WP5 sk,-1g POLICY RESEARCH WORKING PAPER 2015 The Mechanics of Progress Ascountriesgrowrich, education improves in many in Education ways. The sector enjoys more resources for education per Evidence from Cross-Country Data school-aged child, not Evidence from Cross-Country Data primarily because of bigger budget allocations or an Alain Mingat easing of the demographic Jee-Peng Tan burden on the system. but because teacher salaries decline substantially relative to the per capita GNP. The extra resources enable countries to expand coverage and reduce the pupil-teacher ratio, with the latter receiving increasing emphasis during the past 20 years. The implicit tradeoff against coverage is inefficient (and inequitable), particularly in countries where large portions of the population still have no access to basic education. The World Bank Human Development Department Education Group November 1998 | POLICY RESEARCH WORKING PAPER 2015 Summary findings Mingat and Tan explore differences in education in rich their education systems. But bigger budget allocations to and poor countries by first systematically documenting education contribute relatively little to differences in the relationship between per capita GNP and various resources. indicators of educational development. They then exploit Lighter demographic burdens in richer countries is also a simple accounting identity relating the availability of a relatively modest factor. resources to their expenditure, to clarify the sources of By far the most important factor is the decline of rich countries' advantage in education. teacher salaries relative to per capita GNP, which Data for a sample of 125 countries in 1993 confirm accounts for at least half of any educational advantage at the expected favorable relationship bet-ween per capita all stages of economic development. GNP and each of the following dimensions of The extra resources for education associated with educational development: income growth allows a country to expand enrollments * The sector context (as reflected by the demographic and improve classroom conditions by reducing the pupil- burden on the education system, the government's fiscal teacher ratio. Early in income growth, countries allocate capacity, and so on). more of the extra resources to expand coverage; later * The production of education services, including they shift toward reducing the pupil-teacher ratio. such factors as public spending on education and the But, contend the authors, so long as coverage is not yet composition of spending. universal a more efficient strategy for educational - Education outcomes, in terms of coverage and development is to emphasize continued expansion of student learning. coverage rather than a rapid reduction in the pupil- * Efficiency of sector operations. teacher ratio. In the long run, lower levels of educational - Equity in access and distribution of public spending attainment among tomorrow's adults is likely to diminish on education. learning achievement among tomorrow's children. One appealing explanation for why richer countries achieve better results is that they have more resources for This paper - a product of the Education Group, Human Development Department - is part of a larger effort in the department to understand the nature of educational development. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Anahit Poghosyan, room G8-064, telephone 202-473- 0898, fax 202-522-3233, Internet address apoghosyan@worldbank.org. The authors may be contacted at amingat@u- bourgogne.fr or jtan@worldbank.org. November 1998. (46 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center The Mechanics of Progress in Education: Evidence from Cross-Country Data By Alain Mingat Institut de Recherche sur l¶Economie de IlEducation Centre National de la Recherche Scientifique Uiversite de Bourgogne Dijon France And Jee-Peng Tan Human Development Network Education Department The World Bank Contents Executive Summary 1. Introduction 1 2. How Is Education Different In Rich and Poor Countries? 2 2.1 Overall sector context 3 2.2 Production of education services 6 2.3 Education outcomes in quantity and quality 11 2.4 Efficiency of the education system 13 2.5 Equity in access and resource allocation 16 2.6 Summary of the relation between country wealth and education 19 3. What Are the Sources of Rich Countries' Advantage in Education? 20 3.1 A framework for the analysis 21 3.2 Rich countries' advantage in overall educational development 22 3.3 Rich countries' advantage in primary education 27 3.4 Summary of structural differences in education between rich and poor countries 36 4. What Are The Policy Issues? 37 4.1 Expansion of coverage versus lowering pupil-teacher ratios 37 4.2 The scope for policy choice at the country-level 40 5. Conclusion 42 Appendix A: Comparing Education in Rich and Poor Countries in 1993 and 1975 44 Executive Summary A staple feature of cross-country comparisons is that poor countries typically lag behind rich countries on almost any indicators of educational development. What in fact is the relation between country wealth and education? What underlying factors account for rich countries' advantage in education? How does the impact of these factors change in the course of economic growth? What scope is there for policy choice to improve education sector outcomes? We explore these questions by first documenting the relation between per capita GNP and various indicators of educational development in a systematic fashion. We then exploit a simple accounting identity relating the availability of resources to the expenditure of those resources to clarify the sources of rich countries' advantage in education. Data for our sample of some 125 countries in 1993 confirm the expected favorable relation between per capita GNP and each of the following dimensions of educational development: (a) the sector context as characterized by the demographic burden on the education system, the government's overall fiscal capacity, and so on; (b) the production of education services, including such factors as public spending on education and the composition of spending; (c) education outcomes in terms of coverage and student learning; (d) efficiency of sector operations; and (e) equity in access and distribution of public spending on education. An appealing explanation of why richer countries achieve better results is that they have more resources to run well-functioning and efficient education systems. Comparing pairs of countries at contiguous levels of per capita GNP between $200 and $10,000 (e.g. countries at $200 and $400 would form one pair, and those at $400 and $800, another pair), we found that indeed the richer country in each pair enjoys an advantage of almost 50 percent in the resources available per school-aged child. Yet a closer look reveals that bigger budget allocations to education make a relatively small contribution to the differences in resources. The commonly-cited role of lighter demographic burdens in the richer countries is also relatively modest, contributing between 17 and 32 percent of the advantage among the pairs of countries compared. By far the most important factor is the decline of teacher salaries relative to the per capita GNP, accounting for at least 50 percent of the advantage at all stages of economic development. The extra resources per school-aged child that materializes in the process of income growth allow countries to expand enrollments and improve classroom conditions by reducing the pupil-teacher ratio. At the earliest phase of income growth-corresponding to a rise in the per capita GNP from $200 to $400-countries allocate more of the extra resources to support expansion of coverage, but at all subsequent stages, the emphasis shifts in favor of reducing the pupil-teacher ratio. In both the 1970s and the 1990s the shift in emphasis away from expansion of coverage took place in contexts where access to primary education was still not yet universal. Moreover, it occurred at significantly lower enrollment rates-around 50 percent-in the 1990s than in the 1970s, signaling a stronger bias against expansion of coverage in the more recent period. The revealed priorities in spending raise important policy questions. Is an emphasis on reduction of the pupil-teacher ratio at the cost of slower progress in expanding coverage an efficient strategy for the sector? Two considerations suggest an answer in the negative: the first is that over the range that the pupil-teacher ratio typically varies in developing countries, smaller pupil-teacher ratios have little or no impact on student learning; moreover, the slower progress in expanding coverage implies