Policy Research Working Paper 8773 Mobilizing Resources for Education and Improving Spending Effectiveness Establishing Realistic Benchmarks Based on Past Trends Samer Al-Samarrai Pedro Cerdan-Infantes Jonathan Lehe Education Global Practice March 2019 Policy Research Working Paper 8773 Abstract This paper looks at how countries have mobilized addi- as healthy economic growth. Increases in public education tional resources for education and assesses their impact on spending did not generally result in major improvements in access and learning outcomes, using the World Bank’s new average education outcomes. Using the available data, the Learning-Adjusted Years of Schooling measure. The paper paper shows that a doubling of government spending per shows that global spending on education has risen signifi- child led to an increase in learning-adjusted years of school- cantly over the past two decades, although spending as a ing of only half a year. Preliminary findings also show that share of gross domestic product has remained relatively countries with lower efficiency and spending are expected unchanged, at about 4.5 percent. However, global trends to get the most from increases in spending in improved mask large differences across regions and country income education outcomes. The paper concludes by outlining an groups. For example, low-income countries recorded the approach that allows countries to assess their potential for largest increases in terms of the share of GDP spent on increasing education funding and the expected effects on education, but the absolute amount they devoted to educa- their education outcomes, based on benchmarks drawing tion remained low compared to other countries. Economic from the data of comparable countries. It also underscores growth has been the main driver of increases in public edu- the urgent need to improve data on public education spend- cation spending. Yet, countries that achieved the largest and ing and education outcomes, to extend this analysis to cover most rapid spending increases did this through a combina- a wider set of countries and increase the robustness of coun- tion of increases in overall government revenues, a greater try-level benchmarks. prioritization of education in the government budget as well This paper is a product of the Education Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at salsamarrai@worldbank.org. 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 views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Mobilizing Resources for Education and Improving Spending Effectiveness: Establishing Realistic Benchmarks Based on Past Trends* Keywords: Education Finance, Education Outcomes, Fiscal Space, Public Education Spending JEL codes: E62, H20, H52, I21, I22, O15 * We are grateful for comments and advice from Omar Arias, Mohammed Audah, Deon Filmer, Mercedes Garcia- Escribano, Baoping Shan, Igor Kheyfets. Jaime Saavedra and Lars Sondergaard. 1. Introduction and Main Findings When countries around the world pledged to achieve the Sustainable Development Goals (SDGs) in 2015, they also agreed to meet an ambitious set of targets for their education systems. Estimates suggest that an additional $191 billion to $311 billion will be needed each year to expand access and improve the quality of basic education for all children and to attain the other education targets by 2030 (Education Commission 2016; UNESCO 2015).1 Yet mobilizing additional resources alone does not guarantee that education outcomes will improve. It will also be necessary to strengthen the effectiveness and efficiency of education spending if countries are to meet these ambitious goals. This paper seeks to contribute to the global push to improve education systems by analyzing education spending and outcomes across countries and by providing governments with an approach to benchmarking performance that can guide their own policy making. Specifically, the paper does the following three things: First, it documents global-, regional-, and country-level trends in public education spending, identifying the main sources of spending increases around the world and exploring differences in spending trends across income levels and regions. Second, it looks at how countries have mobilized additional resources for education and assesses their impact on access and learning outcomes using the World Bank’s learning- adjusted years of schooling measure. Third, the paper outlines an approach that allows countries to assess their own potential for increasing education funding and the expected effects on their education outcomes, based on benchmarks drawing from data of comparable countries. The proposed approach to benchmarking spending decisions is a key contribution of the paper. The exercise does not seek to provide a definitive answer to whether countries should increase their spending or by how much. Instead, the benchmarking exercise can serve as a starting point for budget discussions. Specifically, the benchmarks seek to influence decision making in several, crucial ways. By comparing both levels of spending and changes in spending across countries, they can provide evidence on whether and by how much countries could increase their funding for education. By predicting how spending increases might impact outcomes, this approach can also identify which countries could be expected to see the largest returns from additional education funding. And by linking spending and expected outcomes, the benchmarks can increase accountability in the education sector. While data limitations currently do not allow us to perform this benchmarking for all countries, the expectation is that this approach can help guide countries’ spending decisions. By identifying countries that have the capacity to mobilize more resources and achieve high expected returns from additional spending, the proposed approach may persuade such countries to increase their funding. Countries categorized as having little fiscal space or low expected returns, on the other hand, may use the opportunity to strengthen accountability and effectiveness in education spending. Overall, this benchmarking approach can help bring about a more efficient allocation of education resources across the globe. The paper’s main findings are: Global public spending on education has risen significantly over the last two decades but spending as a share of gross domestic product (GDP) remained relatively unchanged at about 4.5 percent, and overall, growth has been uneven. Low-income countries recorded the largest increases in terms of the share of GDP spent on education, but the absolute amount they devoted to education remained low compared to other countries. Countries in Africa and Latin America also saw marked growth in terms of education spending as a percentage of GDP, while the Middle East and North Africa saw a significant decline. 1 Unless noted, all $ figures refer to US$. 2 Low-income countries devote the highest share of public spending to primary education and have mostly used funding increases to expand access to education rather than spend more on each student. In contrast, middle-income countries tend to spend a greater share of the government budget on post- primary education and have increasingly used additional funding to spend more on each student enrolled. Economic growth has been the main driver of increased public education spending worldwide. However, countries that achieved the largest and most rapid increases in public education spending did this through a combination of increases in overall government revenues, a greater prioritization of education in the government budget as well as healthy economic growth. Increases in public education spending did not generally result in major improvements in average education outcomes. Using the available data, the paper shows that a doubling of government spending per child led to an increase in learning-adjusted years of schooling (LAYS) of only half a year. Even for countries that achieved the most with an increase in resources, the improvements were fairly limited. For example, a 10 percent increase in spending per child, even for these countries, was associated with a 2.0 percent increase in learning-adjusted years of schooling compared to a 0.8 percent increase for countries at the average. Our preliminary findings also show that the effectiveness of public education spending in any given country depends on its initial levels of efficiency and spending. The historical evidence suggests that countries with lower efficiency and lower spending are expected to get the most out of increases in spending in terms of improved education outcomes. For these countries, outcomes tend to be relatively low and increases in spending can have a bigger impact on enrolling more children in school and achieving basic learning outcomes. Higher spending countries and those with already high levels of outcomes relative to their spending saw “less bang for the buck,” suggesting they need significant reforms to their education and financial management systems to yield further improvements in outcomes from additional spending. The paper also underscores the urgent need to improve data on public education spending and on education outcomes to extend this analysis to cover a wider set of countries and to increase the robustness of country- level benchmarks. The paper is organized as follows. Section 2 documents how public education spending has changed globally over the last 20 years and identifies the sources of these changes. Section 3 looks at how governments have mobilized additional resources for education and how this has impacted learning outcomes. Section 4 outlines how countries can use benchmarks on public education spending changes and the effect of spending increases on education outcomes to inform their own decision making. Section 5 offers some conclusions drawn from our analysis. 