), also exist, sometimes in exacerbated form, in expenditure incidence studies.

statistical limitations

Fiscal incidence studies all rely to some degree on the following propositions: (1) income provides a certain amount of utility to the income receiver; (2) paying taxes reduces the amount of income that can be spent at the discretion of the income receiver and thus reduces the utility obtainable from a given income; (3) the benefits received from public expenditure can be expressed in monetary terms and compared with or added to private income. However, the usual procedure of making incidence estimates ignores some important aspects that presumably ought to be given due consideration in any study of how the public sector affects the welfare of its citizens.

Equating the welfare of families or citizens with their incomes is, for example, a common procedure in economics but one that grossly simplifies an intrinsically unmeasurable experience.

To illustrate, the income differentials that are required to induce workers to engage in socially nonprestigious, hazardous, or unpleasant jobs are often referred to as equalizing differences. Such differentials—which are expressed in easily measurable terms—compensate for a part of the nonmeasurable “disutility” connected with performing these jobs. Can it be said that the worker whose income includes these positive income differentials enjoys a higher level of welfare than his fellow worker who does not receive such differentials? If not, a comparison of their measured income is not a good basis when comparing how the budget affects the welfare positions of these workers. Even where the pleasantness of the job is positively correlated with its remuneration—as often seems to be true—simply equating income with welfare is inadequate. That these factors have not been incorporated in any incidence study reflects the fact that research on this topic has not provided the profession with acceptable guidelines on how to incorporate them. It does not mean that their effect is distributionally neutral or irrelevant, an impression often created by studies of income distribution or by studies that use income distribution tables among their data input.

Income used as a proxy for welfare has still another disadvantage in that it focuses on the financial means of satisfying certain needs while ignoring the needs themselves. Computing effective tax, expenditure, or net-fiscal incidence rates for a group of families that are in a similar income bracket but are faced with disparate needs or responsibilities hides as much information as it reveals. That a net-fiscal incidence of 9 per cent is found for the income class of families with incomes between $2,000 and $3,000 a year has different implications for a family of six, for an individual without dependents, or for a retired couple (Bhatia, 1960). Each of these families is so different from a welfare point of view that grouping them together in one income class and considering this income class as a homogeneous group does not do justice to the original intent of the incidence studies. The example of Brown (1972), who estimates for the United Kingdom the effects of tax changes for different types of family in each income class, proves a worthy exception to the rule. This procedure is more informative for most purposes than are more aggregated results. Unfortunately, it is also quite data intensive and consequently beyond the reach of many developing countries.

Irrespective of the family breakdown used in the incidence study, the particular income concept used also influences the results. Very few authors appear to realize to what extent alternative income concepts systematically bias their results one way or another. Using income data obtained from consumer surveys or by adding consumption to savings, taxable income before exemptions and deductions, income inclusive or exclusive of transfers—each of these procedures as a rule will lead to different patterns of rate progression.

Even if these various income data were available for one year, it is still not clear which would be the most relevant in computing fiscal incidence estimates. Income statistics, whether based on survey or census data, reflect the incomes of families as they are affected in any particular year by many transitory factors. Bad harvests, temporary unemployment, windfall gains or losses, the fact that the income earner has just entered the labor market—these are only a few of the elements affecting current income that may have only a relatively minor impact on the fundamental long-term economic position of a family. Such transitory elements increase the number of families who record low incomes in surveys and also exaggerate the number of families in the upper-income group in the sense that the families thus placed are fundamentally different from those whose permanent income would place them in these income classes. Calculating the indirect tax burden for a family whose permanent income would put it in a certain income class but which, owing to transitory factors, is recorded in a lower-income class overstates the indirect tax burden for the latter. Similarly, for higher-income classes the “permanent burden” of indirect taxation would be underestimated. In the absence of longitudinal income surveys, it is as yet impossible to evaluate the full importance of this factor, although in welfare terms its significance is obvious.