that tomorrow's adults will be less well-educated which in turn is likely to diminish student learning among tomorrow's children, given that adults' educational attainment is a strong predictor of children's school performance. Thus, as long as coverage is not yet universal a more efficient strategy for educational development is to emphasize continued expansion of coverage rather than a rapid reduction in the pupil-teacher ratio. Aside from the systematic differences in education across rich and poor countries, the study also finds evidence of substantial diversity among countries at comparable levels of per capita GNP. Countries set different priorities in the allocation of public spending, pursue different policies that affect how education services are organized and delivered, and make different tradeoffs as to how resources are used to support expansion of coverage and reduction in the pupil-teacher ratio. In countries where education is poorly developed, policy choices that affect the market for teachers (and therefore their cost), as well as the balance between expansion of coverage and reduction in the pupil-teacher ratio have especially strong effects on the prospects for progress in education. The Mechanics of Progress in Education: Evidence from Cross-Country Data 1. Introduction In all countries education plays a key role in social and economic life, so it is not surprising that most governments seek, at least in rhetoric, to build education systems that offer the best possible services to as wide a segment of the population as possible. Despite the common goal, however, countries achieve very different results. On almost any measure of schooling outcome-whether related to coverage or student learning-low-income countries typically lag behind high-income countries. What are the sources of rich countries' advantage in education? What is the role of such factors as demographic conditions, government fiscal capacity, prices, and policy choice? Does the impact of these factors change as countries grow rich, and if so, how? What do the patterns reveal about the nature of policy choice in education? Answers to these questions are explored in this paper. In the process we hope to discover patterns in the relation between country wealth and various aspects of educational development, and to clarify the mechanics of progress in education as a country grows rich.' At the same time, we also expect to improve our understanding of policy options to improve education. Given the breadth of these issues, our study represents a partial treatment, both because it focuses on system-wide features rather than on pedagogical processes at the classroom level, and because it relies solely on quantitative data.2 These limitations notwithstanding the findings reveal several key insights. The first is that systematic links exist between country wealth and most indicators of educational 'For a discussion of similar questions from the perspective of a production-demand framework based on data from 1960 to 1980 see Schultz, T.P. 1988 "Expansion of public school expenditures and enrollments: inter-country evidence on the effects of income, prices, and population growth," Economics of Education Review 7(2): 167-83. 2 The bulk of the data pertains to 125 countries around 1993, the latest year for which data on the most of the indicators used in our study are currently available or can be constructed, supplemented by data for 1975 for a smaller set of indicators and countries. These data come from a larger time-series database prepared by the Institute de Recherche sur l'Economie de l'Education (IREDU), Universite de Bourgogne, Dijon, France. The raw data come from various sources, including international organizations (e.g. UNESCO, the World Bank, ILO, and OECD), published and unpublished country and comparative studies, and doctoral dissertations. Data for all the indicators have been checked for consistency in their relation to each other, as well as across time. Where published data contain obvious inconsistencies they are replaced by data for a nearby year, or by simulations based on related data. Since the database was created, new data have become available, but rather than attempt to incorporate them we have decided simply to use what is already available in it. -2 - development, including steady easing of demographic pressures on the education system as incomes rise, significant expansion, at the primary level of coverage accompanied by a gradual rise in per pupil spending, declines in the pupil-teacher ratio and teacher salaries relative to the per capita GNP. The relation between country wealth and most of the indicators was relatively loose, however, suggesting substantial scope for diversity in policy choice across countries. Exploiting an accounting identity to relate several indicators of expenditure to those on outcomes in primary education, we found that as a country grows rich, the resources for education that become available per school-aged child indeed increase significantly. Strikingly, the main source of the increase is not bigger budget allocations, nor a lighter demographic burden, but the sharp declines in teacher salaries relative to the per capita GNP. Countries use the extra resources to expand primary school coverage and reduce the pupil-teacher ratio, but they have tended to emphasize the latter option, a pattern that has become more obvious in the 1 990s than in the 1970s. In light of the weak links between pupil-teacher ratios and student learning in the context of developing countries, the shift in emphasis raises serious questions about the underlying assumptions of recent policy choices in education, especially in poor countries. The rest of this paper is organized as follows. Section 2 documents the relation between country wealth and educational development in order to identify systematic differences in education between rich and poor countries. Section 3 sets out a framework to clarify the sources of richer country's advantage in education, and to examine how countries allocate the resources that materialize in the process of economic growth to expand coverage and improve schooling conditions. Section 4 discusses the policy issues raised by the findings in the previous two sections, particularly with regard to quantity-quality tradeoffs in resource allocation, and the role of policy choice in shaping progress in the sector. Section 5 concludes the paper. 2. How Is Education Different In Rich And Poor Countries? We make the comparisons along five dimensions of education that together paint a reasonably comprehensive picture of how the sector operates:3 3. For lack of space not only selected indicators of the five dimensions are included below. Supplementary graphs and tables are available upon request from the authors. -3 - * Overall sector context, as described by demographic pressures on the education system, overall fiscal capacity of the government, the educational attainment of the adult population, the structure of employment by economic sector, and the share of education in the public budget; * Production of education services as reflected in the public-private division of service provision, the composition of public spending on education by level and expenditure category, and average public spending per pupil; * Education quantity and quality, as measured by various indicators of coverage, and by student achievement on international mathematics and science tests; * Efficiency of the education system, as captured by the pattern of grade-to-grade student flow, and education outcomes relative to the amount of resources invested to achieve them; and - Equity in education, as revealed by gender disparities in enrollments and by the distribution of public spending on education. 