2. How Has Education Spending Changed across the World over the Last 20 Years? The world has seen significant gains in education outcomes over the last two decades. Specifically, a greater proportion of children now start school at an earlier age and stay in school longer than ever before. The purpose of this section is to document the changes in public education spending that underlay these significant improvements in outcomes. 2.1 Key Variables and Data Sources It was difficult to establish patterns and trends in public education spending because of the low coverage and poor quality of the available data (Box 1). For example, only about half of the 218 economies classified 3 by the World Bank have data on the most commonly reported indicator on education: government education spending as a proportion of gross domestic product (GDP) (Figure 1). Also, of the 124 countries that reported this information in 2000, only 75 reported the same information in 2015. This limited coverage stands in stark contrast to the data available for the health sector. Since 2000, 186 countries have compiled detailed health-financing data each year. We also found discrepancies between statistics provided by UNESCO’s Institute of Statistics (UIS) and the International Monetary Fund (IMF) on the estimated value of countries’ public education spending as a share of GDP, with the IMF’s numbers averaging 0.7 percentage point higher. Similarly, we found an average 0.8 percentage point difference between UIS data for this measure and those reported in the World Bank’s public expenditure reviews (PERs), since 2013 (Figure 1).2 Box 1: Indicator Definitions and Sources a We primarily used data from UNESCO’s Institute of Statistics (UIS) to document country-level changes in education spending between 1999 and 2015 (Table 1). Where there were gaps in the UIS database, we used data from the International Monetary Fund’s (IMF) Government Finance Statistics (GFS) database, with these comprising about 12 percent of all observations. We used data from the same sources to break down spending by level of education (primary, secondary, and tertiary).b Our data on overall government spending and GDP came from the World Bank’s World Development Indicators (WDI), supplemented by data from the Organisation for Economic Co-operation and Development (OECD).c Indicator Source 1. Government expenditure on education (2015 purchasing power parity [PPP] dollars) UIS & IMF 2. Government education expenditure as a percentage of GDP UIS & IMF 3. Government education expenditure as a percentage of total government expenditure UIS & IMF 4. Education level (primary, secondary, tertiary) share of government education expenditure UIS & IMF 5. Government education expenditure per student and by education level (2015 PPP dollars) UIS & IMF 6. Government expenditure as a percentage of GDP WDI 7. GDP in 2011 constant PPP dollars WDI a. Online databases were accessed in November 2018. b. While UIS collects data on the functional classification of government education spending worldwide, we chose not to use them because establishing patterns and trends is difficult given the relatively low coverage of these indicators across countries. c. Annex Table A1 presents the averages and numbers of observations for each of these variables in each year. A full description of data sources for each variable and the approach to the statistics reported in the paper are available from the authors on request. These gaps in data coverage and quality presented challenges for our analysis. With regard to trends in public education spending across all countries and for groups of countries (for example, classified by income or region), the difference in average spending from one year to the next is a function of actual changes in spending at the country level but also of the number of countries reporting each year. For example, average public spending on education for all countries dropped from 4.7 percent of GDP in 2003 to 4.4 percent of GDP in 2004. However, a large share of this drop resulted from the exclusion of several countries with higher-than-average spending which had reported in 2003 but did not in 2004. While it is beyond the scope of this paper to address issues of data quality, we had to find ways to surmount these 2 We also found that related indicators are frequently inconsistent. For example, the raw data on public education spending as a percentage of GDP are often not the same as the figure we get if we multiplied together a country’s public education spending as a percentage of government expenditure and government expenditure as a percentage of GDP. On average, there was a discrepancy of 55 percent between these two estimates of public education spending as a share of GDP. 4 problems in our analysis. To compensate for this volatility, we used different approaches to measure spending for each year, including using the median and four-year country averages, interpolating data to fill data gaps, and excluding countries with less than 10 years of reported data. While these different approaches yielded slightly different estimates, the patterns and trends reported in the paper remained broadly the same. The rest of the paper uses four-year averages when analyzing patterns and trends in public education spending based on country groupings by income and regions.3 Figure 1: The Coverage of Education Spending Data is Limited, and of Low Quality a. Countries reporting public sector spending as a b. Alternative estimates of government education percentage of GDP spending as a percentage of GDP 200 160 91 90 120 75 80 40 health education 0 2000 2005 2010 2015 Sources: WHO Global Health Expenditure Database; World Bank public expenditure reviews (PER); UIS online database. Note: The data labels in the left hand panel shows the number of countries that reported educaiton spending as a percentage of GDP in 2000 that continued to report in subsequent years. Years in parentheses in the right hand panel’s horizontal axis refer to the year of the data. A further data limitation that we faced was a lack of information on household education spending, which meant that we had to focus on public spending rather than total (public and private) spending. The UIS has only recently begun to collect information on private education spending and has data on private household education spending as a percentage of GDP for only 34 countries per year on average between 1999 and 2015 (Box 2). Taking private spending into account can often change what conclusions can be drawn from making comparisons between countries. A recent study that used data from national education accounts to estimate total spending found that public spending in Nepal was much less than in Vietnam. However, when private spending was included, Nepal spent much more than Vietnam (UIS 2016). The exclusion of private spending may help explain why some country governments that allocate a similar share of public funding to education achieve very different levels of education outcomes. 3 These consist of averages of country-level data over five four-year periods (1998–2001, 2002–05, 2006–09, 2010– 2013, and 2014–17) that were then averaged across income or region groups. These averages included all countries with data available for at least one year in the four-year period. There was relatively little information for each country in 2017 for the final period, 2014–17. 5 Box 2: Household spending on education makes up a large proportion of total spending in low- income and lower-middle-income countries Household spending on education tends to account for a greater percentage of GDP in low-income countries than wealthier countries (see figure below). Comparable information on household education spending at the country level is relatively scarce. However, using country averages over four-year periods, we found that, on average, household spending on education in 2010-13 accounted for 2.6 percent of GDP in low-income countries but only 0.7 percent in high-income countries. It also appears that household spending makes up a larger share of total education spending in low-income countries. In 2010–13, we found that average government spending on education accounted for 3.7 percent of GDP in low-income countries and 4.8 percent in high-income countries. Combining the data on household spending with those on government spending, it becomes clear that households in low-income countries provided 41 percent of all education spending compared to only about 13 percent in high-income countries. In terms of regions, households in Africa and Latin America contribute the most to education as a share of GDP. In the four-year period from 2010 to 2013, households in Africa spent the equivalent of 2.1 percent of GDP on education. When household and government spending are combined, African countries spent the equivalent of 6.4 percent of GDP on education and Latin American countries spent 6.8 percent. Only about 40 countries have more than 10 data points on household education spending between 2000 and 2015 in the UIS database. In this limited sample of countries, average