Similarly, life-cycle income is relevant for those who consider that the analysis of the redistributive effects of the budget ought to take into account periods longer than one year (e.g., Polinsky, 1973).

Imperfections in the capital market that make it difficult to borrow on future earnings, the inherent uncertainty of the future income pattern for individuals, and the shortsightedness of consumers are factors that may indicate that the life-cycle-income approach to fiscal incidence might not be particularly relevant for an individual’s welfare. The fact that an individual can expect to be a net beneficiary from the public budget 20 years from now is small solace if he is presently a net contributor to the redistributive activities of the government despite his current low-income level. Given the uncertainty of ever reaching the age at which he will be a net beneficiary and the impossibility of increasing his current consumption by borrowing on future income, he is probably correct in feeling unjustly treated. An entire lifetime is thus perhaps too long a perspective to be very relevant in this connection.

purpose of studies

Three major purposes for making tax incidence studies can be distinguished. Most studies wanted to analyze the redistribution of income through the fiscal system (Snodgrass, 1974) and concentrated on the effects of taxes or expenditure or both on different income size classes of the population. They are concerned with vertical equity.

A second group of incidence studies concentrated on the differential tax burden between the rural and the urban sectors and on the transfer of resources between these two sectors. Most of the Indian studies shared this emphasis, which stems from concern with both horizontal equity and development strategy.

The third category of studies was also concerned with the horizontal equity between various geographical areas (e.g., Consejo Nacional, 1967; Snodgrass, 1974; Azfar, 1972) or between various ethnic components of a country (Snodgrass; Adler and others, 1952).

With respect to horizontal equity, although few authors were very explicit about it, there was apparent agreement that unequal treatment of equals should be eliminated. The policy recommendations made to this end were, of course, based on a priori judgment to this effect. On the question of the transfer of resources, the authors seemed to be unanimous that more resources should flow from the agricultural sector to the nonagricultural sector. Since the agricultural sector was found to pay fewer taxes than the nonagricultural sector, the authors stressed the need for more taxation of the former. These recommendations seemed defensible even without any standard as to how much of a transfer should occur. When the results showed taxes to be more progressive in urban than in rural areas, an additional argument for greater rural taxation was available. This basis for determining the level of agricultural taxation has recently been criticized as irrelevant to the question of the appropriate intersectoral resource flow (Bird, 1974, Chs. 2 and 10).

All studies that were surveyed favored redistribution of income from the rich to the poor. A progressive tax system and a regressive expenditure system were thus deemed desirable features of fiscal policy. Results “proving” the redistributive effects of fiscal policy (e.g., Bhatia, 1960; Consejo Nacional, 1967) were welcomed even though the lowest-income classes were often found to contribute a considerable fraction of their incomes in taxes, and those showing regressive taxes led to recommendations for the elimination of the regressivity.

These recommendations and appraisals—quite apart from the questionable quality of the studies on which they are based—lead to a question that was avoided by most studies: how much redistribution of income should there be? Although several authors extended their analysis to estimate the income distribution in the country for which the incidence analysis was made (McLure, 1971, 1972; Sahota, 1968; Webb, 1972 a), most felt that, without further study, they knew that the prevailing inequality was excessive and that public policy should remedy this situation. There was little doubt in the mind of most that more progressive taxation was desirable. As long as no norms of redistribution and progression are presented, however, all such recommendations are very vague. As noted earlier, only Sahota (1972) attempted to explicitly formulate a norm or standard of progressivity, although Gandhi (1966) and Shetty (1971) also referred to this problem in the framework of agricultural/nonagricultural taxation. The degree of income redistribution desired in any country is, of course, a policy matter, to which economists as such can contribute little other than some factual information.

general summary and comments on limitations

This paper has reviewed and evaluated studies that estimated the redistributive effects of the revenue and/or expenditure system of developing countries. The conclusions can be summarized as follows: The present state of incidence theory is such that no clear conclusions can be reached with respect to overall budget incidence. The estimates provided in the studies reviewed were arrived at by adopting simplifying procedures and methodologies that by their nature yield only very approximate results. Their use as a basis for policy formulation is thus rather dubious, regardless of the qualifications of the individual researcher.

