Human Capital Development HCD Working Papers The Full Social Returns to Education: I Estimates Based on Countries' Economic Growth Performance Alain Mingat Jee-Peng Tan September 1996 HCDWP 73 Papers in this series are not formal publications of the World Bank. They present preliminary and unpolished results of analysis that are circulated to encourage discussion and comment; citation and the use of such a paper should take account of its provisional character. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The Full Social Returns to Education: Estimates Based on Countries' Economic Growth Performance by Alain Mingat Jee-Peng Tan Alain Mingat is a Directeur de Recherche, Centre National de la Recherche Scientific, at the hIstitute de Recherche sur l'Economie de l'Education based at the Universite de Bourgogne, France. Jee-Peng Tan is a Principal Economist in the Human Development Network of the World Bank. This paper has benefited from the generous and helpful comments of friends and colleagues. We owe special thanks to B. Chiswick, J. Hammer, E. Jimenez, J. Lane, J. van der Gaag, H. Patrinos, L. Pritchett and G. Psacharopoulos. As authors, however, we are solely responsible for all remaining errors. The views expressed in the paper are ours alone, and should not be attributed to the organizations with which we are associated. Abstract This paper reports new estimates of the social returns to education, using countries' economic performance during 1960-85 to capture the externalities from education. The results confirm the social profitability of investing in education but indicate that the returns by level of education are sensitive to countries' economic context. To interpret our results meaningfully it is useful to make a distinction between investments to maintain existing coverage, and investments to expand it. Our findings relate mainly to the second type of investment decision in the context of the following observed initial gross enrollment ratios: 58 percent in primary education, 9 percent in secondary education, and 1 percent in higher education in low-income countries; 95 percent, 28 percent, and 4 percent, respectively, in middle-income countries; and 109 percent, 53 percent, and 9 percent, respectively, in high-income countries. Given these initial levels of coverage, our results suggest that low-income countries benefited most from investments to expand primary education, while in middle-income countries, it was investments to expand secondary education that brought the highest social returns. In the high-income countries, investing to expand coverage in higher education yielded the best returns. Investments with poor social returns include expansion of higher education in low- and middle-income settings and expansion of secondary education in high-income settings. Contents Introduction .................................................................................................1 Existing Evidence on the Returns to Education ...............................................2 Clarifying the Costs and Benefits of Investments in Education ...............................................3 Toward Incorporating the Full Benefits from Investing in Education ............................................... 7 Estimating the "Full" Returns to Investments in Education .............................................. 11 Comparing the Results with the Standard Estimates of Social Returns .............................................. 16 Conclusion .............................................. 17 Introduction That education is a form of investment is now a well-established and widely-accepted idea. Like other investments it is natural to ask whether or not the benefits from education are worth the costs. Not surprisingly, therefore, a vast literature has emerged over the years documenting the magnitude of the returns (see Psacharopoulos 1994 for perhaps the most comprehensive catalog to date). In this literature, estimates are reported on the private and so-called social returns to education. Both measures of returns capture the benefits from education by using individual wage data to estimate the increase in productivity from an extra year or cycle of education. On the cost side, private returns take into account schooling costs incurred by individuals (or famriies on behalf of the students), while social returns take into account public subsidies as well. As calculated, the social returns are in fact private returns net of public costs (which explains why they are always lower than private returns). Despite the flaws, the empirical evidence has had a strong influence on government and donor policies regarding investment priorities and public subsidies for education. In recent years, however, doubts about the validity of the findings on social returns continue to be voiced, mainly because community- wide benefits from education have not been taken into account. The purpose of this paper is to report fresh estimates of the social returns to education based on countries' economic growth performance. Our approach conforms more closely to a proper definition of social returns in which costs and benefits are both accounted for from society's perspective. Specifically, we argue that aggregate growth performance captures to a significant degree the externalities from education, unlike conventional estimates based on individual earnings which ignore them altogether.' This approach is also likely to capture the economic consequences of non- economic effects, such as faster growth arising from increased political stability working through the spread of education. Our estimates confirm the social profitability of investing in education, but indicate that the ranking of profitability by level of education is sensitive to countries' economic context: primary education is most sociatly profitable in low-income countries, but its position is replaced by secondary education and higher education, respectively, in middle- and high-income countries. These results do not undernine the importance of investing in primary education, particularly in view of concerns over poverty alleviation and equity, but they do suggest that as countries develop shifting the emphasis of new investments in education toward the secondary and higher levels may well produce the highest social returns. 1 In a recent paper Ram (1996) makes a similar argument for using cross-country data for estimating social returns to education. Our approach differs from his, however, in that we relate countries' growth performance to initial enrollment ratios while he relates it to average educational attainment of the labor force using a Mincerian specification. In addition whereas we incorporate both direct and forgone costs in our calculation of returns, Ram considers only forgone costs. 2 Existing Evidence on the Returns to Education Table 1 summarizes the existing evidence based on a large number of country-specific studies. Despite the diversity in methods and assumptions across studies, the overall pattern is consistent in suggesting two broad conclusions: (i) the so-called social rates of return to education at all levels typically exceed the usual 10 percent cutoff rate for the cost of capital and (ii) the returns are, without exception, highest in primary education.2 The findings have had a persuasive influence on public debate regarding investments in education in general as well as across levels of education. The World Bank's recent policy paper on education, for example, reflects the general consensus that investing in primary education deserves the highest priority in public expenditure programs, particularly in low- income countries (World Bank, 1995). Equity concerns undoubtedly motivate the strong emphasis on primary education, but the pattern of returns (such as they are) provides the added evidence to convince those insisting on good economic returns on public spending. Table 1: Social Returns to Education by Level, percente Region Primary Secondary Higher Sub-Saharan Africa 24.3 18.2 11.2 Asia 19.9 13.3 11.7 Europe/Middle