household spending appears to have fallen during the first half of the 2000s from 1.4 percent in 2000 to 0.8 percent in 2005, but then began to rise and in 2013 accounted for about 1.2 percent of GDP. Households Contribute the Largest Share of National Income to Education in Low-income Countries Household Education Spending as a Percentage of GDP, 2006–13 3.0 2006–09 2010–13 2.5 2.0 1.5 1.0 0.5 0.0 LIC LMIC UMIC ECA (24) EAP (7) HIC MNA (5) LCR (13) AFR (13) Income group Region Source: World Bank calculations using UIS and IMF online databases. Note: Income groups are defined by World Bank country income group classification in 2017. Figures in parentheses record the average number of countries in each region for which data are available. LIC = low-income country, LMIC = lower-middle-income country, UMIC = upper-middle-income country, and HIC = high-income country. AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, EAP = East Asia and the Pacific, MNA = Middle East and North Africa. There was no information for South Asia. 6 2.2 Trends in Education Spending Average government education spending around the world doubled in real terms between 1999 and 2015 but spending in terms of share of GDP increased only slightly (Figure 2). Much of the 4 percent annual growth in public education spending was fueled by economic growth rather than significant increases in the share of GDP devoted to education. The share of GDP devoted to public education spending increased from an average of 4.6 percent in 1999 to an average of 4.7 percent in 2015. Figure 2: Real government spending on education has grown significantly since 1999 Public education spending, constant 2015 PPP dollars and percentage of GDP, 1999-2015 40,000 5.0 Public education spending (% GDP) (Milllions, constant 2015 PPP $) Total public education spending 30,000 4.5 20,000 4.0 10,000 3.5 0 3.0 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Public education spending Public education spending as % of GDP Linear trend (spend) Linear trend (spend as % of GDP) Source: World Bank calculations based on UIS and IMF online databases. Note: The bars show the country average of total public spending on education for all countries with data available in each year. Low-Income Countries Have Seen Faster Growth of Public Education Spending as Share of GDP Low-income and upper-middle-income countries have registered relatively large increases in the share of GDP devoted to government education spending since 1999 (Figure 3). For example, countries that were classified as low-income in 2017 increased their public education spending from 3.2 percent of GDP in 1998–2001 to 4.0 percent in 2014–17. In contrast, government education spending as a share of GDP has remained relatively constant for upper-middle-income countries (4.6 percent) and high-income countries (4.8 percent) (Figure 3). These trends have resulted in a narrowing of the gap between country income groups in terms of share of GDP allocated to public education spending. 7 Figure 3: Government Education Spending as a Share of National Income Has Risen Fastest in Low-income and Lower-middle-income Countries a. Public education spending as percentage of b. Public education spending as percentage of GDP by income group classifications as of 2017 GDP in low-income countries by graduation status 6 6 remained LIC between 1999 and 2017 graduated between 1999 and 2017 5 HICs 5 UMICs 4 LMICs 4 LICs 3 3 1998–2001 2002–05 2006–09 2010–13 2014–17 1998–2001 2002–05 2006–09 2010–13 2014–17 Source: World Bank calculations based on UIS and IMF online databases. Note: World Bank income group classifications in 1999 are used to group countries and are as follows: LIC = low-income country, LMIC = lower-middle-income country, UMIC = upper-middle-income country, and HIC = high-income country. Overall, countries that graduated out of low-income status between 1999 and 2017 devoted a larger share of GDP to public education spending than those countries that remained in the low-income group (Figure 3). Between 1998 and 2013, countries that graduated from low-income status spent approximately 4.1 percent of GDP on education compared to 3.5 percent for countries that did not graduate.4 However, this gap has narrowed over time because of a decline in the spending share among graduating countries and an increase in the share among non-graduating countries. Between 2006–09 and 2014–17, government education spending as a share of GDP declined from an average of 4.4 to 4.1 percent among graduating countries but increased from an average of 3.7 to 4.1 percent among non-graduating countries. Government Education Spending Has Grown in All Regions in Absolute Terms, But Trends in Spending as a Percentage of GDP Have Varied Real public education spending growth has been most rapid in East Asia and the Pacific, Latin America and the Caribbean, and Africa. While public education spending has grown in real terms in all geographic regions since the late 1990s, the rate of growth differs enormously (Figure 4). For example, since the late 1990s, spending in East Asia and the Pacific has more than trebled and in Africa it has increased by 120 percent. Trends in other regions were more muted. For example, public education spending in the Middle East and North Africa increased by only 12 percent over the same period. 4 Between 1998 and 2013, low-income countries and countries that graduated from that status allocated between 15 and 16 percent of total government expenditure to education. In 2014–17, education as a share of total government expenditure declined to 14 percent for countries that graduated but increased to 17 percent for non-graduating countries. 8 Figure 4: Government education spending has grown rapidly in East Asia and the Pacific but has not kept pace with economic growth in the Middle East and North Africa a. Public education spending b. Public education spending as a percentage of (Billions, constant 2015 PPP $) GDP 8 60 EAP 1998–2001 2002–05 2006–09 2010–13 2014–17 6 40 SAR ECA 4 20 LCR MNA 2 AFR 0 1998– 2002–05 2006–09 2010–13 2014–17 2001 0 AFR ECA LCR SAR EAP MNA Source: World Bank calculations using UIS and IMF online databases. Note: Statistics on public education spending are averages for countries in each region that have data available for at least one year in the four-year period. World Bank regional classifications are used to group countries as follows: AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, EAP = East Asia and the Pacific, MNA = Middle East and North Africa, and SAR = South Asia. Even as all regions have registered growth in government spending on education, the share of GDP they have allocated to education has varied widely. In 1998–2001, Africa and South Asia devoted a similar share of GDP to public education spending. However, by 2014–17, the percentage for Africa exceeded that for South Asia by 0.6 percentage point. Latin America and the Caribbean has also increased considerably the share of GDP used for public education spending since the late 1990s (Figure 4). In contrast, in the Middle East and North Africa, the share of government education spending declined from 5.4 to 4.5 percent of GDP over the same period, which explains the relatively slow absolute growth in such spending for the region since 2000. The available data show that patterns and trends in public education spending in countries that are classified as Fragile and Conflicted-Affected Situations (FCS) are similar to those in other countries at a similar stage of development (Box 3). 9 Box 3: Trends in Public Education Spending in Fragile and Conflict-Affected Countries Fragility, conflict, and violence can have a profoundly negative impact on a country’s education system and its ability to provide a quality education to all children. A lack of funding often exacerbates the difficulties faced by the governments of these countries (Commins 2017; UNESCO 2011). On the face of it, countries on the World Bank’s Fragile and Conflict-Affected Situations (FCS) list seem to devote similar shares of GDP to public education spending as other low-income and lower-middle-income countries, and to spend the money in similar ways. The figure below compares the public education spending aggregates for these three categories of countries. The data show that patterns and trends in public education spending for FCS countries are similar to those for other countries at a similar stage of development. Public Education Spending in FCS Countries is Similar to Funding in Other Low-income and Lower-middle-income Countries a. Public education spending b. Public education spending by education level, (% GDP) 2014–17 (%) 5 24% 20% 17% 4 27% 29% 42% 3 2 50% 51% 1 42% 0 2006–09 2010–13 2014–17 LIC FCS LMIC LIC FCS LMIC Primary Secondary Tertiary Source: World Bank calculations using UIS and IMF online databases. Note: World Bank income group classifications in 2017 are used to group countries and are as follows: LIC = low-income country, and LMIC = lower-middle-income country. FCS = country classified as a fragile and conflict-affected situation. The lack of data on financing in fragile and conflict-affected countries and the need to go beyond national aggregates to understand spending inequalities make it difficult to draw any firm conclusions on public spending on education in these countries. Fifty-five countries have appeared on the Fragile and Conflict-Affected Situations list since its introduction in 2006. Data on public education spending as a percentage