2.1 Overall sector context The demographic composition of the population describes a key feature of the sector context because children and young people depend on adults to finance the system; thus, the more children there are relative to adults, the heavier is the fiscal burden that education places on each adult. The government's fiscal capacity-as reflected by total public spending relative to the GNP-also matters because a bigger overall budget is likely to improve the availability of resources for education, not least because in most countries the government provides most of the funds for education. A third aspect of the sector context is the educational attainment of adults, its relevance stemming from the fact that adults make decisions about children's schooling and they also shape the broader out-of-school learning environment to which children are exposed. Yet a fourth features is the distribution of employment by sector; it matters because it closely mirrors the economy's demand for educated labor, which in turn affects the opportunity cost of schooling and the employment prospects of school leavers. Finally, the intensity of the inter-sector competition for public funds, as reflected in the share of education in overall public spending, forms yet a fifth dimension of the overall context. Below we examine how the foregoing aspects of the sector context relate to country wealth, as proxied by the per capita GNP. -4- Demographic burden and overall fiscal capacitv. As a measure of the demographic burden we express the population aged 6-11 (corresponding roughly to the primary school aged population) as a percentage of the total population4 Figure 1 (panel A) shows that this indicator (which we shall call the dependency ratio) declines steadily as per capita GNP rises, from an estimated average of 23 percent in countries at a per capita income of $200, down to only 14 percent, on average, among OECD countries. The difference is large, because it implies that, all other things being equal, achieving the same coverage in primary education implies twice as heavy a resource burden (relative to the GNP) in the poorest countries as in the richest countries. Figure 1: Relation between demographic constraint and fiscal capacity, and per capita GNP, circa 1993 (A) Demographic comtraint (B) Fiscal constraint 25 0 00 80s 120~ e\< 0 0°0 0 0 0 5, 0 0 ~~~~~~t 00 lo000 0 0 0 00 * 5~~~~~~~~~~~~~~~~~4O~~~~~6 0 000 20 0 0 0 0 0 S 0 0 ° o 0 0 ~~~~00~ 0 0 0 0000 0 I 0' 804 I =0 I 00 200 400 SOO 1500 3000 10000 0 10000 20000 The line in the graph, estimated from regression analysis, describes the average relationship between the dependency ratio and per capita GNP (expressed in logarithmic units).5 The B? statistic for the regression equation is relatively high, at 0.65, suggesting a fairly tight relationship between the two variables. Thus, although the scatter plot suggests variation across countries in the dependency ratio, the broad pattern is that countries generally follow a predictable path of demographic transition as they grow rich. 4TIhe pattr is simila when the numerator includes children in a binader age band (6-14) corresponding to basic education. S We estimated the following equation: dependency ratio = a + I5. In (per capita GNP). + ,u, where oa and ,B are the regression estimates, and 1i is the exmr tern. The value of ff is the change in the dependency ratio associated with a percentage point change in the per capita GNP. ln thin formulation, a change in per capita GNP from, say, $200 to $400 (a 100 percent rise) would have the same impact as a change from, say, $1,000 to $2,000 (which is also a 100 percent rise). -5- Turning now to panel B in the figure, we note that the overall size of the government budget as a percentage of GDP tends to expand with country wealth-from an estimated average of 27 percent among the poorest countries, to 47 percent among the OECD countries. The pattern is consistent with the fact that as countries becomes richer and the economy more formalized, the tax base tends to broaden and tax administration generally becomes more efficient. The positive association is relatively weak, however, as the R2 statistic of the regression equation is only 0.25 for the sample as a whole, and an even smaller 0.04 among countries with incomes no higher than $4,000 per person.6 Thus, in this income group, the size of the public budget relative to the GNP varies almost independently of a country's level of economic development, suggesting that the variable is generally open to policy choice. Adults' educational attainment and the economy's employment structure. To economize on space we examine only the adult literacy rate and the share of workers in agriculture as measures of these aspects of the sector context.7 Data for our sample show that the literacy rate rises with the per capita GNP, from an estimated 50 percent in countries at $200 in per capita GNP in 1993, to nearly universal literacy by the time the per capita GNP exceeds $10,000. With regard to the structure of employment, the graph shows that the share of workers in agriculture drops precipitously as a country grows rich, from an estimated average of 71 percent at $200 in per capita GNP, to 35 percent at $1,500, and then to only about 5 percent at $20,000; correspondingly, employment in industry and services expands as country wealth rises. Both the literacy rate and the employment share of agriculture relate fairly tightly to the per capita GNP, with a R2 statistic of 0.54 and 0.77 respectively; the result implies that countries face relatively similar constraints in these aspects of the context for educational development. Inter-sector competition for public spending. The more intense the competition for resources, the smaller is likely to be the share of education in the public budget. Do rich countries tend to favor education over other sectors more than poorer countries? The evidence suggests no such tendency: rich countries allocate about the same share as poor countries, 6The regression equation is the same as that in the previous footnote, except that it contains an extra terms, the square of hn (per capita GNP). The new item is added to improve the fit of the regression to the data, which improves a higher R2 value. Subsequent regressions use one of these functional forms, and the choice is decided based on the R2 value of the regression. 7 For details on other indicators see Mingat Alain and Jee-Peng Tan (1998) "Education in rich and poor countries: a systematic comparison" mimeo, Human Development Department, The World Bank. -6- ranging between 13 and 17 percent of the total public budget. However, as a share of the GNP, public spending on education rises more noticeably with per capita GNP, mainly because countries' overall fiscal capacity expands with country wealth. Both measures of the availability of resources for education show a fairly weak relation to the per capita GNP, with a R2 statistic of 0.08 and 0.22 respectively. The results confirm the expectation that countries do have considerable leeway in deciding how much to spend on education. 2.2 Production of education services Consider below the following aspects of the production of education services: the public-private provision in education; the distribution of public spending on education; and the level and pattern of public spending per pupil. Public-private roles in service provision. The available data relate only to the share of enrollments in public and private schools at the primary and secondary levels. Because private schools often receive public subsidies and public schools sometimes charge fees, the public-private distinction does not correspond to a clear-cut separation in financing arrangements. The division is more reasonably interpreted as reflecting different arrangements for school management. Over the per capita income range in our data, private schools account for between 10 and 13 percent of all children at the primary level, and between 13 and 18 percent of those at the secondary level. At both levels, the share of private sector enrollments is 8 unrelated to the per capita GNP, a pattern consistent with the findings in James (1993). The role of private education thus reflects the influence of factors other than country wealth, including that of policy choices, that affect the incentives for private sector participation in education. The distribution of public spending on education. Given the government's predominant role in education, we examine in more detail the allocation of spending by level and across pedagogical and non-pedagogical inputs. Table 1 shows simulations of the sub-sector shares of spending based on regression estimates of the relation between the shares and country wealth. In general, the share of primary education decline as countries grow rich, while those of James, Estelle. 1993. "Why do different countries choose a different public-private mix of educational services? "Journal of Human Resources. 