The quantitative results in these studies were obtained simply by applying a set of restrictive assumptions to the available data for a specific country. It is doubtful whether such an exercise provides a better basis for judgment on the effects of the budget incidence than would a more qualitative discussion of the subject. Presented in a quantitative format, the results are perhaps potentially powerful educational devices, methods with which to stir the imagination of the public and politicians, yet they can do little more in reality than to make explicit what the researcher already knows (or thinks he knows). Furthermore, these studies convey an unwarranted impression of precision that cannot possibly be removed by customary introductory notes of caution. The researcher who attempts to evaluate overall budget incidence in fact attempts the impossible: no methodology available can lead him to the answers he wants. This skepticism does not arise from the quantification of assumptions in itself but from the aggregation of the results and from the fact that incidence studies often do not even ask the most important questions about income redistribution through the public sector.

This stress on the conceptual and statistical problems facing incidence studies is important not only from a methodological point of view but also from that of political reality. Results such as those reproduced in Table 5 may suggest that budget policies in some countries are redistributive. These results could be offered as “proof” that government policies improve income distribution; they may thus support Myrdal, who says that “radical declarations can act as tranquilisers for the radicals and as substitutes for action” (Myrdal, 1968, p. 768). This use of the estimates is not only theoretically indefensible but may thus reduce public concern with redistributive policies, contrary to the apparent intention of most researchers in this field.

This concern should not be overstressed, however, for few incidence studies appear to have influenced policy making in the past. In the United States, for instance, a country for which a large number of tax incidence studies have been made in recent decades and where the apparent irrationality of the incidence pattern has been amply illustrated, only cursory attention is given to incidence patterns when tax reform measures are discussed (Pechman, 1972 a, p. 180). Myrdal’s (1968) statement that “all social mechanisms in India operate mainly in accord with the power structure” (p. 766) is applicable to fiscal policy and applies not only to India. The interest of the politically dominant group combined with the ignorance of the inarticulate mass of the population are far more determining factors with respect to the income redistributive policies of the public sector than are incidence studies.

Incidence studies, through educating the public as to the “facts,” may, of course, affect the level of public awareness of the distributional impact of budget decisions and may eventually alter those decisions. This would be a lengthy process, however, and one that is most likely to be overtaken in countries where the need for redistributional policy is the greatest by less scientific but immensely more realistic experiences with the prevailing income inequality. The contribution of incidence studies is at best only a minor one in the mobilization of the political will to redistribute incomes. Their appearance is quite independent and perhaps politically irrelevant and not unlikely to tip the balance one way or the other.

The realization of the marginal role played by economists in this respect should not reduce their academic zeal to further investigate these issues, but in presenting their results, economists should be careful to display the kind of modesty indicated by the quality of their knowledge on the subject.

Once the political determination exists to influence income distribution, incidence studies covering different aspects of public policy would become more relevant. Such research, concentrating on those aspects of public policy (including tax and expenditure policy) that are likely to be most relevant for income distributional purposes, could serve as a guide in the trial-and-error process of bringing the income distribution closer to a politically acceptable pattern. Viewing the scope of incidence studies in this light may make them appear less ambitious but more in line with what we know on the subject, as well as more relevant for public policy.

possible directions for research

In light of the rather critical evaluation made here of incidence studies and of the fact that public policy—and budget decisions are among the most important policy tools available—will likely be concerned increasingly with its redistributional impact, it might be useful to indicate what directions for research seem to be more promising than generating more studies of the kind surveyed here. First, however, it seems useful to state clearly what public policy in most countries appears to be really interested in with respect to income distribution.