East/N. Africa 15.5 11.2 10.6 Latin Arnerica/Caribbean 17.9 12.8 12.3 OECD n.a. 10.2 8.7 Source: Psacharopoulos 1994. a/ The reader is referred to the source for similar summaries of the private rates of return to education. Yet the empirical evidence is based on an incomplete calculation of social returns that invariably excludes economic extemalities. As Schultz (1988) notes, "... it is rare for studies to assign monetary values to social externalities of education ...". The incomplete calculation has led some observers to question the use of the available evidence in formulating public policies, especially with regard to allocations of public spending across levels of education. Birdsall (1996), for example, argues that "[we] do not really know the true ranking of social returns across primary, secondary and higher education", because the magnitude of externalities may well differ by level of education. In certain areas of higher education, such as research and postgraduate training in science and technology, and public administration, she points out that the magnitude of externalities may be sufficiently large to cause the social returns to investing in these areas of higher education to exceed those from investing in secondary or even primary education. In a commentary on Birdsall's argument, Psacharopoulos (1996) 2 See below for elaboration of how these rates have been calculated and problems associated with interpreting them. 3 notes that while research activities in higher education may generate externalities, the probability of a real breakthrough is infinitesimally low. In contrast, the probability "must be nearly 1" that someone who has not been to school will remain a life-long illiterate. Based on this comparison, he argues that even in the absence of specific estimates on the magnitude of externalities, it would be socially efficient for education investments to emphasize primary education in countries that have yet to secure full coverage and adequate levels of student learning at this level. The heuristic arguments on both sides find support among some but not all people, and are therefore unlikely the settle the matter. Skeptics will continue to reserve judgment pending concrete evidence on the magnitude of externalities. Such evidence is not easily obtained because externalities are notoriously difficult to quantify. Below we attempt to make some headway in this regard. Clarifying the Costs and Benefits of Investments in Education There is relatively little controversy regarding the concepts of costs and benefits in education. It is in estimating their magnitudes and in implementing the cost-benefit analysis, and interpreting their results, that problems typically arise. Defining the concepts. The costs and benefits of education accrue to individuals and society at large (Table 2). Individuals, or their families, incur two types of costs directly: (i) out-of-pocket outlays (C. 1 in the table) to pay for such items as books, transportation to school, and school fees; (ii) and the value of forgone production (C.2), either in household activities or in the labor market, while the student is in school. Costs that are shared collectively by society consist of tax-financed public subsidies for education (C.3).3 The full cost of education is the aggregation of the costs incurred individually by the students, or their families, and those that are shared at the societal level. On the benefit side, individuals (or their families) capture benefits from education in the form of increased labor productivity (B. 1). For example, better-educated workers may earn more than others with less education, and farmers with some schooling may produce more output than farmers with no schooling. In a well-known article, Haveman and Wolfe (1984) identify 19 other benefits besides increased labor productivity. These non-market effects include the impact of education on personal health, capacity to enjoy leisure, and efficiency in making a variety of personal choices (B.2). Although seldom recognized as such, another non-market benefit of education is the option it creates for proceeding up the education ladder. This benefit applies mainly to the lower levels of education and stems from two features of education systems everywhere. First, admission to a given level of study is typically contingent upon prior attendance at the previous level. Second, most course curricula are designed on the assumption that entering students have already acquired the skills and knowledge 3 Because tax-financed public subsidies may distort prices and impose deadweight losses on the economy, their true economic cost may be greater than their face value. Making the appropriate adjustment is difficult in practice, however, because few empirical estimates exist regarding the magnitude of tax-induced deadweight losses. 4 Table 2: Generic Education Costs and Benefits and Their Accrual to Individuals and the Rest of Society a' INDIVIDUALS SOCIETY w' Cl. Direct costs C3. Public subsidy C (including school fees) (net of cost recovery and adjusted for O possible deadweight losses of tax- S financed public spending) T S C2. Forgone production (lost earnings or other production) B 1. Increased market Droductivity B3. Spillover effects in worker productivity (as reflected in earnings or other work (as when a person's education enhances outputs) the work productivity of his or her co- workers) B E B4. Expanded technological possibilities N (such as those arising from the E discovery, adaptation and use of new knowledge in science, medicine, industry, and elsewhere) T B2. Private non-market effects B5. Community non-market effects (better personal health, expanded (greater social equity, more cohesive capacity to enjoy leisure, increased i communities, stronger sense of efficiency in job search and other nationhood, slower population growth personal choices) I and related alleviation of environmental stress, reduced risks from infectious diseases, crime reduction, and so on) Source: authors' construction. a/ Dotted lines in the table indicate that the identified categories of benefits are not completely separable. b/ Strictly speaking, this column shows the costs and benefits accruing to the rest of society (i.e., to people other than the individuals being educated). 5 taught at earlier cycles. Thus, even when admission criteria are not used (e.g., for some programs offered by open or distance universities), a student may not be able to follow a course of study effectively unless he or she had in fact passed through earlier cycles in the system. Education also yields benefits for society beyond those enjoyed by the educated individuals themselves. According to Lucas (1988), for example, a worker's schooling enhances his or her own productivity as well as those of co-workers, thereby giving rise to classical externalities or spillover effects (B3). The general level of education in the workforce also expands production possibilities, by facilitating the discovery, adaptation and use of more economically rewarding, albeit technologically more demanding and knowledge-intensive, production processes (B.4). These effects are probably closely related which explains the use of a dotted line to separate them in Table 2. Besides economic performance, there are other community-level benefits from education (B.5). These non-market effects include the possible contribution of education to improving social equity, strengthening national cohesiveness, reducing environmental stress through its effect on fertility and population growth, lowering crime rates, and so on. In a recent article, Olson (1996) makes a distinction between "public good human capital" which affects incomes by influencing people's knowledge about and choice of public policies and institutions, and traditional "marketable human capital" which increases people's personal income. To the extent education produces "public good human capital" it also contributes to the production of community wealth. The community-wide effects from education are not always separable from the non-market benefits