28(3): 571-592. James found that religion and language exert a particularly strong influence on the share of private enrollments. In addition, the share is affected positively by the fact that the government subsidizes private education, and negatively by the magnitude of public spending on education. -7 - secondary and higher education rise. The trends are relatively flat, however: the share of primary education, for example, falls only from 45 percent to 31 percent as per capita GNP rises a hundred-fold from $200 to $20,000. Moreover, there is wide variation across countries in the distribution of spending, as indicated by the small R2 statistic for all the regressions (last two columns in the table). The result implies that countries have substantial leeway to set priorities for spending across levels of the education. Table 1: Simulations of the sub-sector share of public spending on education at selected per capita GNP, circa 1993 (%) al Per capita GNP, 1993 $ Sub-sector 200 400 800 1,500 3,000 10,000 20,000 (1) (2) Primary 45.1 43.0 40.8 38.9 36.8 33.1 31.0 0.21 0.02 Secondary 25.8 26.8 27.8 28.7 29.7 31.5 32.5 0.06 0.03 Higher 17.1 17.7 18.4 19.0 19.6 20.7 21.4 0.05 0.05 a! Excludes data for countries where primary and secondary are combined as basic education. Note: The R2corresponds, in column (1), to a regression of the form noted in footnote 5, based on data for the whole sample; and in column (2), to the same regression based on data for countries with a per capita income below $4,000 in 1993. Consider now the allocation of spending by expenditure category. The available data pertain to spending on financial aid for students (in the form of grants, scholarships, bursaries and welfare services); and on pedagogical materials. Figure 2 (panel A) shows that the share of financial aid in total spending on education follows a U-shape, falling from about 10 percent in countries at $200 in per capita GNP, to about 5 percent among countries at $3,000, and then rising to about 8 percent among countries at $20,000. The U-shape is even more pronounced if we exclude the data for primary education where hardly any spending is used to provide student financial aid. The relation between the share claimed by student aid and the per capita GNP is relatively loose, however, with a R2 statistic of only 0.06.9 It means that country wealth have little influence on the pattern of allocation. With regard to spending on textbooks and pedagogical materials, we focus on the pattern at the primary and secondary levels where these inputs are a key determinant of student achievement. The figure (panel B) shows that as countries grow richer this category claims an 9 The R2 statistic for separate regressions on secondary and higher education is, respectively, 0.23 and 0.38. -8- achievement. The figure (panel B) shows that as countries grow richer this category claims an increasing share of public spending on primary and secondary education, rising from less than 2 percent among the poorest countries, to more 8 percent among the richest countries.10 The relationship between spending share and per capita GNP is moderately loose, the R2 statistic being 0.27 for the whole sample. It suggests that countries have some leeway in setting the level of spending on pedagogical materials, although poverty does constrain its range. Figure 2: Relation between shares of public spending on student aid and pedagogical materials, and the per capita GNP, circa 1993 (A) Stiit aid (B) Pe&Vgoical nmtaials 0 10 520 - 0 0 0 '- 5 -o 0 110 0 0~~~~~~~~~~~~~00 0 0 0 ~~~~~~~0 400 10 - ~ ~ 0 oO0 0 0 .. 0 0 ~~~~~~ ~~ ~~~0 0 6 0 0 0 00 010 0 6' 0 0 0 20 0 0 o s o 0o0 o o 0 ° ° O oo 0 0 00 0000 0 0 0 0 0 0 C 0 0 0 0 200 400 800 1500 3000 10000 20000 200 400 800 1500 3000 10000 20000 Per cspita GNP (log scale) Per capita GNP (log scale) Per pupil public spending on education. We consider first the relation between country wealth and spending per pupil (expressed as a percentage of the per capita GNP), and then turn to examine its relation to two underlying components of unit spending-the pupil- teacher ratio and teacher salaries." Figure 3 pertains to spending per student at the primary and tertiary levels; the pattern for secondary education, excluded to save space, is intermediate between that for the other two levels. In primary education, average spending per pupil among countries with '° Because rich countries' total spending on pruiaiy and secondary education as a share of GNP also tends to be bigger, the gap in real spending on pedagogical mateials between rich and poor countries is in fact wider than suggested by the difference in the shares of speding. The two indicators refer to averages for the sector as a whole. Thus, a low level of unit spending may reflect low aggregate spending or large shares of enrollment in unsubsidized private education. For the pupil-teacher ratio, both the numerator and denominator refer to totals in public and private schools. These definitions are appropriate mainly for comparisons of aggregate patterns across sub-sectors and countries. -9- incomes between $200 and $3,000 is more or less flat at around 0.12 to 0.14 times the per capita GNP. It then rises as incomes increase beyond $3,000, reaching about 0.2 times the per capita GNP among the OECD countries. Figure 3: Relation between per-pupil public spending on education and per capita GNP, circa 1993 (A) Prinary e3din (B) Higher edcation .4 14 0 0 o 12 00 .3 0 0\ 0 0~~ 1 0 ~~~0 kI 0 0 ~~~0 000.0 0~~~~~~~~~~~ .2 ~~~~~~0 0 .2- ° 0 °° o ° , S-0 0 6 0 a~0 I _ 0 °- °o o - .0 ° o °o 2 \° 0 0 0 0 0 :0~6 0 0 00 0 c 0 0 0 0 ~000000 01 200 400 800 1500 3000 1000 20000 200 400 800 1500 3000 10000 20000 Per capita GNP (log scale) Per ca GNP (log scae) In higher education, the relation between the two variables follows the shape of a rectangular hyperbola: per student spending is highest among the poorest countries, averaging more than 4 times the per capita income for those at $200 in income, and flattening out at about 0.6 times the per capita income at $10,000 and beyond. In terms of the tightness of the relation between public spending per student and per capita GNP, the regression R2 statistic ranges from 0.14 for primary education, to 0.38 for higher education, both values being smaller when the data are restricted to countries below $3,000 in per capita GNP. Thus, while there is a predictable pattern of per pupil spending and country wealth, the relation is loose, implying that countries do make quite different choices in education finance. Per pupil spending can be decomposed into two underlying components: the pupil- teacher ratio and average teacher salaries. At the primary level where teachers typically account for the bulk of all spending, per pupil spending is approximately equal to average teacher salaries divided by the pupil-teacher ratio. Figure 4 shows how the two variables relate to the per capita GNP. The decline in pupil-teacher ratio as incomes rise is familiar from causal observation of schooling conditions in rich and poor countries, but the decline in teacher salaries is perhaps less familiar. The pattern stems directly from the fact that as countries develop, educated labor becomes - 10- more plentiful, and the earnings of teachers therefore generally rise less rapidly than the per capita GNP. The behavior of the two variables taken together is consistent with the basic economic logic that when an input is costly, it tends to be used more sparingly than when it is less costly. Figure 4: Relation between the pupil-teacher ratio and teacher salaries at the primary level and per capita GNP, circa 1993 (A) Pupil-t-cle ratio (B) Tehad salaries 0 15 o oO 0 60 ~~00 6D~~~ 0 0 0. 0 0 0~~~~~~~~~~~~~~1 0~~~~~~~~~~~~~~~~~~~~~ 0 10~~~~~~~~~~~~~~~~~~~~ ~~40 0 ,~. 0 0 40 400 0 0 0 a ~~0 000 0 0 0 0~~~~~~~~~~~~~ 0 0 CO, 00 0 0 0 0 0 0 00 0 00 ~ ~ ~ ~ 5 0 0 OD 00 0~~~~~~~~~~~~~~00 GNP over the entire income range in our data ismoderatelytight,asindicatedbf0 0 00 0 0 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 200 400 80 1'5003000D 10000 20000 200 400 80 1500 3000 10000 20000 Per capita (GNP (log scale) Per capita GNP (log scale) For both the pupil-teacher ratio and teacher salaries, the relation to the per capita GNP over the entire income range in our data is moderately tight, as indicated by R.2 values of 0.45 and 0.31 respectively; among lower income countries, however, the corresponding values are smaller, at 0.28 and 0.23. The results suggest that countries at the same level of economic development make appreciably different choices in how they organize teaching and learning. Given the scope for choice, the issue clearly is to discover options that produce the biggest impact on schooling outcomes. At the secondary and tertiary levels, data for a sufficiently large number of countries exist only for the pupil-teacher ratio. The variable declines with per capita GNP in secondary education, from an estimated average of 24 pupils per teacher for countries at $200 in per capita income, to about 12-13 for countries at $10,000 and above. In higher education, however, the pattern is flat, at about 16-17 students per teacher across the entire income range represented in our data. Moreover, the R2 