of GDP are only available for approximately half of these countries at any one time. It is possible the countries that fail to report are those in which systems have broken down and, as a result, they spend less on education. Even when FCS countries have managed to report spending data, weaknesses in their public financial management systems make it more likely that the funds are not distributed fairly and are not used for their intended purposes. Education’s Share of Government Budgets Has Remained Relatively Stable over the Last 20 Years Governments must make difficult trade-offs when deciding how to allocate public funds to support a variety of priorities and policy objectives. Channeling more resources into education inevitably means channeling less into other sectors and it can be difficult for governments to take such action. In some cases, increased peace and stability–and corresponding declines in military spending–have allowed governments to devote a greater share of public spending to education. During its war with Eritrea, Ethiopia spent nearly one-third of the government budget on military expenditures, but this figure began to decline as the conflict wound 10 down, falling to 4 percent in 2017. Meanwhile, education expenditures as a share of the budget almost doubled, rising from 15 percent in 2000 to 27 percent in 2013. In other cases, reductions in fuel subsidies (e.g., Ghana and Indonesia) have allowed governments to allocate more to education (Van Der Burg and Whitley 2016). Overall, the share of government budgets allocated to education has remained relatively stable over the last 20 years across all regions and income groups (Table 1). Only in upper-middle-income countries has the share decreased. However, looking at regional averages, education expenditures as a percentage of government budgets have increased in Africa but declined in the Middle East and North Africa and in East Asia. Low-income and lower-middle-income countries tend to devote a greater share of government spending to the education sector than richer countries (Table 1). For example, in 2014–17, low-income countries spent an average of 16.4 percent of government funds on education compared to 12.6 percent for high-income countries. Regional differences tend to follow a similar pattern, with lower-income regions such as Africa devoting a greater share of government spending to education than higher income regions like Europe and Central Asia. Table 1: Share of Government Budget Used for Education, by Income Group and Region 1998–2001 2002–05 2006–09 2010–13 2014–17 Income Group LIC 15.0 15.8 16.0 15.9 16.4 LMIC 15.7 16.7 16.3 16.0 15.7 UMIC 16.0 15.5 14.9 14.9 14.5 HIC 13.1 13.2 13.1 12.7 12.6 Region AFR 14.8 16.2 16.6 16.7 16.1 ECA 11.9 12.2 12.2 12.4 12.2 LCR 16.2 15.7 16.1 17.3 17.4 SAR 15.5 16.3 14.1 13.5 15.0 EAP 16.1 17.2 15.4 14.5 14.0 MNA 17.3 16.5 15.4 12.9 13.6 Source: World Bank calculations using UIS and IMF online databases. Note: Income groups are defined by country income group classification in 2017. LIC = low-income country, LMIC = lower-middle-income country, UMIC = upper-middle-income country, and HIC = high-income country. AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, SAR = South Asia, EAP = East Asia and the Pacific, and MNA = Middle East and North Africa. Low-Income Countries Have Spent More on Primary Education, Richer Ones on Post-Primary Wealthier countries have devoted a greater share of government spending to post-primary education than less well-off countries. In 2014–17, less than one-third of public education spending in high-income countries went to primary education compared to about one-half for low-income countries (Table 2). This suggests that as countries become wealthier and their education systems expand and develop, additional public funding tends to be spent on post-primary education (Box 4).5 5 The correlation coefficient between per capita GDP and the share of public education spending going to secondary education is 0.4 and for tertiary education it is 0.3. This is statistically significant at the 1 percent level. 11 Box 4: Financing Education Development in the Republic of Korea The Republic of Korea has taken a progressive approach to universalizing primary and secondary education. In 1954, the government launched its first plan for free compulsory primary education financed by the Education Tax Act of 1958 and foreign aid. Between 1954 and 1959, public education spending as a share of the overall government budget increased threefold from 4.2 percent to 14.9 percent. Nearly 80 percent of this increase was devoted to primary education, and by the late 1950s and early 1960s, primary school enrollment was nearly universal (Kim 2002). After the expansion of primary schooling was complete, the government shifted the focus of its funding to secondary education (see figure below). Secondary education spending as a percentage of total public education spending surged from 21 to 39 percent between 1970 and 1987. This funding provided schools with enough resources to increase the secondary school gross enrollment rate from 39 percent in 1971 to 101 percent by 1996.a While primary public education spending as a proportion of GDP fell, absolute levels of public education spending at the primary level continued to rise in real terms over the same period. Since the universalization of secondary schooling in Korea in the late 1990s, the government’s focus has shifted, to expanding higher education. As Universal Primary Enrollment Was Achieved, Government Spending Shifted toward Secondary Education a. Education spending as a percentage of GDP b. Secondary school learning outcomes (Harmonized Learning Outcomes (HLO) scale) 7 Total 600 6 Primary Secondary 500 5 400 4 300 3 200 2 1 100 Mathematics Science 0 0 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 Sources: World Bank calculations using UIS, IMF online databases; Altinok, Angrist, and Patrinos (2018). Learning outcomes have improved alongside the continual increase in public education spending as a share of GDP. While information on secondary school learning outcomes only goes back to 1985, the data show an upward trend that corresponds to increases in spending on secondary education. More recently, average learning outcomes have remained relatively unchanged, even while public education spending on secondary education has continued to increase rapidly. While learning outcomes only measure one objective of the education system, this situation points to a potential lack of efficiency in the education system. a. The secondary gross enrollment rate is the number of children enrolled in secondary school divided by the number of children of official secondary school going age. 12 Over the last 20 years, the focus of public education spending in middle-income countries has shifted away from primary education towards secondary and tertiary education (Table 2), even as absolute figures for public spending have generally increased for all levels of education. In 1998–2001, lower-middle-income countries spent half of their public education resources on primary education, but in 2010–13, this share dropped to 42 percent. The shift of public education spending away from primary education has mainly benefited secondary education, with its share increasing from 33 to 42 percent over the same period. A similar trend is evident for upper-middle-income countries but not for low-income and high-income countries. However, since 2010–13, high-income countries have seen a shift in funding emphasis from secondary to tertiary education (Table 2). Table 2: Composition of Education Spending, by Income Groups and Regions (%) 1998–2001 2002–05 2010–13 2014–17 Pri. Sec. Tert. Pri. Sec. Tert. Pri. Sec. Tert. Pri. Sec. Tert. Income Groups LIC 51 29 20 58 25 17 51 28 21 50 27 24 LMIC 50 33 17 46 34 19 42 38 20 42 42 17 UMIC 43 36 21 43 38 18 36 42 22 35 41 24 HIC 31 45 23 30 46 24 31 43 26 31 41 28 Regions AFR 48 32 19 50 31 19 47 32 21 45 31 24 ECA 26 49 25 29 47 24 27 47 26 28 46 26 LCR 47 34 19 45 37 18 42 35 22 40 36 24 SAR 48 36 17 42 45 13 42 38 20 45 39 15 EAP 44 33 23 43 34 23 40 36 24 38 40 23 MNA 36 44 19 34 43 22 32 45 23 37 38 25 All countries 41 38 21 41 39 21 39 39 23 37 39 24 Source: World Bank calculations using UIS and IMF online databases. Note: The figures in the tables show the percentage of the public budget spent on each education level. Pri. = Primary, Sec. = Secondary, and Tert. = Tertiary. Income groups are defined by World Bank country income group classifications as of 2017. LIC = low-income country, LMIC = lower-middle-income country, UMIC = upper- middle-income country, and HIC = high-income country. AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, SAR = South Asia, EAP = East Asia and the Pacific, and MNA = Middle East and North Africa. How regions spend their public education funds often mirrors the spending patterns of the income group into which most of a region’s countries fall. For example, trends in public education spending in Africa and Europe and Central Asia tend to resemble spending patterns for low-income countries, while trends for Europe resemble those for high-income countries. Patterns and trends in South Asia appear to mirror those for lower-middle-income countries. Both Latin America and the Caribbean and East Asia and the Pacific have shifted spending from primary to post-primary education over the last 20 years, reflecting the regions’ strong economic growth and the resulting expansion of