that accrue to individuals (i.e. B2 and B5 in the table). Using data for the United States, Haveman and Wolfe (1995) found that parents with more education have a significantly smaller chance than other parents that their children will drop out of school, that their daughters will become unmarried teenage mothers, and that their grandchildren will become economically inactive as adults. Although parents suffer personally when their offspring fail to grow up well, the effects extend to the community as well, and substantial public programs have indeed been mounted to address these social problems. The treatnent of costs and benefits in the empirical literature. Two sorts of issues arise in applying the foregoing concepts in empirical work: measurement of the costs and benefits; and taking proper account of them in making the cost-benefit calculation. With regard to the first issue, data on the direct costs accruing to individuals and to society as a whole are usually available, or can be estimated using relatively straightforward procedures. Where appropriate the costs can be adjusted for the effects of repetition and dropping out. More problematic is the estimation of forgone earnings or output, especially for children in the lower grades. Although young children contribute to household production, it is often difficult to estimate the value of their contribution because children typically receive no formal compensation for their labor. Problems can also arise in measuring the benefits. When estimating the private retums to education, personal earnings provide an appropriate measure of the benefits, because earnings are all that matters to the individual. When estimating social retums, however, it is important that earnings reflect the marginal productivity of labor. The match between observed earnings and productivity is 6 closest in competitive labor markets, but such markets may not always exist: in low income countries, for example, most formal sector workers are civil servants whose pay are publicly administered rather than market-determined; and in transition economies where markets are in disequilibrium, observed earnings are likely to be contaminated by the effects of temporary skill shortages and surpluses. Beyond the measurement problems, there is also imprecision in the literature arising from the inclusion of different items of costs and benefits in calculating the returns to education. One of the most common sources of rates of return estimates are Mincerian earnings functions. These functions apply regression analysis to earnings data on a cross-section of individuals to relate people's earnings and to their level of schooling and their work experience. The regression coefficient on the schooling variable can, under certain assumptions, be interpreted as the private returns to education.4 Although it is possible to adjust the result to incorporate data on direct public and private costs, most estimates assume that forgone earnings are the only cost of education. Thus, in terms of table 2 only C2 (forgone earnings) and B 1 (increased earnings) enter the cost-benefit calculation. Another common method of computing returns to education involves estimating the relation between earnings and age within each education group, and then combining the resulting age-earnings profiles with information on the direct costs of education to calculate rates of return (i.e., following the long method in the terminology of Psacharopoulos (1994)). The private returns to education obtained in this way take account of direct costs (CI) and forgone production (C2) on the cost side, and increased individual earnings (BI) on the benefit side. The social returns referred to in this literature take account of an additional item on the cost side, namely the public subsidies to education (C3), while on the benefit side no other effects are included. Summers (1994) is an example of a cost-benefit calculation of education investments where the benefits include only the non-market effects of education. The social returns he calculates differs in definition from uses of the term elsewhere in the literature, however, in that they include only public spending on education (C3) and non-market effects (B4) in the calculus. Although the non-market effects quantified in this example-reduced fertility, as well as child and maternal mortality-are strictly private benefits, the author treats them as public benefits on the argument that society has already defined them as such by using public resources to finance health and family planning services. To summarize, existing estimates of the social retums to investment in education typically fail to take account of extemalities in the cost-benefit calculation. Yet extemalities are what matters for justifying public subsidies. If the extemalities are sizable and differ substantially across levels of education, taking account of them may well modify the ranking of priorities for public investment across levels of education, as Birdsall (1996) has argued. 4 See Becker and Chiswick (1966), Mincer (1974), Willis (1986) and Rosen (1987) for further details and discussion on this issue. 7 Towar-d Incor-porating the Full Ben.efits frioii Inivestinig in Education One possible approach consists of adding up all the benefits identified in Table 2. But beyond the benefits that matenralize as increased personal earnings, the other benefits are likely to prove elusive to quanti y fully. The list is long, perhaps even endless, so one is never sure that all the benefits have been taken into account. And some of the benefits materialize only after a lag, or take time to unfold. As a result, the impact of education observed at a particular time may well under-estimate the fill effects from the investment. An alternative approach is to assume that all the benefits from investing in education are internalized in the perforinance of the economy over the medium-term. In other words, if the education of one person improves his or her own productivity as well as that of co-workers; if education improves personal and community health, leading to increased labor productivity, or indeed produces the same effect through any of the other routes identified in table 2 above, the aggregate impact is likely to show up in the long-tenn performance of the economy. In this interpretation, the relation between education investments and economic performance at the country level represents the "reduced form" effect of all the separate influences of education on economic productivity broadly defined. Economic growth is admittedly a flawed proxy for country performance, since it does not capture all aspects of social welfare. It nonetheless goes beyond the usual practice of focusing only on individual earnings. Given its promise of comprehensiveness, we chose the second approach to estimate the full benefits of investing in education. The relation between schooling investments and economic growth is the subject of a large body of specialized research, and a full discussion of the specification issues would tak-e us well beyond the scope of the present paper. For our purpose, we follow the relatively conventional regression model adopted in the World Bank's study on 7he Ezat Axian Miracle, using the same data as in that study.5 Specifically, we relate real per capita income growth (over 1960-1985) to an array of country characteristics: initial education in 1960, growth rate of the population (over 1960 and 1985), investment spending as a share of the GDP (averaged over 1960 and 1985), and GDP in 1960 relative to the U.S. Because we are interested mainly in the role of education, we treat the non-education factors simply as control variables. Other authors such as Alesina and Rodrik (1994) have also included in the regression equation measures of initial equity (as proxied through Gini-coefficients for income and land distribution). Unfortunately, the scarcity of data on these indicators meant that we could not stratify the sample by income group-an important step in our analysis-without a drastic reduction in subsample sizes. 