statistic is modest, particularly among countries below $3,000 in per capita GNP, at 0.13 and 0.01, respectively, at the two levels of education. As 2.3 Education outcomes in quantity and quality We focus here on school life expectancy and the intake rate in primary education as measures of quantity (i.e. coverage), and scores on international mathematics and science tests as a measure of quality (i.e. student learning).12 Coverage of the education system. The first indicator of coverage, the school life expectancy, is defined as the number of years of schooling the average child can expect to attain as he or she grows up, given the education system's current structure of enrollments. The indicator therefore succinctly captures the system's overall coverage. The other indicator, the intake rate to grade 1, is defined as the percentage of each age cohort that enter school. Figure 5 shows the expected positive relation between the two indicators and the per capita GNP. Children in the poorest countries expect, on average, to be in school less than half as long as their counterparts in the richest countries; and the entry rate to first grade ranges from an estimated 75 percent in the former countries, to nearly universal entry in countries at or above $4,000 in per capita GNP. There is nonetheless substantial disparity across countries, especially in the intake rate. Among the poorer countries in the sample, the variable ranges from a low of less than 30 percent in Mali and Niger, for example, to nearly universal intake in Laos, and Kenya. The wide disparity is reflected in the moderately low value of the R2 statistic for the regression between the intake rate and the per capita GNP-0.30 for the whole sample, and 0.24 for the restricted sample. 12 Data are available on other indicators of coverage, but we exclude them here to save space. Gross enrollment ratios are a common measure of coverage. The data indicate that they relate positively to the per capita GNP at all three levels of education, with a reasonably close fit, as indicated by a R2 statistic of 0.31 in primary education, 0.65 in secondary education, and 0.55 in higher education. Another variable of coverage is the transition rate between cycles of education; the available data relate to the transition between the primary and secondary cycles. The data again show that it relates positively to the per capita GNP: among the poorest countries, slightly more than half of primary school leavers enter secondary school, compared with nearly universal entry for countries with per capita GNP at or above $10,000. The relation is also moderately tight, with an R2 statistic of 0.46 for the whole sample; the corresponding value for the restricted sample is much more modest at 0. 15. - 12 - Figure 5: Relation between school life expectancy and grade 1 intake rate and the per capita GNP, circa 1993 (A) Overall coverage (B) Entry rate to grade I 16 ~ o 0 >° ~~ 0 00 ~~~5 0 o%o~~~ 0 0 10o000OD8&gC04 6 - o 0 o 00 60 0 0 C 0. o~~~~~~~~~~0 ~v0 0 0~~~~~~~ 50 4- 0 o o 20 000 0 0- 0 2 8 20 VI 4 0 0 0~~~~~~~~~~~~~~~III 200 400 800 1500 3000 100 20000 200 400 800 1500 3000 10000 20000 Per capita GNP (log scale) Per capita G;NP Student learning. The cognitive skills that school leavers bring to the labor force is the other important dimension of educational outcome. 13 For our purpose we amalgamate test scores from several rounds of international mathematics and science tests administered to 9-14 year old children in the 1990s.14 The resulting indicator of achievement is calibrated to range between 0 and 100 percent, and is interpreted here as a summary measure of the achievement of children around the end of primary education and the beginning of lower secondary education. Figure 6 shows the relation between the indicator and per capita GNP in 1993. Predictably, test scores are generally higher among children in richer than poorer countries. The relation is relatively weak, however, with a regression R2 statistic of only 0.17, implying substantial variation among countries. It therefore appears that differences in student learning is not just a function of country wealth, but is probably also the result of differences in policy choices that ultimately affect the effectiveness of education processes within the classroom. 3 For examples of studies on the relation between worers' cognitive sklfls and work productivity, see Murnane, Richard, J. John B. Willet, andFrank Levy (1994). 'Ihe growng impn tance of cognitive skills in wage determination," Harvard Graduate School of Education (rnimeo); and Bishop, John (1991) "Achievement, test scores, and relative wages," in Marvin H. Kosters (ed.) Workers and Their Wages, Washington D.C.: The AEI Press. 4 The raw data pertain to some 53 countres, and come from the various international science and mathematics tests (e.g. the Third International Mathematics and Science Study) and conveniently reported in Jong-Wha Lee and Robert J. Banro, 1997 "School quality in a cross-section of countries" National Bureau of Economic Research, Working paper no. 6198, Cambridge, Mass. Not all countries patcipated in all the tests We used regression estimates relating one test score to another to fill the gaps, and to produce a single test score incoporating all the available information. The procedure produced data on 44 observations. - 13 - Figure 6: Relation between international achievement score and per capita GNP, circa early 1990s 60 0 0 0 0 0 0 O ~ ~ ~~~~ %o 0 50 ~~~~~0 0 0 0O t500 0 0~~~~ 0 ~~~~~~~~0 00 240 0 0 0 200' 400 S0O 1500 3000 10000 20000 Pe oapita GNP 2.4 Efficiency of the education system By efficiency we refer to two aspects of the education system's operation: student flow patterns, and output per unit of public spending on education. Efficiency of student flow. The available data pertain to the first 8 grades of schooling. In primary education, high dropout rates imply inefficient systems because children who quit before completing the cycle, or who do so with less than four years of schooling, are unlikely to become permanently literate and numerate. Grade repetition also leads to wastage because repeaters use at least twice as much resources as other pupils to attain the same amount of education. To measure the combined influence of these problems we construct an index of student flow efficiency which is benchmarked against a system in which no child repeats or drops out.15 An index of 100 means that all pupils complete primary school and that no one repeats; a smaller value, say 60, means that with a given amount of resources the system produces only 60 percent as many primary school completers as a system with no dropouts and no repeaters. The relation between the index and the per capita GNP appears in figure 7. 5 For details on the constuction of the summary index, see Mingat, Alain and Jee-Peng Tan (1998) "Analyzing problems in the structure of sudent flow," mimeo, Hmnm Development Department, The World Bank. -14- Figure 7: The relation between efficiency of student flow in primary education and the per capita GNP, circa 1993 0 100 0 0 0 ° ° 0 , $1,500 $3,000 $800 $1,500 e $20 $400 00/ 20% 40% 609/6 X/% 100% Percentage allocation of the increase in resources Note. top bar in ea block refas to 1993; bottom bor, to 1975. Given that the enrollment rate has an upper bound of 100 percent, it is not surprising that as countries grow rich and coverage expands, they allocate a declining share of the incremental resources for the sector to expand coverage. Yet when we plot the share allocated to expand coverage against the enrollment rate of the poorer country in each pair, the shift in emphasis away from the expansion of coverage occurs well before universal coverage is achieved, as figure 14 shows (for completeness, the figure shows on the right axis, the -36- complementary percentage used to reduce the pupil-teacher ratio). In 1993, for example, at an initial enrollment rate of 60 percent in primary education, only about 40 percent of any extra resources that materialize in the process of economic development are used to expand coverage. What is especially interesting is that the shift away from coverage occurs at much lower levels of initial coverage in the later year. At an enrollment rate of 50 percent, for example, about 75 percent of the extra resources is used to expand coverage in 1975, compared with only 50 percent in 1993. Figure 14: Relative emphasis on expansion of coverage and reduction of pupil-teacher ratio at various levels of initial enrollment ratios, 1975 and 1993 , 100- 0 u °~ < 80 - 20 **.1975 C) 60- -40 1993 40- __80 60 20 100 "I0 20 40 60 80 100 Enrollment rate in primary