Public concern about income distribution and how it can be “improved” through government policies follows from the realization that market forces and government activities do not yield an income distribution that conforms to the equity norm held by the public or its representatives, however vague this notion of equitable income distribution may be. As shown earlier, incidence studies are unable to isolate in a satisfactory way the contribution of the public budget to the observed income distribution. But the fact that the present income distribution is regarded as unsatisfactory is reached independently from any incidence studies, and is seldom influenced by it. Net-fiscal incidence estimates of +5.2 per cent or –1.4 per cent for a specific income class do not change the public’s opinion on the relative income position of that income class. Judgments on the fairness of the income distribution are made quite independently of such estimates.

The fact that estimates of net-fiscal incidence are trying to measure the impossible is thus not as great a loss as might appear. Whether incidence analysis can assist policymakers (or social reformers) in formulating methods of closing the gap between the present and desired distribution of income is more important. A change of emphasis away from general studies of budget incidence to studies that trace the distributional implications of specific policy measures is what seems needed. Fortunately, such partial and incremental studies are also more feasible from a methodological point of view.

When a certain distribution of income is considered unsatisfactory, one need not completely overhaul the revenue and/or expenditure system to provide a more satisfactory distribution. Short of a social revolution, such an overhaul would be politically and administratively impossible. Even total abolition of the income taxes or customs duties is rarely considered, let alone the abolition of defense spending or public education. Rather than such drastic changes, most policy decisions are concerned with the relative expansion of different revenue sources and the spending of additional revenues on alternative projects. Public policy decisions are thus on the whole marginal, pertaining to relatively small tax or expenditure changes. Incidence studies can make a valuable contribution to the income distributional implications of these measures. The aggregation of results obtained from static partial analysis—each cast within its own time dimension—makes it a conceptually invalid exercise to estimate total budget incidence. This objection disappears, however, if only smaller budget elements are subjected to analysis (Shoup, 1969, p. 572). The effects on incidence of each of these marginal budget decisions can be presented in its appropriate time dimension. Effects on other variables can be discussed, if not fully incorporated, and it can also be assumed for the most part that these marginal changes will not affect the income distribution so much that tax burdens and expenditure benefits expressed as a share of presently recorded incomes will become irrelevant.

This conclusion has an important bearing on the use that can be made of some of the results obtained in the studies surveyed here. The results of these studies were obtained simply by adding up the income redistributive effects of the different components of the budget. This aggregation constitutes their major weakness. But the incidence estimates for separate items may still be valuable. For example, the estimated incidence of expenditure on higher education or of the tobacco excise may be more appropriate guides for policy than are estimates of total tax or expenditure incidence, because it is the smaller items that are really the policy instruments.

The policy value of these disaggregated estimates will, of course, depend on the adequacy of the respective allocation formulas and the degree of disaggregation used. The better studies provide several estimates for some of the taxes and expenditure whose allocation presents serious conceptual problems. Such effective tax or benefit rates then provide a band within which the “real” incidence is expected to fall. The width of this band may perhaps indicate how helpful incidence research can be in providing assistance in framing the policy decisions about income distribution and budget policy.

Some authors who were otherwise critical of the informative value of most of the incidence studies published so far have suggested that intertemporal comparisons may yield more interesting policy conclusions (Bird and DeWulf, 1973) and are “conceptually easier” (Prest, 1968, p. 92) than incidence studies for a single year. The argument is that the use of an identical methodology (allocation assumption and data) for two different time periods allows some conclusions concerning the evolution of the budget’s impact on the distribution of income.

Intertemporal comparisons also have their shortcomings and problems, however. The major one is that it is a doubtful exercise to compare two aggregate incidence estimates that in themselves are conceptually invalid. In addition, changes in incidence patterns can result from budget measures or from other factors, and it may be difficult to single out these factors from those that the researcher is primarily interested in. When structural changes of the budget have occurred between the two periods considered, it is crucial to consider carefully which taxes or expenditure underwent the greatest relative change and what allocation formulas were used. The comparative results obtained for the two years may differ greatly, depending on the allocation procedure used. Alternative allocation procedures may even provide reversals with respect to the changes in incidence pattern. When this is true, no clear-cut conclusion can be drawn from the intertemporal comparison.