5 The data (which are publicly available) relate to 113 countries drawn from Barro (1991), Summers and Hcston (1991), and Dc Long and Summers (1991). The specification is consistent with models of the determinants of long-run economic growth which highlighlt the role of hunman capital, ideas and externalities from education. See, for example, Lucas (1988), Barro (1991), Azariadis and Drazen (1990), and Mankiw et. al (1992) for specific applications; and Patrinos (1994) for a review of the "new" growth modcls. Although the use of enrollment ratios in grow,th models is standard practice in the literature, the work of Benhabib and Spiegel (1994) and Pritchett (1996), for example, has raiscd questions about their validity. The issues remain unresolved, however, so we proceed on the basis of the predomiinaiit approach in this area of research. 8 Dollar (1992), among others, has highlighted the role of macroeconomic policies (e.g. openness to trade) in economic growth. Again, lack of a good measure of such policies for a sufficiently large number of countries prevented us from incorporating the variable in our regression. We modify the basic model used in 7he East Asia Miracle in two aspects. First, with regard to initial education, we include enrollment ratios in 1960 for all three levels of education, rather than limiting it to only the first two levels.6 Second, we examine the data separately for three country groupings: low-income countries (with per capita GDP below 20 percent of the U.S. in 1960; 74 countries), middle-income countries (with per capita GDP in 1960 ranging from 20 to 40 percent of that of the US; 19 countries) and high-income countries (with per capita GDP at least 40 percent that of the U.S.; 20 countries). The choice of cut-off points separating the three groups of countries is admittedly arbitrary, but our objective is to allow (within the limitations of the data set) for possible differences in the impact of different levels of schooling according to the initial economic context.7 We used a linear specification and estimate it using median regression to reduce the influence of outliers in the sample.8 The regression results appear in table 3. With regard to the non-education variables, the results are largely unsurprising. First, the negative coefficient on initial income relative to the United States in 1960 in the regressions for all three country groups suggests that poor countries tend to catch up in per capita income over the two and a half decades of observation. For low and middle income countries the coefficients are not statistically significant, however, indicating that a signification proportion of these countries in fact lagged further behind instead of catching up. Second, the coefficients on investment in equipment are positive and statistically significant (and are also of comparable magnitude) in all three country groups. Finally, the coefficient on population growth rates is statistically significant only in the high-income group; in the other two groups, the coefficients are opposite in sign and not statistically significant, suggesting the variable's lack of a systematic influence on economic growth. Focusing now on the influence of education on growth, the results are consistent with previous studies showing that initial levels of investment in education affect subsequent economnic growth. But the impact appears to differ widely both by initial level of economic development and by level of schooling. For the full sample, the results reported in the East Asia Miracle indicate that only primary education is statistically significant, with a magnitude that implies that a 10 percentage point increase in 6 WC supplement the World Bank data by adding data on higher education enrollment ratios in 1960 from UNESCO (1968). For countries for which enrollment ratios were not available in that source, we computed the ratios by dividing the number of students by the population in the 20-24 age group, based on World Bank data. In the few countries for which the computation could not be accomplished, we obtained the enrollment ratio by converting data on the number of higher education students per 100,000 population, based on the average of conversion indices for a sample of countries for which data are available on both the number of students per 100,000 population and the higher education enrollment ratio. 7 The expectation that the impact of, and therefore returns from, education investment are likely to differ according to context is consistent with evidence from micro-level studies (see for example Rosenzweig (1995)) as well as the broad cross-country patterns documented in Psacharopoulos (1994). 8 We also estimated the equation using ordinary least squares. The results proved to be largely consistent with those reported in Table 3. 9 the primary enrollment ratio in 1960, other things being equal, leads to a 0.26 percentage point gain in the rate of econonic growth over the period 1960 to 1985.9 Table 3: Relating Per Capita GDP Growth (1960-85) to Selected Country Characteristics Country characteristics Low-income Middle-income High-income countries countries countries Primary enrollment rate, 1960 0.033 *** 0.031 * Secondary enrollment rate, 1960 0.034 0.070 *** -0.008 Tertiary enrollment rate, 1960 - 0.129 -0.100 0.062 ** Per capita GDP relative to US, - 0.085 - 0.072 - 0.021 ** 1960 0.064 * 0.085 * 0.073 ** Average investment/GDP, - 0.175 0.490 - 0.670 ** 1960-85 Growth of population, 1960-85 - 0.002 0.029 0.024 Constant Number of observations 74 19 20 Pseudo R2 0.273 0.546 0.493 Source: median regressions based on data from World Bank (1993). Note: one star (*) denotes statistical significance at the 10% level; two stars (**) at the 5% level; and three stars ***) at the 1% lcvel; full regression results are available on request from the authors. In the sub-sample regressions, the coefficient of primary education is positive and statistically significant in both low and middle income countries, with a comparable magnitude of 0.33 and 0.31 respectively.10 In other words, a 10 percentage point advantage in the 1960 primary enrollment ratio would boost the annual growth rate of per capita GDP between 1960 and 1985 by an average of 0.32 percentage points in these countries. Thus, if we compare two countries starting from the same initial per capita income in 1960 but with one of them having a 10 percentage point advantage in the initial primary school enrollment rate, their average per capita GDP in 1985 would differ by about 8 percent. 9 One potcntial source of bias in the estimate arises from the omission of a mcasure for educational quality in the regression. If countries with vider coverage also tendcd to be the ones wvith a stronger commitment to educational quality, the coefficicnt on the education variable would overestimate the impact of educational expansion. On the other hand, if those countries are also the oIIes operating under constrained public spending on education, the bias would work in the opposite direction. 10 In the high income group, all countiries had already achieved universal primary education in 1960. making it meaningless to include the cnrollmiienit ratio in primary education as a regressor. 