education (%) 3.4 Summary of structural differences in education between rich and poor countries In the foregoing discussion we have decomposed the differences in education between rich and poor countries, using an accounting identity involving terms pertaining to the availability of resources for education on one hand, and the deployment of those resources on the other. When pairs of countries at contiguous income levels are compared, the richer country in each pair invariably enjoys a substantial resource advantage; a country at $3,000 in per capita GNP, for example, has more than 4 times the resources per school-aged child of a country at $200. The advantage arises from three sources: bigger budget allocations, lighter demographic burdens; and smaller teacher salaries relative to the per capita GNP. The resource advantage of richer countries allows them to expand coverage and lower teacher-pupil ratios. -37 - Looking more closely at primary education, we found that at all stages of economic growth, differences in budget allocations generally make only a modest contribution to richer countries' resource advantage, while the decline in teacher pay relative to the per capita GNP makes the biggest contribution, accounting for between 56 and 68 percent of the advantage. The contribution of the easing of the demographic burden shows dramatic changes over time, ranging between 17 and 32 percent in 1993 across the income spectrum represented in our exercise (from $200 to $10,000), and between 1.7 and 62 percent in 1975. The much wider range in 1975 is consistent with the fact that the transition to lower fertility rates occurred at higher income levels in 1975 than in 1993. With their resource advantage, richer countries have the option of emphasizing an expansion of coverage or a reduction in the pupil-teacher ratio. They have emphasized the former only at the earliest stages of economic growth, when incomes rise from $200 to $400 in 1993 and from $200 to about $1,500 in 1975. At all subsequent levels of income growth, the emphasis shifted to lowering the pupil-teacher ratio. Noteworthy is that the shift occurred in contexts where coverage in primary education is still not yet universal. 4. What are the policy issues? Two kinds of policy issues arise from the foregoing analysis. The first relates to the long-term shift in favor of lowering pupil-teacher ratios that we have documented: to what extent does it represent an efficient path for the development of primary education in developing countries, particularly in the poorest countries where a significant share of children still have no access to primary education? The second issue relates to the fact that the patterns refer to the average for countries at each stage of economic development: how much scope is there for deviation from the average pattern, and what difference in educational outcomes does it make? 4.1 Expansion of coverage versus lowering pupil-teacher ratios In all countries improving access and student learning are key objectives of educational policies. The accelerated decline in pupil-teacher ratios over the past two decades suggests that policy makers have succumbed to the hope that lowering this indicator would translate into gains in student learning. Yet there is persistent and widespread evidence that simply lowering pupil-teacher ratios does not generally produce the desired gains in learning.32 While recent evidence from a randomized experiment in the United States do suggest a 32 See, for example, Hanushek, E. A.(1995) "Interpreting recent research on schooling in developing countries." World Bank Research Observer 10(2): 227-46. -38 - connection between smaller classes and achievement33, the finding has limited relevance in developing countries for several reasons. Most importantly, the experiment involves reducing class size from a much smaller initial level than is typical in the average low-income country. Gains in student learning also depends on changes in teaching and learning practices-from passive learning by students listening to a teacher talking in front of the class, to hands-on learning-that may be difficult to implement in systems without adequate support and resources for teacher training and classroom experimentation. Evidence from the cross-country data used in the present study support the claim that lowering the pupil-teacher ratios has few payoffs (if any) in student learning in the typical context of developing countries. In table 9 we regressed achievement scores on international mathematics and science tests among 9-14 year-old children against selected regressors.34 The results strongly suggest that variation in pupil-teacher ratios in the sample range has no impact on student achievement; in all three regression models, the coefficient on the indicator is statistically not different from zero. Instead, the educational attainment of adults in the country and the adult literacy rate show strong and positive impact. The estimates indicate that a one- year rise in adults' average school attainment raises a country's performance on international mathematics and science tests by 0.84 points (or 0.17 of a standard deviation from the sample mean), and that a one percentage point advantage in the adult literacy rate raises test scores by 0.30 points (or 0.06 of a standard deviation from the sample mean). Table 9: Regression estimates of the correlates of scores on international mathematics and science tests, 1990s . ~~1.11- - Ln (per capita GNP, 1993) (1.5) Adults' average years of school attainment a/ 0.84** ______________(2.2) Adult literacy rate (%) 0_(3.3)0*** Pupil-teacher ratio in primary education -0.11 (011 (0.7) Constant 40.6 43.7 17.7 Number of countries 35 33 34 R 2 0.19 0.30 0.38 Source: the estimates are based on country data on achievement from Lee and Barro 1997 (op. cit.); and on the other variables from the IREDU database. 33 See Krueger, Alan B. (1997) "Experimental estimates of education production functions," National Bureau of Economic Research Working Paper # 6051, Cambridge, Mass. 34 See section 2.3 above for a description of the data used in this analysis. -39- That adults' educational capital affect children's learning achievement should hardly be surprising. The link arises not just because educated parents provide more effective support for a child's education, but also because in more literate societies, children are exposed to many learning opportunities in daily living-through newspapers and other printed materials-that reinforce what is taught in the classroom. The finding argues not so much for massive investment in adult literacy programs, as for a dedicated and sustained effort to expand coverage so that each child has the opportunity to enter school and complete at least primary schooling. Especially in countries where substantial numbers of children still do not enroll and where too many of those who enroll drop out before finishing the cycle, a re-orientation toward expanding coverage and away from reducing the pupil-teacher ratio, requires serious consideration by policy makers. Where budgets for education are limited, these choices present a stark tradeoff that is probably more efficient to resolve in favor of expanding coverage. This does not mean there is no place for policies to rationalize the distribution of teachers within the system; such policies are clearly relevant where wide disparities in pupil-teacher ratios exist (e.g. across urban and rural areas and across rich and poor neighborhoods). However, the appropriate way to address them is to redistribute resources across schools, rather than to reduce the overall pupil- teacher ratio at the expense of expanding coverage in the system as a whole. As a final observation it is of interest to take note of the choices that Korea (whose education system is recognized as arnong the best in the world today) made with regard to the pupil-teacher ratio as the country developed economically.35 Between 1950 and 1970, as the per capita GNP rose from $500 to $1,000, this indicator was maintained at the astonishingly high level of nearly 60 pupils per teacher; at the same time, primary education expanded to reach universal coverage in the 1 960s. It was only at subsequent stages of income growth that the ratio was allowed to decline, reaching 30 only after the per capita GNP had risen to $8,000 (in the early 1990s). Although we have no information on student learning in Korea in the 1950s and 1960s we do know that in the 1980s and 1990s, the average Korean pupil achieved among the highest scores on international science and mathematics tests, despite the country's generally higher pupil-teacher ratio relative to other countries. Korea's experience suggests that a strategy of emphasizing expansion of coverage over lowering of the pupil-teacher does not necessarily harm student learning in the long run; on the contrary, by broadening the base of human capital development, it may have contributed to the outstanding achievement of its young people today. 