their education systems. The Middle East and North Africa has experienced strong shifts in public education spending from secondary to tertiary education even though public education spending in the region has seen limited growth in absolute terms. Spending Trends Can Vary Widely across Countries in the Same Region or Income Group These regional and income group averages mask large differences among individual countries. For example, Figure 4 (above) highlighted the large increases in public education spending’s share in GDP in Africa over the last 20 years, but government education spending as a share of GDP actually declined in some countries, such as Sierra Leone and Cabo Verde. There has also been significant variation in the magnitude of the increases. In Latin America and the Caribbean, the share of GDP allocated to public 13 0 1 2 3 4 5 6 7 8 0 2 4 6 8 10 12 Cambodia Armenia Lao PDR Azerbaijan Indonesia Kazakhstan Thailand China Romania percentage point. EAP Samoa Albania Mongolia Georgia Malaysia Vanuatu Bulgaria ECA Kiribati Turkey Bangladesh Russian Federation Pakistan Sri Lanka Belarus Source: UIS and IMF online databases. SAR Maldives Tajikistan Nepal Ukraine Bhutan Guinea Kyrgyz Republic Uganda Moldova Mauritania Dominica Gabon Madagascar Peru 14 Cameroon El Salvador Namibia 2014–17 Nicaragua Gambia Sierra Leone Paraguay Tanzania Colombia Rwanda St. Lucia Mali EAP = East Asia and the Pacific, and MNA = Middle East and North Africa. Benin Ecuador Congo, Rep. LCR Mexico AFR Comoros Jamaica Ethiopia Cote d'Ivoire Argentina Malawi St. Vin. & Gr. Mauritius Brazil Togo Cabo Verde Belize Kenya Costa Rica Burundi Niger Bolivia Ghana Iran Figure 5: Trends in Regional Averages Mask Country-Level Differences in Spending South Africa Jordan 2014-17 MNA Senegal Note: AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, SAR = South Asia, Public education spending as percentage of GDP (excluding high-income countries), 1998–2001 to Tunisia 1998 – 2001 Eswatini education in Brazil increased by nearly 2 percentage points, while in Colombia, it increased by only 0.6 2.3 Sources of Increase in Public Education Spending Identifying the main drivers of change in public education spending can provide important insights into how countries have mobilized additional funding for education. As we saw in the previous section, government spending on education has increased over the last 20 years in the aggregate and in many countries. Do these increases simply reflect growth in the overall government budget due to economic growth? Or are they the result of changes in the size of the public sector or in the priority that policy makers give to education in the government budget? We used a simple accounting identity to disaggregate changes in public education spending over the last 20 years to try to answer these questions (Box 5). Box 5: Decomposition of Public Education Spending into Main Fiscal Components A simple accounting identity is used to disaggregate changes in public education spending over the last 20 years into three component parts: ≡ (1) where T is total public education spending, Y is GDP, G is total government spending as a percentage of GDP, and E is total public education spending as a percentage of total government spending.a Using Taylor’s expansion, the percentage change in total public education spending can be decomposed as: (2) We ignore second and higher order terms of the Taylor expansion, which implies that our decompositions are approximations to the overall total change. These approximations are more accurate when proportional changes in total public education spending are small.b Where they are large, differences between the sum of the three components in equation 2 and actual changes in education spending can be large.c a. It would be possible to break down total government spending further to analyze revenue, borrowing, and development assistance, but we did not pursue this line of analysis in this paper. b. See Rolleston (2009) for a similar approach to decomposing trends in public primary education spending. c. See Section 3 for further discussion. Economic Growth Has Been the Main Driver of Increased Public Spending This decomposition shows that economic growth has been the key driver of public education spending increases (Figure 6). Using equation 2, economic growth accounted for 88 percent of global increases in government spending on education between 2000 and 2015. Except for the aftermath of the 2008 financial crisis, this period was characterized by global GDP growth of 2.9 percent per year, which accounts for most of the near-doubling of real public education spending over the same period. Income groups that registered the most rapid economic growth also had the fastest growth in public education spending. For example, low-income countries grew at 4.8 percent while upper-middle-income countries grew by 5.7 percent. Overall changes in the size of the government sector played a limited role overall, and only played an important role in low-income and upper-middle-income countries. Growth in overall public spending accounted for approximately 28 percent of the increase in education spending in low-income countries and 18 percent for upper-middle-income countries (Figure 6). In low-income countries, average public spending as a proportion of GDP increased from 13.3 percent in 1998–2001 to 19.5 percent in 2014–17 (Annex Table A2). However, overall public spending as a share of GDP still remains relatively low compared to other 15 income groups, and it is a key driver of the lower proportion of GDP devoted to public spending on education outlined in Figure 3 (see Box 6).6 Box 6: Key Drivers of Overall Levels of Public Spending Governments mainly finance public spending through tax and other domestic revenue streams, concessional aid, and net borrowing. The combination of sourcing for public spending will depend on a country’s specific context and overall level of development. On the whole, tax revenues tend to be higher in wealthier countries. For example, in 2014–17, average tax revenues for low-income countries were 15 percent of GDP compared to 19 percent in high-income countries. The gap in total government revenues (from tax revenues, social contributions, and other revenues such as rent, fees, and income from property, but excluding grants) is much bigger between these two income groups. In 2014– 17, average total government revenues were 31 percent of GDP in high-income countries compared to only 18 percent in low-income countries. Overall levels of deficit financing also differ among income groups. For example, in 2014–17 government budget deficits (net borrowing) were 3 percent of GDP on average in low-income countries compared to 1.4 percent and 2.6 percent in lower-middle and upper-middle income countries, respectively. Total Official Development Assistance (ODA) as a Percentage of GDP (%), 2002–2017 2002–05 2006–09 2010–13 2014–17 Income Groups LIC 1.1 1.1 0.9 0.8 LMIC 0.9 0.9 0.7 0.6 UMIC 0.5 0.6 0.6 0.5 HIC 0.1 0.1 0.1 0.1 Regions AFR 1.0 0.9 0.7 0.6 ECA 0.2 0.2 0.2 0.2 LCR 0.2 0.2 0.1 0.1 SAR 0.6 0.7 0.5 0.3 EAP 1.9 2.1 1.9 1.7 MNA 0.4 0.6 0.5 0.5 Source: World Bank calculations using OECD Creditor Reporting System and World Bank online databases. Note: Overseas development assistance is measured as gross disbursements, and figures only include countries that received some assistance. Income groups are defined by country income group classification in 2017. LIC = low-income country, LMIC = lower-middle-income country, UMIC = upper-middle-income country, and HIC = high-income country. AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, SAR = South Asia, EAP = East Asia and the Pacific, and MNA = Middle East and North Africa. While overall levels of official development assistance (ODA) have been declining over the last 15 years, such aid provides an important source of education funding for low- and middle-income countries (see table above). Given average levels of public spending on education, ODA on average made up about 21 percent of total public education spending in low-income countries. However, this average masks large disparities among different low-income countries. For example, in 2017 Malawi received ODA for education equivalent to 2.4 percent of GDP compared to Madagascar’s 0.4 percent. 6 Average public spending as a percentage of GDP in lower-middle income countries was 24 percent in 2014–17 (see Annex Table A2). 