10 The coefficient on the variable is statistically significant at the 1 percent level in the low income sub- sample whereas it is so only at the 10 percent level in the middle-income group; we interpret the difference to mean that a larger share of countries in the first group conforms more closely to the overall pattern of a positive relationship between initial coverage of primary education and subsequent economic growth. The impact of secondary education is less consistent across the three groups of countries. The coefficient is positive and statistically significant only in middle-income countries. In low-income countries, it is positive but not statistically significant, implying that although expanding secondary education tends, on average, to boost subsequent economic growth, the impact is highly variable across countries. For the high-income group of countries, the enrollment ratio in secondary education in 1960 has had no impact on economic growth during 1960-85. With regard to higher education, the coefficient is positive and statistically significant only in the high income group. For the other two groups of countries, it is negative but statistically insignificant, making it safe to conclude that higher levels of initial investments at this level of education would have had, on average, no impact on economic growth during 1960-85. Summarizing the regression results, we indicate in Table 4 our interpretation of their implication regarding the impact of initial enrollment ratios and subsequent economic growth across the three country groups. We base our judgment on the magnitude of the regression coefficient as well as the level of its statistical significance. The symbol "0" indicates that there is basically little impact, while the symbol "+" indicates a positive impact, with the strength of the impact indicated by the number of times the symbol is repeated. It is important to recall that the regression coefficients capture marginal effects. In other words, they do not reflect the absolute impact of education investments, but rather the effect of expansion in coverage beyond the averages observed in 1960 in the three groups of countries. Table 4: Qualitative Interpretation of the Impact of Initial Education Levels on Subsequent Economic Growth During 1960-85 Level of education Low income Middle income High income Primary ... ++ 0 Sccondary + +++ 0 I-lighcr 0 0 ++ Source: authors construction based on regression results in table 3. 11 Taken as a whole the results suggest that education indeed yields tangible economic benefits for society."I The evidence is consistent with the idea that education produces valuable human capitaL and not merely "pieces of paper" whose only use is to help employers select prospective employees. Analyses based on individual data may be open to this interpretation because people with more education may be selected for their innate ability. Aggregate country-level data are free from this possible difficulty because it is unlikely for countries to vary systematically in the average innate ability of their workers. 12 The regression coefficients on the education variables do differ, however, in magnitude and statistical significance, according to country context, suggesting that not all education investments had been equally productive. The lower the per capita income of the country in 1960 the more productive was expansion of primary education. It was only in the middle income countries that increased coverage at the secondary level made a difference to growth. Expanding coverage in higher education-again from the average level observed in the sample-mattered for economic growth only in the richest countries. One explanation for the differences across country settings is that the benefits from education are effectively realized largely when opportunities exist for exploiting the potential advantages of having skilled labor. In low-income economies, for example, production technology, economic processes, and commercial and legal institutions are relatively simple and opportunities for specialization of labor are probably limited. These conditions may explain why an increase in the availability of highly educated people in such economies would have had little or no impact on aggregate economic performance. The foregoing results are interesting in themselves, but they bring us only part of the way toward identifying what would have been in retrospect socially efficient investments in education the context faced by countries in the 1960s. For this purpose we also need to take account of the cost of the investments. We compute the full rates of return to education below, using the foregoing regression results and incorporating data on the direct and forgone cost of education. Although our calculation conforms more closely to the theoretical definition of a social rate of return, we eschew the term in favor of "full rate of return" because it has by sheer force of habit acquired a somewhat ambiguous meaning in the empirical literature. Estimating the "Full" Returns to Investments in Education We calculate the returns to education using the full discounting method which involves setting up the relevant benefit and cost streams over an assumed investment life-time (Psacharopoulos, 1994). 11 The omission of such variables as initial income equity and openness of the economy may have biased estimates of the regression coefficients. Below we recognize this possibility explicitly by performing the rate-of- return calculations under altemative assumptions about the magnitude of the regression coefficients. 12 See Weale (1993) for an elaboration of the biases to which analyses based on individual data are subjected; and Ram (1996) for a discussion of how these difficulties are addressed to a large extent by the use of aggregate country-level data. Within-country aggregate data can presumably also be used to overcome these problems, but analysis of the data may be hampered by small sample sizes. See, however, Lau et. al (1993) for an analysis of the relation between education and growth using state-level data from Brazil. 12 For our purpose, we specify a life-time of 25 years, matching the duration of our data on growth performance. For each country group we construct the benefit stream by level of education following these steps. First, based on the observed average rates of growth in per capita income and population, we project a stream of aggregate GDP over 25 years indexed on a arbitrary base. Next, we project what the stream would have been had the per capita income growth during the period been hastened by a 1 percentage point increase in the relevant initial enrollrment ratio in 1960.13 The difference in GDP between the two projections gives the benefit from expanding by 1 percentage point the education system's coverage at the relevant level. Regarding the cost stream, we note that three types of costs arise with the 1 percentage point increase in initial enrollment ratio: public direct costs, private direct costs, and forgone earnings or production. Data on the magnitude of each of these components are scanty, but we can obtain rough estimates by weaving together several pieces of information. We make separate estimates for each of our country groups in order to allow for well-documented differences among them in this regard. On the public direct cost of expanding education coverage, we base our estimate on the following information: (i) public spending on education as a share of GDP; (ii) the distribution of public spending by level of education; and (iii) the initial enrollment ratio.'4 In low-income countries, for example, public spending on education averages about 2.8 percent of GDP, with about 45 percent going to primary education, 28 percent going to secondary education, and the remaining 27 going to higher education. Given that primary enrollment ratios in low-income countries averaged 58.2 percent in 1960, it would cost an extra 0.022 percent of the prevailing GDP to sustain a one percentage point increase in the 1960 level of primary education coverage. With regard to the magnitude of private spending on education, we have assumed on the basis of a variety of ad-hoc evidence for such countries as the Philippines, Sri-Lanka, India, Indonesia, Kenya, and Venezuela, that it averages about 20 percent of total government and private spending at the primary level, and about 30 percent at the secondary and tertiary levels (World Bank 1995). Forgone earnings or production, the third component of cost, refer to what students at each level of education forgo while they are in school. We obtain estimates on the basis of the following assumptions and data. For investments in higher education, the forgone costs correspond to the earnings of those with secondary education. We assume that secondary school leavers earn about the same wages as primary school teachers, on the observation that in most countries during the sixties teachers at this level tend to be recruited with secondary education. The earnings of primary school teachers in high-income countries (the only country group for which we will compute the returns to higher education) are about 1.8 times the per capita GNP.'5 To make the cost-benefit calculations we 13 The calculation is based on the regression coefficients reported in Table 3. 