3 See Alain Mingat (1988), "The strategy used by high-performing Asian economies in education: some lessons for developing countries," World Development 26(4): 695-715. -40 - 4.2 The scope for and consequences of policy choice within countries We turn now to examine deviations from the average patterns in the structure of progress in education, focussing on the diversity across countries in terms of budget allocations for education, demographic pressures and labor market conditions for teachers, as well as education outcomes as reflected in coverage and pupil-teacher ratios. We use examples from selected low-income countries where the scope for policy intervention appears to be particularly clear. Consider Laos and Niger which had a similar per capita GNP in 1993-$280 and $270, respectively. The resources available per school-aged child for primary education in Laos were 4.3 times as high as that in Niger, however. What is the source of this enormous gap? The latter country admittedly faced a heavier demographic burden, with the school-age population representing 24 percent of the total population rather than only 21 percent. But this factor by itself would have implied an advantage for Laos of only 1.17 times. Indeed the advantage should have been completely reversed by the fact that in Laos public spending on primary education was only 0.64 times as high as in Niger. Instead the advantage enjoyed by Laos widens to a startling 4.3 times. The reason is that the cost of teachers is much smaller in Laos, with salaries averaging 1.7 times the per capita GNP, compared with 9.7 times in Niger. Because of its resource advantage, Laos achieved more than twice Niger's coverage in primary education, enrolling 53 of the primary school age population. instead of only 20 percent. In both countries, pupil-teacher ratios were comparable at about 31-34 pupils per teacher. Niger allocated more public spending for primary education than Laos, but achieved less in coverage, because teacher salaries were so high. High teacher salaries are indeed a common feature of the education sector in Francophone African countries. In countries where budget allocations for primary education are not as favorable as in Niger, the result has been lower coverage as well as severe pressures on classroom conditions in the form of very high pupil-teacher ratios. Compare, for example, Chad and Nepal-countries with per capita GNP of around $200 in 1993. Public spending on primary education in Chad was only 80 percent as high as than in Nepal in 1993, and teacher salaries were 1.72 times as high. Both factors put Chad at a resource disadvantage, which translated into smaller coverage and higher pupil-teacher ratios: Chad enrolled only 25 percent of its primary school-age children, compared with 54 percent in Nepal; and its education system had an average of 61 pupils per teacher, compared with 39 in Nepal. -41- The contrast between Chad and Nepal, and between Laos and Niger highlights the fact that poverty is not the only factor that hampers educational development. Demographic pressures tend to be more burdensome in poor countries, increasing the difficulty of expanding coverage and improving the quality of the learning environment. Yet among poor countries, policies that affect budget allocations to education and especially policies that influence teacher salaries do vary and account for widely different paths of educational development. Cross country comparisons as well as within country evidence point to possible directions for policy development. In Niger, the scope for further increases in budget allocations for primary education is probably more limited than in Chad, given that current spending levels are already relatively high. In both countries, as in the majority of Francophone African countries, substantial headway depends on the future evolution of teacher salaries. In many of these countries, teacher salaries have declined in real terms (i.e. relative to the per capita GNP) between 1980 and 1993, reflecting the impact of general price inflation and structural adjustment policies36. But several observations among countries in the region indicate that much deeper declines are probably feasible: (a) Salaries in private schools are much lower than in public schools, by as much as 40 percent or more; (b) Excess demand for public sector teaching jobs, even when the jobs are offered at much lower salaries than those received by incumbents. In Burkina Faso, the government created a new category of lesser-qualified teachers-"instituteur adjoints" which require only a lower secondary school certificate. It received 18,000 applications for the 800 positions, many from people with much more education than was required. Similarly in Senegal, the government hired "volontaires" at one third the salary of incumbent teachers, and even so received 28 applications for each available job in this category. (c) Expansion of community schools where teachers are hired directly by the local community. Teachers typically receive a fraction of the pay of public school teachers and none of the non-pecuniary benefits of public sector employment. Examples include the "ecoles de base" in Mali and the "ecoles spontan6es" in Chad. 36 See Mingat, Alain (forthcoming 1999) "Assessing priorities for education policy in the Sahel from a comparative perspective," Comparative Education. -42 - Hiring lesser qualified teachers and encouraging cheaper types of schools are obviously feasible in many of the countries where public sector teacher pay is relatively high. A direct consequence would be to allow coverage to expand; at the same time the policy does not necessarily compromise student learning. A recent evaluation suggests that students in Mali's "ecoles de base" in fact outperform their counterparts in regular public schools; and that students taught by the "volontaires" in Senegal outperform pupils taught by the more qualified teachers.37 Thus, although gains in coverage and improvement in classroom conditions generally materialize as part of the process of economic development, countries achieve better or worse results than other countries at comparable levels of economic development according to the policies they adopt, particularly with respect to budget priorities, teacher salaries, and the relative emphasis on expansion of coverage and decreases in pupil-teacher ratios. 5. Conclusion In this study we have used aggregate data to explore the relation between country wealth and various aspects of educational development; examine the underlying sources of differences in education among rich and poor countries; and raise questions about promising directions for policy development in the sector, particularly in the context of lower income countries. As countries become richer, they generally spend more per pupil, expand coverage of the school-aged population, lower the pupil teacher ratio, achieve improvements in student learning, enhance the efficiency of student flow patterns and sector operations, and improve equity in access to schooling and the distribution of public spending on education. On some dimensions, such as coverage, access and per pupil spending, the improvement follows a relatively steep rise with per capita GNP, whereas on other indicators, such as student learning, the rise is less steep, and on yet other indicators, such as the share of enrollments in private schools, there is hardly any relationship to the per capita GNP. An appealing explanation for the generally superior outcomes in richer countries is that they have more resources to support well-functioning and effective systems of education. Indeed richer countries have more resources for education per school-aged child. But bigger budget allocations account for a modest share of the advantage, since public spending on education relative to the GNP rises only modestly as incomes grow. Focussing on primary 37 Fomba, M. (1996) "L'enseignment primaire au Mali: mode de financement et acquisition des eleves," PhD dissertation, Uinversite de Bourgogne, Dijon, France. -43 - education for which more detailed analysis was feasible, we found that this factor contributed no more than 20 percent of a richer countries' resource advantage as incomes rise from $200 to $400 (in 1993 prices), and no more than 14 percent of the advantage