16 Changes in the priority that policy makers gave to education in government budgets also generally contributed little to the increases in public education spending. For example, upper-middle-income countries had the largest increases in public spending on education overall but devoted a smaller share of their government spending to education in 2015 than they did in 2000. The exception was low-income countries, where making education a higher priority in government budgets did lead to greater public education spending (Figure 6), although this represented a relatively small proportion of the overall increase. Over this period, the share of the budget devoted to the education sector increased from an average of 15 to 16 percent in those countries (see Table 1). Figure 6: Economic Growth Has Been the Main Driver of Increased Public Spending on Education Decomposition of public education spending changes (%), 1998–2017 350 %ge change between 2000 and 2015 300 250 200 150 100 50 0 -50 High-income Lower-middle-income Low-income Upper-middle-income Contribution of GDP growth to growth in education spending Contribution of share of education in total government budget to growth in educaiton spending Contribution of total government spending growth to growth in education spending Total change in education spending - 2000–2015 Source: World Bank calculations based on UIS and IMF data. Note: Average percentage growth in all variables calculated as the percentage change from the four-year average in 1998–2001 to the four-year average in 2014–17, for all countries with data available in at least one year across both four-year periods. Regions with the fastest growth in public education spending were those that experienced rapid economic growth and managed relatively large increases in government spending as a share of GDP. For example, about one-third of increased public education spending in the East Asia, Latin America, and South Asia regions was due to an increase in the size of the government sector (Table 3). Between 1998–2001 and 2014–17, government spending as a percentage of GDP increased from 20 to 26 percent in East Asia and the Pacific and from 20 to 24 percent in Latin America and the Caribbean.7 In contrast, the impact of economic growth on increased public education spending in the Middle East and North Africa was reduced because both government spending as a share of GDP and the share of government spending allocated to education declined. 7 Annex Table A2 presents a summary of patterns and trends in government spending as a share of GDP. 17 Table 3: Regional Decomposition of Public Education Spending Changes (%), 1998–2017 Real change in Source of changes in public education spending public education Economic Total public Education spending growth spending as a prioritization – share of GDP education spending as a share of the total government budget (1) (2) (3) (4) East Asia and Pacific 251 215 69 -33 Europe and Central Asia 42 34 6 2 Latin America and the Caribbean 137 86 37 14 Middle East and North Africa 15 37 -8 -14 North America 134 189 13 -68 South Asia 81 75 22 -16 Africa 116 91 15 9 Source: World Bank calculations based on UIS and IMF data. Note: Education spending is measured in PPP (constant 2015 dollars). Average percentage growth in all variables calculated as the percentage change from the four-year average in 1998–2001 to the four-year average in 2014–17, for all countries with data available in at least one year across both four-year periods. Column 1 shows the total percentage change in education spending. Columns 2 – 4 break down the overall change into its component parts (see Box 5). Column 1 is equal to the sum of columns 2 – 4 with small differences due to rounding errors. Boosting the priority given to education in government budgets contributed to increased public education spending in Africa and Latin America but not in many other developing regions. Giving a higher priority to education accounted for 8 percent of the increase in public education spending in Africa and 10 percent in Latin America. Giving a higher priority to education combined with fast economic growth and increases in government spending as a share of GDP resulted in these two regions having the largest regional increases in public education spending. In all other regions except Europe and Central Asia, the share that education claimed in government budgets declined between 2000 and 2015. 2.4 Expanding Places versus Increasing per Student Spending On the whole, public spending per primary and secondary school student has risen steadily in low-income and middle-income countries (Figure 7). However, the per student average has grown much more rapidly in wealthier countries and has led to a widening of the per student spending gap between low-income and middle-income countries. For example, in 1998–2001, lower-middle-income countries spent more than three times as much per primary school student than low-income countries ($522 in purchasing power parity dollars compared to PPP $136). By 2014–17 they spent almost 4.5 times as much as low-income countries (PPP $916 compared to PPP $207).8 The only exception to the upward trend in per student spending was public spending on secondary school students in low-income countries. Between 1998–2001 and 2014–17, per secondary student spending for this group declined by 19 percent from PPP $418 to PPP $339. Since low-income countries’ overall spending on secondary education rose over the same period, the decline in per student spending resulted from faster growth in the number of students attending secondary school.9 8 Annex Table A3 reports per student spending for regional groupings. With the exception of secondary education, spending per school-aged child follows similar trends to those discussed in the main text and are reported in Annex Table A4. 9 Annex Table A4 reports spending per school-aged child, which increased for secondary school children in low- income countries, unlike spending per student. 18 Figure 7: Real Spending per Student Has Generally Risen in Low-Income and Middle-Income Countries, but the Gap between Income Groups Has Widened Public education spending per student (constant 2015 PPP $), 1998–2001 to 2014–17 a. Primary education b. Secondary education 3,500 3,500 UMIC 3,000 3,000 2,500 UMIC 2,500 2,000 2,000 1,500 1,500 LMIC 1,000 LMIC 1,000 500 500 LIC LIC - - 1998–2001 2002–05 2006–09 2010–13 2014–17 1998–2001 2002–05 2006–09 2010–13 2014–17 Source: World Bank calculations using UIS and IMF online databases. Note: LIC = low-income country, LMIC = lower-middle-income country, and UMIC = upper-middle-income country. Recent Spending Increases Have Mostly Financed School Expansion in Low-Income Countries Increases in government funding for education can go towards expanding the number of student places, increasing the average amount spent on each student, or a combination of the two. To determine if countries are finding a balance between quantity and quality, it can be useful to explore how policy makers have used additional public education funding and how their country’s level of education development has influenced their choices. The data can also help to identify the transition points where countries shift spending priorities from increasing access to improving quality. These transitions are where spending inefficiencies potentially can increase if policy makers fail to pay due attention to how increases in levels of spending per student are used. To understand how governments have historically used funds over the last two decades, we broke down public education spending by education level (primary, secondary, and tertiary) and decomposed it into spending per student and enrollment (Box 6). 19 Box 6: Decomposition of Changes in Public Education Spending between Changes Due to Enrollment and Spending per Student A second simple accounting identity is used to disaggregate changes in public education spending over the last 20 years into two component parts: ≡ (3) where Tl is total public education spending for education level l, Sl is public education spending per student in level l and El is total enrollment in level l. In a similar way as in equation (2), the percentage change in total public education spending in level l can be decomposed as: , , , , , , (4) , In low-income countries, increased public education spending over the last 20 years has largely funded efforts to expand access to primary and secondary education rather than to increase the amount spent per student (Figure 8). In countries with complete data, we found that 66 percent of the total increase in spending at the primary level and 80 percent at the secondary level went towards expanding access (Table 4). This partly reflects the relatively low rates of access in these countries in the early 2000s. For example, between 1998 and 2001, the average net enrollment rate in primary school in low-income countries was 56 percent while the average rate in secondary schools was only 16 percent. Figure 8: Primary and Secondary Spending Growth in Low-income Countries was Driven by Increases in Access Decomposition of public education spending changes (%), 1998–2017 800 800 Percentage change 600 600 400 400 200 200 0 0 -200 -200 Madagascar Togo Togo Benin Chad Burundi Mali Nepal Burundi Benin Mali Nepal Chad Malawi Gambia Malawi Uganda Tanzania Rwanda Comoros Rwanda Burkina Faso Burkina Faso Niger Senegal Senegal Primary Education Secondary Education Estimated change in total spending due to changes in per student spending Estimated change in total spending due to changes in enrollment Total spending change over period Source: World Bank calculations based on UIS and IMF data. Note: Only countries for which there are data that span across at least four out of five periods (1998–2001, 2002– 05, 2006–09, 2010–13, and 2014–17) between 1998 and 2017 are included in the figure. For each country, spending is from the earliest four-year period for which data are available to the latest four-year period for which data are available. 