14 UNESCO (1993) provides data on the first two indicators, while our regression data set provides information on enrollment ratios. 15 This estimate relies on data for individual countries reported in Mingat (1995) and ILO (1995). In the presence of externalities in production, average wages-which reflect the returns only to private individuals-may underestimate the forgone earnings associated with education investments. While we do not correct for this possible bias directly, we take account of it by performing sensitivity tests on the baseline forgone costs assumed in the rate of return calculations below. 13 need to generate a stream of forgone cost based on this information. We do so by noting that the annual forgone costs are roughly 2.4 times the annual direct public cost per student in higher education in high-income countries.'6 Applying this ratio to the stream of direct costs already calculated above, we obtain the requisite stream of forgone costs associated with expanding higher education coverage. For investments in secondary education, the forgone costs correspond to the earnings of people with primaiy education. We assume that the earnings of unskilled laborers provide a reasonable estimate. According to ILO (1995), they range from 1.43 times the per capita GNP in low income countries, to 0.66 times in the high-income group. Following the same boot-strap procedure as for higher education, we derive the stream of forgone costs by multiplying the corresponding data on direct costs by a ratio of 2.0 in low-income countries and a ratio of 2.2 in middle-income countries, the two groups for which the calculations are relevant.'7 Finally, for investments in primary education, forgone earnings refer to the work production of young children. We assume that they average about 0.25 times the per capita GNP (rising from 0 in the first grade to 0.5 times in the fifth grade), which imply that they are about 2.0 times the direct public cost of education at this level. As before we multiply the stream of direct public cost by this ratio to obtain the desired stream of forgone costs. Based on the foregoing assumptions on benefits and costs we compute the "full" rate of return to education by level of education for each of the three country groups. The results appear in Table 5 as the base case. We also report a range in the estimates corresponding to a 25 percent variation in the assumed opportunity cost. We focus on this item of cost for two reasons: it is subject to greater uncertainty than the other two items; and it is the single most important component of cost. The table also shows how the estimates of returns would shift according to a one standard error change in the regression coefficients on which the benefit streams are based. For context, the table also shows the average enrollment ratio by level of education in the three groups of countries in 1960. Before discussing our interpretation of the findings, we note two important points. First, the retums calculated in Table 5 correspond to the historical profitability of investing in education over the last three decades; and second, the returns reflect the marginal profitability of expanding investments in education (i.e., the economic gain from increasing the education system's coverage from the levels prevailing in 1960). If the returns to a particular level is negative or low, we interpret this to mean that new investments to increase coverage were in retrospect probably not economically justified. Strictly speaking, a negative rate of return in fact implies that it would have been economically efficient to reduce coverage. Recall that coverage (i.e., enrollment ratios) is defined in our analysis as the number of students relative to the population in the relevant age group. In the context of growing populations, a reduction in coverage would not necessarily imply an absolute contraction in the size of any part of the education system, only that it grows more slowly than the relevant population group. 16 The ratio is based on data for 1993 from UNESCO (1995). In our cost-benefit calculation all the data are measured in relational rather than absolute units. This justifies using current rather than historical cost data. There is also the added advantage that recent data are more plentiful. 17 As before, the ratio between forgone costs and direct public cost per student is based on data on unit cost reported in UNESCO (1995). 14 Table 5: Estimated FuOl Rates of Return by Level of Education and Country Group (%) Low-income Middle-income High-income Level of education group group group (74 countries) (19 countries) (20 countries) Primary: Base Case 47 39 25% change in opportunity cost 41 -- 55 34 -- 46 1 s.e. change in regression coefficient 32 -- 61 21 - 56 Secondary: Base Case 8 52 < 0 25% change in opportunity cost 5 -- 12 45 -- 61 1 s.e. change in regression coefficient <0 -- 23 37 -- 66 Higher education: Base Case < 0 < 0 20 25% change in opportunity cost 17 -- 25 1 s.e. change in regression coefficient 7 -- 30 Average enrollment ratio in 1960 (%)a Primary 58.2 95.0 109.3 Secondary 9.4 28.4 53.1 Higher 1.2 4.4 9.4 Source: authors' calculation based on assunptions discussed in text. Data on enrollment ratios in 1960 correspond to the averages for countries included in the regression analysis for the indicated sub-samples. a/ Enrollment ratios are defined as the number of pupils enrolled at a given level of education relative to the population in the age group which officially corresponds to that level of schooling. The presence of over-age students makes it possible for enrollment ratios to exceed 100 percent. 15 Bearing in mind the clarifications above, we may summarize the pattern of returns by level of education as follows: * For low-income countries, investing in pfimary education proved to be the most sociatly profitable of the three levels of education. In such countries, productive processes and economic transactions are relatively simple and primary education probably supplies the skills for most jobs (even though the curriculum typically does not have an explicit technical content). We also computed a rate of return for secondary education despite the lack of statistical significance in the regression coefficient on which it is based. The resulting rate of return is positive but low, implying that expanding secondary education (beyond the level observed in 1960 in the sample) in such countries did not, in retrospect, prove to be a priority. For higher education, we made no estimate of the returns because the regression coefficient was negative, which in itself would have implied a negative social rate of return; moreover the costs for this level of education, both in direct spending and forgone production, are typically quite substantial in low-income countries. It is interesting to note in passing that East Asian countries' investment strategies in education during the sixties and seventies were consistent with the implications of these patterns in the returns (Mingat, 1995). Whereas primary education was heavily subsidized to stimulate rapid increases in coverage at this level higher education (and to a lesser extent, secondary education) was financed largely by private resources which tended to restrain the growth in coverage at this level. * For middle-income countries, expanding primary education continued to remain socially profitable; this result is obviously relevant only for countries where full coverage had still not been reached by 1960. But expansion of secondary education would have yielded, in hindsight, the highest social returns. Thus, whereas the economic justification for expanding secondary education was