at all subsequent stages of income growth up to $10,000. Easing of the demographic burden makes a bigger difference, its contribution ranging between 17 and 32 percent of richer countries' advantage as incomes rise from $200 to $10,000. At all stages of economic development, by far the most important source of increased public spending available per school-aged child in the richer countries is the dramatic decline in teacher salaries relative to the per capita GNP as countries grow rich, accounting for as much as two-thirds of the resource advantage in the richer country at some stages of income growth. As the resources available per school-aged child materializes with income growth, countries are able to expand coverage and improve schooling conditions via a reduction in the pupil-teacher ratio. They initially allocate more of the resources to expand coverage, but the emphasis quickly shifts in favor of lowering the pupil-teacher ratio even though a substantial share of the primary school-aged population remains out of school. This trend toward lower pupil-teacher ratios was stronger in the 1990s than in the 1970s. Is the shift toward smaller pupil-teacher ratios an efficient way to use the extra resources that become available for education in the course of economic growth? According to the cross-country evidence in the present study as well as findings reported in the broader literature the answer is probably in the negative, particularly in the context of the poorer countries. One reason is that over the typical range of the pupil-teacher ratio in developing country contexts, a reduction in the variable has little impact on student learning. Another is that reducing the pupil-teacher ratio implies slower progress in expanding educational coverage. The long run result is lower levels of educational attainment among tomorrow's adults, which in turn is likely to diminish learning achievement among tomorrow's children. Besides the systematic differences in education associated with country wealth, substantial diversity is also evident among countries at comparable levels of economic development. In all low income countries, for example, high teacher salaries relative to the per capita GNP reduce the scope for expanding educational coverage, but the salaries are much higher in African countries than in comparable countries in other regions, leading to wide gaps in coverage and well as in pupil-teacher ratios. Where a country deviates substantially from the income-related structural patterns, a likely reason is that it has made different policy choices than other countries. The scope for choice is indeed wide, ranging from the overall pattern of resource allocation in education, to how education services are organized and delivered. Setting aside the impact of country wealth, policy choices, particularly with regard to teacher salaries -44 - and the tradeoff between coverage and reductions in pupil-teacher ratios, can and do have significant influence on the path of educational development in developing countries. -45 - Appendix A Comparing Education in Rich and Poor Countries in 1975 and 1993 In section 3.3 in the text we compared richer countries' advantage in education in 1975 and 1993 by looking at pairs of countries at contiguous income levels in each year. Another way to examine the data is to compare pairs of countries--one from each year--at the same level of per capita GNP (in constant dollar terms). The patterns that emerge (summarized in the three graphs below) are consistent with those discussed in the text. At each stage of economic development, more public resources are available for primary education in the later than earlier year. Larger budget allocations account for little of the advantage in the later year, but differences in demographic burdens and teacher pay relative to the per capita GNP make significant contributions. A country at $800 in per capita GNP in 1993, for example, derives nearly 40 percent of its resource advantage relative to its 1975 counterpart from the fact that its school aged population was only 20.2 percent of the total population compared with 24.1 percent in the earlier year. Declines in relative teacher pay make the largest contribution, however, accounting for between 80 and 60 percent of the resource advantage in the latter year. With regard to how the resource advantage is used, the pattern confirms what we have already seen above: in the poorest countries, the overriding emphasis has been on expansion of coverage, but this focus quickly switches in favor of reducing the pupil-teacher ratios once the per capita incomes exceeds $800. Figure A.1: Resource availability for primary education in 1993 and 1975 10000 9 3000 6@ : 1500 2o 800 ; - 400 200 0.0 0.5 1.0 1.5 2.0 Resource availability in 1993 as multiple of that in 1975 -46 - Figure A.2: Sources of the resource advantage for primary education in 1993 relative to that in 1975 LBudget f Demographic burden * Teacher pay 10,000 3,000 E - 1,500 . . - - _11 u:> o 800 400 1 200 -I I- III -20 0 20 40 60 80 100 120 Percentage contribution Figure A.3: Deployment of the resource advantage for primary education in 1993 relative to that in 1975 U Coverage Pupil-teacher ratio 10000 I 11 /001 Z P 0O 3000 1500 _uwazofls i > ~800 3 7571137171300#§, 7 3711117FIRg^g 000000 4) ~ 40 200 __ 0 20 40 60 80 100 Percentage allocation Policy Research Working Paper Series Contact Title Author Date for paper WPS1993 Financial Safety Nets and Incentive Philip L. Brock October 1998 K. Labrie Structures in Latin America 38256 WPS1994 Estimating Wealth Effects without Deon Filmer October 1998 S. Fallon Expenditure Data - or Tears: Lant Pritcheft 38009 with an Application to Educational Enrollments in States of India WPS1995 What Macroeconomic Policies Mansoor Dailami October 1998 8. Nedrow Are'Sound?" Nadeem ul Haque 31585 WPS1 996 Namibia's Social Safety Net: Kalinidhi Subbarao October 1998 P. Lizarondo Issues and Options for Reform 87199 WPS1997 On Measuring Literacy Kaushik Basu October 1998 M. Mason James E. Foster 30809 WPS1998 The Structure and Determinants of Sudharshan Canagarajah October 1998 A. Garscadden Inequality and Poverty Reduction Dipak Mazumdar 38400 in Ghana, 1988-92 Xiao Ye WPS1 999 Heterogeneity among Mexico's Wendy V. Cunningham October 1998 T. Gomez Micro-Enterprises: An Application William F. Maloney 32127 of Factor and Cluster Analysis WPS2000 GATT Experience with Safeguards: J. Michael Finger October 1998 L. Tabada Making Economic and Political 36896 Sense of the Possibilities that the GATT Allows to Restrict Imports WPS2001 Measuring the Dynamic Gains from Romain Wacziarg November 1998 S. Crow Trade 30763 WPS2002 Accounting for Toxicity Risks in Susmita Dasgupta November 1998 Y. D'Souza Pollution Control: Does It Matter? Benott Laplante 31449 Craig Meisner WPS2003 Thailand's Corporate Financing Pedro Alba November 1998 R. Vo and Governance Structures Stijn Claessens 33722 Simeon Djankov WPS2004 What Can Be Expected from African Alexander Yeats November 1998 L. Tabada Regional Trade Arrangements? 36896 Some Empirical Evidence WPS2005 Fiscal Federalism and AnwarShah November 1998 S. Valle Macroeconomic Governance: 84493 For Better or For Worse? WPS2006 Household Welfare Measurement Jesko Hentschel November 1998 PREM and the Pricing of Basic Services Peter Lanjouw 87736 Policy Research Working Paper Series Contact Title Author Date for paper WPS2007 Regional Integration Arrangements: Dean A. DeRosa November 1998 L. Tabada Static Economic theory, Quantitative 36896 Findings, and Policy Guidelines WPS2008 Volatility and Contagion in a Pedro Alba November 1998 D. Fischer Financially Integrated World: Amar Bhattacharya 38656 Lessons from East Asia's Recent Stijn Claessens Experience Swati Ghosh Leonardo Hernandez WPS2009 Poverty and the Economic Transition: Peter Lanjouw November 1998 P. Sader How Do Changes in Economies Branko Milanovic 33902 of Scale Affect Poverty Rates for Stefano Paternostro Different Households? WPS2010 The Real Impact of Financial Shocks: liker Domac November 1998 M. Greaves Evidence from the Republic of Korea Giovanni Ferri 81876 WPS2011 Measuring Poverty Using Qualitative Menno Pradhan November 1998 P. Sader Perceptions of Welfare Martin Ravallion 33902 WPS2012 Export Quotas and Policy Constraints Sanjay Kathuria November 1998 L. Tabada in the Indian Textile and Garment Anjali Bhardwaj 36896 Industries WPS2013 A New Database on Investment and Al Crego November 1998 P. Kokila Capital for Agriculture and Donald Larson 33716 Manufacturing Rita Butzer Yair Mundlak WPS2014 Land Institutions and Land Markets Klaus Deininger November 1998 M. Fernandez Gershon Feder 33766