20 Recent Spending Increases Have Gone to Providing More Spending per Student in Middle-Income Countries In middle-income countries where access to primary and secondary education started out much higher, increases in public education spending have tended to be used to boost the level of spending per student. Average enrollment rates in 1998–2001 in upper-middle-income countries were 92 percent for primary school and 62 percent for secondary school. This has meant that in upper-middle-income countries more than half of all the public spending increases in primary school and three-quarters of the public spending increases in secondary school was devoted to increasing the amount of funding spent on each student (Table 4). These increases in spending per student were often implemented as part of reform attempts by governments to improve the quality of education. As countries make the transition between spending for expansion and spending for quality, it is important to focus on the efficiency of public spending to ensure that increases in spending are used effectively to improve education outcomes. Table 4: Public Education Spending Changes by Level and Income Groups, 1998–2017 Low income Lower-middle Upper-middle High income income income Primary (total % change) 148 89 83 62 % devoted to access 66 30 38 26 % devoted to per student spending 34 70 62 74 Secondary (total % change) 241 121 162 43 % devoted to access 80 49 25 43 % devoted to per student spending 20 51 75 57 Tertiary (total % change) 167 147 118 136 % devoted to access 69 64 73 55 % devoted to per student spending 31 36 27 45 Source: World Bank calculations based on UIS and IMF data. Note: Only countries for which there are data that span across at least four out of the five periods (1998–2001, 2002–05, 2006–09, 2010–13, and 2014–17) between 1998 and 2017 are included in the figure. The percentage devoted to access and percentage devoted to per student spending sum to 100 percent of the total percent change over five-year episodes for each level of education. 2.5 Summary While the overall patterns and trends in public education spending vary widely across income groups and regions, we can draw some general conclusions from our analysis in this section.  Global averages: o Average government education spending doubled in real terms between 1998 and 2017, but spending as a share of GDP increased only slightly. o The composition of public education spending remained relatively stable over the same period. o Economic growth has been the main source of financing increases in government education spending.  Differences among regions: o Real government education spending has increased more than threefold in East Asia and the Pacific since 1999. o Government education spending as a share of GDP has declined in the Middle East and North Africa region. o Both Africa and Latin America and the Caribbean have seen rapid increases in terms of share of GDP going to government education spending. 21  Differences among income groups: o Low-income countries had the largest increases in terms of share of GDP allocated to government education spending. o Countries in wealthier income groups devoted a greater share of their public education funding to tertiary education and less to primary education. o Low-income countries have tended to use increases in their government education budgets to expand access to education while middle-income countries have increasingly used their additional funding to spend more on each student. 3. How Have Governments Secured More Education Funding and Improved Outcomes? The previous section focused on long-term trends in government spending on education. This section zooms in on specific country experiences on two fronts. First, we zoom in on changes in spending over a shorter time period. We use five-year periods, a time horizon that is more aligned with the average length of a political cycle and the medium-term planning cycles of ministries of finance. In looking at these episodes, we seek to provide guidance for countries based on the experience of other countries. In particular, we explore how countries have been able to mobilize resources for education over a five-year period, as well as the extent to which their initial conditions affected their pathway. Second, we explore how countries have translated these boosts in spending into changes in learning outcomes over the subsequent period, also exploring how this depends on initial conditions of the country. In deciding whether to increase spending on education, policy makers should consider the potential impact of such a move on learning outcomes. Countries may have the resources to boost their education spending, but that does not necessarily mean they will translate bigger budgets into better learning outcomes. Experience varies widely among countries, and a range of factors besides funding can help determine success or failure. These factors include, for example, the quality of existing teachers, a country’s governance structure, and the incentives associated with its financing of the education sector. However, even if the precise returns that a country will get are uncertain, one can estimate a range of possible outcomes based on the experience of other countries with similar levels of initial spending and education outcomes. This range of potential outcomes can serve as a helpful guide for policy makers as they seek to make decisions about education spending. Even if a country should increase education spending from an efficiency standpoint – that is, if it expects big returns from its increased spending – the political process must be aligned to ensure that a boost to funding materializes. The previous section identified prioritization of education in budget allocations as a driver of increased education spending. But even if relevant departments such as the ministry of finance agree to an increase, spending more on education means spending less on other priorities if the overall budget remains unchanged. Governments need to be able to justify such a move, and data showing a potentially positive impact of funding increases on learning outcomes can provide evidence to support arguments of this kind. The proposed approach to benchmarking spending increases and changes in outcomes in five-year periods seeks to inform these decisions in practice. We present this approach in more detail in Section 4. 3.1 Historical Evidence on the Magnitude and Pace of Resource Mobilization for Education We examine public education spending growth at the country level for every five-year period between 1999 and 2015. There were 852 instances in the database where data were available on a country’s total public education spending and its three underlying components (economic growth, aggregate public spending, and education prioritization) at both the beginning and the end of a five-year period. 22 Figure 9: Magnitude of Changes in Education Spending over Five-Year Periods a. Distribution of five-year growth episodes for total b. Distribution of five-year growth episodes for spending education spending, 1999–2015 (constant 2015 PPP as a % of GDP, 1999–2015 $) Source: World Bank calculations based on UIS and IMF data. The overall distribution of these 852 five-year growth episodes shows that spending has grown in most cases (Figure 9). Consistent with our findings from Section 2, real public education spending increased in 83 percent of these five-year growth episodes. Over all five-year growth episodes, public education spending grew by a median of 18 percent, but there was wide variation between countries and time periods. The changes ranged from a 71 percent decline in spending in Malta between 2003 and 2008 to a 166 percent increase in spending in Afghanistan from 2006 to 2011. However, public education spending as a proportion of GDP changed only slightly, with the median five-year episode being associated with a 0.1 percentage point increase in public education spending as a percentage of GDP. There was considerable variation in the characteristics of five-year changes in public education spending across and within income groups. The average increase in public education spending growth was much higher for the low-income group (39 percent) than the high-income group (13 percent) (Table 5).10 Real public education spending for countries in the top growth quartile11 grew by an average of 36 percent for high-income countries and 92 percent for low-income countries. Patterns among countries in the lowest growth quartile were less clear and ranged from an average decline of 4 percent for low-income countries to an average decline of 15 percent for the upper-middle-income group (Table 5). 10 Annex Table A5 includes details on the second and third quartiles. 11 Those country growth episodes in the 75th percentile in terms of the growth rate in real public education spending. 