Nevertheless, such intertemporal studies may be useful for some purposes. Kaldor’s (1964) intertemporal incidence study of Chile, for instance, suggested that the difference in tax incidence might be explained by a deterioration of the administrative apparatus over the period considered. Further study of this apparatus was thus recommended. To the extent that incidence studies thus draw attention to possible problem areas, they may indeed perform a useful role—although it is unlikely that any Chilean thought that the tax administration was perfect before reading Kaldor. Webb (1972 b), in an attempt to find out “what happened to tax incidence under the (relatively) redistributive governments of Belaúnde (1963–68) and of Velasco (1968–)” (p. 1), estimated the tax incidence for Peru for the years 1961, 1966, and 1969. He concluded that “the strong redistributive intent of the Belaunde and Velasco governments has done nothing to improve that distribution and it could be argued that both have contributed to a slight worsening by raising taxes most on middle groups and least on the very rich” (pp. 34–35). Webb’s own discussion, however, illustrates how these interesting conclusions could have been arrived at quite independently from his incidence estimates.

As noted earlier, in all the studies reviewed, benefits were valued according to their cost of production. As an alternative to this accounting approach, which ignores real questions about the benefits received, it may be useful to investigate further the “behavioral” approach.

Neenan (1970), for example, in an attempt to verify the thesis of “exploitation” of the central city by suburbs explicitly assumed that “perceived benefits of local government services rise proportionately with income” (p. 129). Benefits received were thus not restricted to being equal to the value of the inputs used up in their production.

To obtain the perceived benefits for each suburb, Neenan (1970) assumed that the cost of production equaled the benefits perceived in the central city (Detroit) in which the expenditure decisions for the expenditure analyzed in his paper were made. The suburbs were estimated to receive a benefit from those services, the perceived value of which equaled the benefits as perceived by Detroit adjusted by the ratio of the median income in the suburb to Detroit’s median income. The result of this procedure is that the same service is taken to yield a greater benefit for suburbs with a median income in excess of Detroit’s median income and a lower benefit for suburbs with incomes that are less than Detroit’s. The author claimed that “on a conceptual basis I think it [the behavioral approach] is preferable to the more traditional method of measuring the benefits of public sector expenditures” (Neenan, 1972, p. 605). The benefits perceived in Detroit, where the expenditure decisions are made and where an identity between preferences of the citizenry and resource allocation (between private and public sector and within the public sector) was assumed to exist, are evaluated according to the traditional “accounting” procedures.

In a formalization of Neenan’s (1970) contribution, Greene (1973) explicated the decision-making model used by Neenan and by Aaron and McGuire (1970) and demonstrated that the valuation of the benefits of general government expenditure depends on the specific model of collective decision making adopted in the analysis. Greene concluded, however, that “given some plausible alternative models and our present state of knowledge about the demands of various groups for public services, the assumption that such benefits and costs are equal may be no worse than any other assumption” (p. 184). Further research on the decision-making process, on the demand for public services, and on the distribution of the tax burden as this is perceived by the various income groups is thus needed to arrive at more specific answers about the benefits derived from the general public expenditure.

Estimates of the benefits derived from specific government expenditure by various subgroups of the population could be similarly analyzed, and the considerations regarding decision-making processes would probably alter the results, in a rather predictable fashion. The more the decision making is concentrated in a small segment of the population (excluding the hypothetical omniscient benevolent dictator), for example, the more the package of services provided by the government will probably coincide with the preferences of the decision-making group. The groups of the population that do not partake in the budgetary decision will in these circumstances be less interested in the particular package of services provided than will the decision takers themselves. Thus, the cost of providing these services appears to be an inadequate approximation of the benefits received from their provision.