weak in low-income settings, it became strong in the context of middle-income countries.18 The high returns suggest that investing in this level of education offered the most appropriate support for boosting economic growth in countries at this stage of development. Investing in higher education did not prove to be socially efficient. * For high-income countries, the returns turned out to be highest for expanding higher education. Investing to boost coverage at this level by 1 percentage point would have yielded an estimated social return of 20 percent a year, based on the pattem of countries' economic performance during 1960-85.19 Primary schooling was already universal by 1960, so there was no basis for computing social rates of return to this level of education. For secondary education, the benefits from expanding coverage were insufficient to outweigh the costs of expansion, leading to negative returns. As indicated before, investing to maintain coverage in primary and secondary education- averaging 100 and 50 percent respectively in 1960-continue to be justified, not least because these 18 Of course countries could still have placed the highest priority on universalizing primary education. The choice would have been valid for equity reasons only, however, since the full economic returns to primary education are exceeded by those to secondary education. 19 As a matter of interest, we note that our result is close to Becker's (1975) upper bound estimate of 25 percent for the social rate of return to college education in the United States. His calculation was based on evidence from the sources-of-economic-growth literature. 16 levels provide the pool of candidates for higher education. But given the impact on the growth process and the magnitude of costs, new investments to expand coverage in higher education would have offered society the best returns. Comparing the Results with the Standard Estimates of Social Returns The results presented above differ from the standard estimates reported in the literature in two respects: (i) the structure of returns across levels of education and country income; and (ii) the size of the estimates. The differences arise for two possible reasons. First, the standard estimates typically rely on earnings data for individuals, and these data may not provide accurate measures of costs and effective social benefits, particularly in low- and middle-income countries where wages are often influenced to a large extent by factors (e.g., social institutions) other than market forces. Second, the earnings data refer to a cross-section of people who are already employed. Because of possible rigidity in wages arising from legal and other constraints, the observed structure of earnings may not capture the relative scarcity of new entrants to the labor market. Our estimates escape both these flaws (although they may suffer from other shortcomings) because they relate to aggregate rather than individual data, and because they relate to actual growth performance of economies over the entire period for which the calculations are made. For the comparison of results, recall that conventional estimates reveal a systematic structure in which the returns are typically highest for primary education, followed by secondary education, and then higher education. At all three levels, the returns exceed the traditional 10 percent cut-off rate for the cost of capital. Our results suggest that the pattern of returns is sensitive to the economic context in which the investments are made. The returns to investing in primary education were highest in low- income countries. They remain high in rniddle-income countries, but were exceeded by the social returns to investing in secondary education in such settings. In the context of the most advanced economies, higher education offered the best social returns of the three levels. With regard to the size of social returns to investing in education, we may summarize the differences between our estimates and those reported in the literature as follows: where our results suggest a strong economic justification for investing, the estimated magnitude of the social returns was, on average, about twice that of conventional estimates; but where they suggest a weak justification, our estimates were much lower than the conventional estimates. If we accept that aggregate growth performance is a more complete measure of the economic productivity of investment in education, encompassing benefits for individuals as well as for society at large, our estimates suggest that the size of externalities can be very large indeed under certain conditions.20 It is common practice in policy analysis to invoke the existence of externalities as a blanket argument for justifying investments in education. Our results suggest that the argument is not universally applicable because the existence of externalities is highly sensitive to the context in which the investments are made. 20 That the returns to schooling is sensitive to the economic context is consistent with Schultz's (1975) observation that schooling is more valuable where opporturities exist for exploiting the advantages of being more educated (or having more educated people). Recent work by Rosenzweig (1996) among others using micro level data provide additional support for this expectation. 17 Conclusion We have attempted in this paper to estimate the 'flill" economic returns to education by level, with benefits defined to include externalities captured in economic growth performance. Because we use country-level data to compute the returns, the estimates refer to the "macro" returns to education. Given the cross-country perspective of our analysis, limitations in the data, and issues regarding model specification, the results remain tentative. They nonetheless go beyond the conventional literature which by and large ignores externalities altogether. Our findings support the view that investments in human capital through formal education can produce high economic returns to society. Our results thus find no support for the view that education is useful only for the purpose of screening individuals by ability. But they also suggest that the impact of education investments on aggregate economic productivity depends critically on the context in which they take place. What implications for policy emerge from the results? The fact that our results rely on historical data suggests obvious caution in using them to set investment priorities in the context that countries face today. The process of economic development is nonetheless relatively stable, so experience from the past, especially with regard to such a basic investment as education, may continue to offer useful insights. To support economic growth a socially productive investment strategy in education would place more emphasis on expanding basic education at early stages of development, shifting the focus of expansion to secondary and higher education only as the economy matures. References Alesinia, Alberto, and Dani Rodrik. 1994. "Distributive politics and economic growth," Quarterly Journal of Economics, 109 (2): 465-490. Azariadis, Costas, and Allan Drazen. 1990. "Threshold externalities in economic development," Quarterly Journal of Economics, 105 (2): 501-26. Barro, Robert. 1991. "Economic growth in a cross-section of countries," Quarterly Journal of Economics, 106 (2): 407-43. Becker, Gary. S. 1975. Human Capital. A Theoretical and Empirical Analysis with Special Reference to Education. Second Edition. Chicago: University of Chicago Press. Becker, S. Gary; and Barry R. Chiswick. 1966. "Education and the distribution of earnings," American Economic Review, 56 (May): 358-69. Benhabib, Jess, and Mark M. Spiegel. 1994. "The role of human capital in economic development: evidence from aggregate cross-country data," Journal of Monetary Economics, 34 (1): 143-173. Birdsall, Nancy. 1996. "Public spending on higher education in developing countries: too much or too little?" Economics of Education Review, 15 (4). DeLong, J. Bradford, and Lawrence H. Summers. 