In addition to the fact that the benefits perceived by the subgroups of the population are not necessarily related directly to the cost of producing the particular services, the “consumers” of the services were not the only beneficiaries of the services so provided. A better public education system, for instance, can be viewed as a substitute for private training facilities by businessmen, so that even if their children attend private schools they may receive some benefits from public education. Also, the elimination of poverty may promote a stable social climate, the benefits of which go not only to the direct beneficiaries of public hospitals and transfer payments.

Much more research into the politics of decision making in specific countries and into the evaluation of publicly provided services by various subgroups of the population thus seems needed before it will be possible to estimate the benefit incidence of public expenditure. In the absence of such knowledge, studies that estimate on whose behalf expenditure is made may shed some light on the problem.

The discrepancy between the results obtained through the usual accounting approach and the behavioral approach would presumably narrow down or become distributionally neutral when effective mass participation in the political process is established or when differences in incomes and preferences between different subgroups of the population become less important.

All studies reviewed in this paper concentrated on the income distributional results of budget policies. The effects of these policies on distribution are not negligible, but they are not necessarily the most important distributional aspect of public sector activity. Price and wage policies, import policies, land reform, and industrial legislation are only a few of the nonbudgetary policies that may have important redistributional effects and that have received scanty attention thus far. A more systematic analysis of the effects of such policies would also run into serious methodological problems, yet might provide relevant information that should be incorporated in an overall assessment of the income redistributional impact of public policy.

Finally, once it is realized how conceptually unsatisfactory all available incidence studies are, one is naturally inclined to question whether the results are really worth the efforts required to obtain them. Might it not be more economical to limit the analysis to more qualitative or descriptive statements with respect to budget incidence? With general knowledge about the characteristics of the economy under investigation and about the incidence theory pertaining to the budget items studied, it is generally possible to provide relevant speculations, as good as those in conventional incidence studies, on the distributional effects of budget policies. Probably the direction of the distributional effect could generally be indicated adequately. While it might be possible, for instance, to indicate that a specific change in customs duties will have a progressive impact, it might be difficult to indicate just how progressive and whether this additional progressive feature extends itself all the way up to the highest-income class or not. Quantification of the probable effects of such a change would, in most instances, give a clearer picture of their anticipated incidence effects, and might thus assist the making of public policy decisions. An additional advantage (and danger) to results cast in a quantitative format is that the general public and policymakers are more easily convinced by quantitative results than by a lengthy verbal discussion of the same phenomena. Overconfidence in the significance of these results is the main potential drawback of quantification.

Nevertheless, in view of the limitations of quantitative studies of tax or expenditure incidence, considerable scope remains for nonquantitative discussions, which often are able to grasp the most salient income distributional features of specific taxes and expenditure. In fact, the best analyses of budget incidence are generally accompanied by such discussions. Webb (1972 b), for instance, has an excellent discussion of the evolution of the tax burden in Peru during the 1960s. While this discussion is presented as explaining his tables on the tax incidence for 1961, 1966, and 1969, it could equally well have been presented quite separately from those tables. His general conclusion does not really depend on the tabular presentations, which, rather, may be taken to illustrate the conclusions arrived at by careful analysis of the tax changes over this period in Peru.

This paper has discussed many problems concerning incidence studies, problems that cast doubt on the results of such studies. Despite these problems, researchers will no doubt continue for various reasons, good or bad, to make more such studies. It may therefore be hoped that future studies will at least employ alternative formulas to allocate taxes and expenditure among the subgroups of the population, thus illustrating the sensitivity of the results to the various possible allocation methods. It is also hoped that more attention will be given to incremental and partial incidence studies, to incidence comparisons over time, and to some experimentation with the behavioral approach of allocating the benefits of government expenditure. Above all, it is suggested that, given the present highly defective state of the art, researchers should be honest and clear about the severe limitations of these quantification exercises.