1991. "Equipment investment and economic growth," Quarterly Journal of Economics, 106 (2): 445-502. Dollar, David. 1992. "Outward-oriented developing countries do really grow more rapidly: Evidence from 95 LDCs ,1976-85," Economic Development and Cultural Change, 40 (3): 523-44. Haveman, Robert, and Barbara Wolfe. 1984. "Schooling and economic well-being: the role of non-market effects," Journal of Human Resources, 19 (3): 377-407. - 1995. Succeeding Generations. On the Effects of Investments in Children. New York: Russell Sage Foundation. ILO. 1995. Statistics on Occupational Wages and Hours of Work and on Food Prices, Geneva: International Labour Office. Lau, Lawrence J., Dean T. Jamison, Shu-Cheng Liu, and Steven Rivkin. 1993. "Education and economic growth. Some cross-sectional evidence from Brazil," Journal of Development Economics, 41 (1): 45-70. Lucas, Robert E. Jr. 1988. "On the mechanics of economic development," Journal of Monetary Economics 22 (1): 3-42. 19 Mankiw, N. Gregory, David Romer, and David N. Weil. 1992. "A contribution to the empirics of economic growth," Ouarterly Journal of Economics, 106 (2): 407-37. Mincer, Jacob. 1974. Schooling. Experience, and Earnings. New York: National Bureau of Economic Research. Mingat, Alain. 1995. "Towards improving our understanding of the strategy of high-performing Asian economies in the education sector," paper presented at the Asian Development Bank Conference on the Financing of Human Resources in Asia, November 17-18, 1996, Manila, Philippines. Olson, Mancur. 1996. "Big bills left on the sidewalk: why some nations are rich, and others poor," Journal of Economic Perspectives, 10 (2): 3-24. Patrinos, Harry A. 1994. "Notes on education and economic growth: theory and evidence," HRO Working Papers, HROWP39, Human Resources Development and Operations Policy, World Bank, Washington D.C.: The World Bank. Pritchett, Lant. 1996. "Where has all the education gone?" Policy Research Working Paper 1581, Policy Research Department, World Bank, Washington D.C.: The World Bank. Psacharopoulos, George. 1994. "Returns to investment in education: a global update," World Development, 22 (9): 1325-1343. Psacharopoulos, George. 1996. "Commentary on 'Public spending on higher education in developing countries: too much or too little"', Economics of Education Review, 15 (4) Ram, Rati. 1996. "Level of development and rates of return to schooling: some estimates from multicountry data," Economic Development and Cultural Change, 44 (4): 839-55. Rosen, Sherwin. 1987. "Human Capital," in The New Palgrave: A Dictionary of Economics, eds. John Eatwell, Murray Milgate, and Peter Newman. New York: Stockton. Rosenzweig, Mark R. 1995. "Why are there returns to schooling?" American Economic Review. Papers and Proceedings 85 (2): 153-158. Schultz, T. Paul. 1988. "Education investments and returns," Handbook of Development Economics, Vol. 1: 544-621. Schultz, T.W. 1975. "The value of the ability to deal with disequilibria," Journal of Economic Literature, 13 (3): 827-47. Summers, Lawrence. 1994. Investing in all the people, EDI Paper No. 45, Washington D.C.: The World Bank. Summers, Robert; and Alan Heston. 1991. "The Penn World Tables (Mark 5): an expanded set of international comparison, 1950-88." Quarterly Journal of Economics, 105 (2): 327-68. 20 UNESCO. 1995. World Education Indicators. Paris: United Nations Educational, Scientific, and Cultural Organization. - - 1968. Statistical Yearbook. Paris: United Nations Educational, Scientific, and Cultural Organization. Willis, Robert. 1986. "Wage determinants: a survey and reinterpretation of human capital earnings function," in Handbook of Labor Economics, eds. Orley Ashenfelter and Richard Layard. New York: North-Holland. Weale, Martin. 1993. "A critical evaluation of rate of return analysis," The Economic Journal, 103 (May): 729-37. World Bank. 1993. The East Asia Miracle. Economic Growth and Public Policy, Washington D.C.: Oxford University Press for the World Bank. - 1995. Priorities and Strategies for Education. A World Bank Review. Washington D.C.: The World Bank. Human Capital Development Working Paper Series Contact for Title Author Date paper HROWP43 Public and Private Secondary Marlaine E. Lockheed November 1994 M. Verbeeck Schools in Developing Emmanuel Jimenez 34821 Countries HROWP44 Integrated Approaches to T. Paul Schultz November 1994 M. Espinosa Human Resource Development 37599 HROWP45 The Costs of Discrimination in Harry Anthony Patrinos November 1994 I. Conachy Latin America 33669 HROWP46 Physician Behavioral Nguyen X. Nguyen December 1994 M. Espinosa Response to Price Control 37599 HROWP47 Evaluation of Integrated T. Paul Schultz January 1995 M. Espinosa Human Resource Programs 37599 HROWP48 Cost-Effectiveness and Health Philip Musgrove January 1995 0. Shoffner Sector Reform 37023 HROWP49 Egypt: Recent Changes in Susan H. Cochrane February 1995 0. Shoffner Population Growth Ernest E. Massiah 37023 HROWP50 Literacy and Primary Kowsar P. Chowdhury February 1995 M. Espinosa Education 37599 HROWP51 Incentives and Provider Howard Barnum March 1995 0. Shoffner Payment Methods Joseph Kutzin 37023 Helen Saxenian HROWP52 Human Capital and Poverty Gary S. Becker March 1995 M. Espinosa Alleviation 37599 HROWP53 Technology, Development, and Carl Dahlman April 1995 M. Espinosa the Role of the World Bank 37599 HROWP54 International Migration: Sharon Stanton Russell May 1995 0. Shoffner Implications for the World 37023 Bank HROWP55 Swimming Against the Tide: Nancy Birdsall May 1995 A. Colbert Strategies for Improving Equity Robert Hecht 34479 in Health HROWP56 Child Labor: Issues, Causes Faraaz Siddiqi June 1995 I Conachy and Interventions Harry Anthony Patrinos 33669 HCOWP57 A Successful Approach to Roberto Gonzales July 1995 K. Schrader Partcipation: The World Bank's Cofino 82736 Relationship with South Africa HCOWP58 Protecting the Poor During K. Subbarao July 1995 K. Labrie Adjustment and Transitions Jeanine Braithwaite 31001 Jyotsna Jalan HCOWP59 Mismatch of Need, Demand Philip Musgrove August 1995 Y. Attkins and Supply of Services: 35558 Picturing Different Ways Health Systems can go Wrong HCOWP60 An Incomplete Educational Armando Montenegro August 1995 M. Bennet Reform: The Case of 80086 Human Capital Development Working Paper Series Contact for Title Author Date paper Colombia HCOWP61 Education with and with out the Edwin G. West September, 1995 M. Espinosa State. 37599 HCOWP62 Interactive Technology and Michael Crawford October 1995 P. Warrick Electronic Networks in Higher Thomas Eisemon 34181 Education and Research: Lauritz Holm-Nielsen Issues & Innovations HCOWP63 The Profitability of Investment George Psacharopoulos December 1995 M. Espinosa in Education: Concepts and 37599 Methods HCDWP64 Education Vouchers in Practice Edwin G. West February 1996 M. Espinosa and Principle: A World Survey 37599 HCDWP65 Is There a Case for Antonio Zabalza March 1996 M. Espinosa Government Intervention in 37599 Training? HCDWP66 Voucher Program for Alberto Calder6n Z. May 1996 M. Espinosa Secondary Schools: The 37599 Colombian Experience HCDWP67 NGO-World Bank Toshiko Hino June 1996 A. Thomas Partnerships: A Tale of Two 31151 Projects HCDWP68 The Disability-Adjusted Life Nuria Homedes July 1996 L. Arias Year (DALY): Definition, 35743 Measurement and Potential Use HCDWP69 Equitable Allocation of Ceilings Philip Musgrove August 1996 Y. Attkins on Public Investment: A 35558 General Formula and a Brazilian example in the Health Sector HCDWP70 The Economics of Language: Barry Chiswick September 1996 I Conachy The Roles of Education and 33669 Labor Market Outcomes HCDWP71 Agricultural Growth and Rashid Faruqee September 1996 C. Anbiah Poverty in Pakistan Kevin Carey 81275 HCDWP72 Measuring the Opportunity Andrew D. Mason September 1996 D. Ballantyne Cost of Children's Time in a Shahidur R. Khandker 87198 Developing Country: Implications for Education Sector Analysis and Interventions HCDWP73 The Full Social Returns to Alain Mingat September 1996 J. Yang Education: Estimates Based Jee-Peng Tan 81418 on Countries' Economic Growth Performance