23 Table 5: Average Size and Decomposition of Five-Year Growth Episodes by Income Group, 1999– 2015 Real change Decomposition into sources of five-year growth: in public Economic Total public Education education growth spending as prioritization – spending in a share of education spending as five- year GDP a share of the total period government budget (1) (2) (3) (4) Low income 39 27 5 6 Slowest growth quartile -4 25 -10 -19 Fastest growth quartile 92 34 15 44 Lower-middle income 24 20 6 -2 Slowest growth quartile -6 12 -2 -16 Fastest growth quartile 57 26 23 8 Upper-middle income 18 17 0 1 Slowest growth quartile -15 12 -12 -14 Fastest growth quartile 48 20 10 17 High income 13 11 2 0 Slowest growth quartile -7 6 -5 -8 Fastest growth quartile 36 21 7 7 All low- and middle-income 26 21 4 1 countries Slowest growth quartile -8 15 -8 -16 Fastest growth quartile 63 26 17 20 Source: World Bank calculations based on UIS and IMF data. Note: Slowest and fastest growth quartiles refer to the 25th and 75th percentile growth rates within an income category. Column 1 shows the total percentage change in education spending. Columns 2 – 4 breakdown the overall change into its component parts (see Box 5). Column 1 is equal to the sum of columns 2 – 4 with small differences due to rounding errors. In Figure 9, we used actual percentage changes in spending, whereas in this table we use the results from the decomposition using logs, which explains the slight differences in some of the variables. Countries in the top quartile of public education spending growth almost universally derived growth from all three sources:12 they enjoyed fast-growing economies, their government spending accounted for a growing share of GDP, and they increasingly prioritized education over other components of their government budgets. In contrast, for countries in the slowest growth quartile, reductions in education spending were due to two main factors: a drop in aggregate public education spending as a percentage of GDP and a decrease in the percentage of the national budget allocated to education (Table 5). For example, low-income and middle-income countries in the fastest-growing quartile increased their public education spending by 26 percent due to GDP growth, 17 percent due to increased government spending, and 20 percent due to increased prioritization of education. In contrast, decreases in government spending as a share of GDP and in education spending as a share of total government spending contributed to an 8 and 16 percent decline, respectively, in spending among the slowest growth quartile. 12 We use the same approach, outlined in Box 5, to decompose overall spending changes. In some five-year country growth episodes, there are discrepancies between calculations on the change in total government education spending and those on the change in its three components (GDP, government spending as a percentage of GDP, and education spending as a percentage of government spending). However, the overall magnitude of these discrepancies is small. Overall, the average log change in total government education spending is 20 percent over five years, while the average sum of the log changes in the three component variables is 21 percent over five years. 24 Initial conditions of a country are key to explaining differences in the magnitude of spending changes over a five-year period. When we examined initial levels of government spending as a percentage of GDP and the share of government spending devoted to education, we saw that those countries registering the fastest growth in public education spending were frequently those that started with more fiscal space, that is, a low share of the government budget spent on education and with a relatively small share of government spending in GDP. For example, in low-income countries, education accounted for only 13 percent of the government budget during the fastest episodes of growth compared with an average of 16 percent for all growth episodes.13 In upper-middle-income countries, the fastest growth episodes occurred in countries with initial levels of total government spending accounting for 27 percent of GDP, compared to an average of 30 percent across all growth episodes. A simple regression analysis confirmed this negative relationship between initial government spending and changes in government spending (Table 6). Countries with higher initial levels of national income, higher rates of government spending as a percentage of GDP, and higher shares of education expenditures in the total budget had lower levels of public education spending growth during the five-year periods. Table 6: Association Between Changes in Education Spending and Initial Country Conditions Explanatory Variable (1) (2) Initial GDP -0.03*** -0.01*** (0.005) (0.005) Initial Gov’t Spend (% of GDP) -0.11*** -0.06*** (0.02) (0.02) Initial Educ. Spend (% of Gov’t) -0.17*** -0.24*** (0.04) (0.04) Income Group Fixed Effects No Yes Region Fixed Effects No Yes N 852 852 R-squared 0.13 0.21 Source: World Bank calculations based on UIS and IMF data. Notes: The table shows regression estimates of five-year change in total education spending in constant PPP-adjusted dollars (dependent variable) on initial country conditions. Standard errors in parentheses. * p<0.10; ** p<0.05; *** p<0.01. All variables are logged. 3.2 Evidence on How Governments Have Translated Spending into Education Outcomes In the previous section, we explored how countries have increased their spending and how this was influenced by their initial conditions. However, having the ability to mobilize resources for education is not enough to guarantee that those resources should be mobilized from an efficiency point of view or that they will be mobilized even if they should. So, the first question is: Should a country always aim for an ambitious increase in resources for its education sector? The answer is no. The desired level of public spending on education can depend on numerous factors, including the importance of competing demands on the public purse. However, a crucial consideration in determining whether a country should spend more on education, regardless of other demands, is the outcomes that can be realized from such an increase. Defining Our Outcome of Interest: Learning-Adjusted Years of Schooling (LAYS) The first step in answering these questions is to establish the outcomes and spending measures of interest. Should governments focus on increasing enrollments, improving learning, or both? Since we are interested in both access to education and learning, we used a measure that captures the levels of educational attainment and learning of the whole population (not just of those who are in school). We used the Human 13 See Annex Table A6. 25 Capital Index (HCI) data set developed by the World Bank and its measure of Learning-Adjusted Years of Schooling (LAYS). The LAYS measure is made up of two components:  Expected years of schooling. This measures the number of years of school a child born today can expect to obtain by age 18. It is based on age-specific enrollment rates between ages 4 and 17 and has a maximum value of 14.  Harmonized measures of learning from global and regional learning assessments (Altinok, Angrist, and Patrinos 2018; Patrinos and Angrist 2018). Relative scores for learning are calculated for each country by dividing its harmonized score by 625, which corresponds to the “advanced” benchmark on the TIMSS and PIRLS learning assessments. These two measures are multiplied together to arrive at the expected learning-adjusted years of schooling (LAYS) we use as our measure of education outcomes (Filmer et al. 2018; Kraay 2018). For example, if a country had 10 years of expected years of schooling and an average test score of 500 (80 percent of 625), then its LAYS would be 8 years. A crucial benefit of the LAYS measure is that it captures changes in both educational attainment and average learning outcomes. If a country maintains its average learning levels while increasing enrollment rates, our measure will increase since the average levels of learning of the population are now higher. If average learning levels increase because some children drop out of school and enrollments decrease as a result, the level of the LAYS will depend on the magnitude of those changes. Its limitations are mainly related to a lack of data in some countries. We propose to measure spending as total expenditure per child calculated as total spending in primary and secondary education divided by the total number of school-aged children in the country. This captures how much a country is spending on the population that it needs to serve (not just the population that it is currently serving). Thus, if a country is spending a lot on its students, but enrollment rates are very low, the country would effectively be spending too little on the population that it should be serving. Because it measures total spending in U.S. dollars (constant PPP-adjusted dollars), this measure is strongly correlated with per capita GDP. 26 Figure 10: Association Between Spending and Enrollment-Adjusted Learning Expenditure Per Child and Learning-Adjusted Years of Schooling (LAYS), 1991–2015 a. All countries b. Countries spending less than $2,500 per child (constant 2015 PPP dollars) (constant 2015 PPP dollars) 14 14 Bulgaria, 2005–10 12 12 France, 2010–15 Learning Adjusted Years of Schooling Learning Adjusted Years of Schooling 10 10 Moldova 2010–15 Cyprus, 2005–15 8 8 Colombia, 2010–15 6 6 Ecuador, 2010–15 Malawi, 2015 Madagascar, 2010–15 4 4 Burkina Faso, 2010–15 2 2 Chad, 2015 0 0 0 12,000 16,000 4,000 8,000 0 500 1,000 1,500 2,000 2,500 Spending per child ($) Spending per child ($) Source: World Bank calculations based on HCI, UIS and IMF data. Note: Calculations only incorporate countries with available data. There is a clear relationship between our measures of spending and outcomes, which change with the level of spending. Figure 10 plots our measures of spending and outcomes averaged for each country within five- year periods (1991–1995, 1996–2000, 2001–05, 2006–10, and 2011–15) and includes a stochastic frontier curve that estimates the maximum learning-adjusted years of schooling that is expected for a given level of spending per child.14 The frontier does not estimate the relationship between learning-adjusted years of schooling and spending per child, but the vertical distance from a particular country observation to the frontier provides a measure of how well the country is able to translate spending per child into education outcomes.15 However, there are large differences among countries at similar levels of spending, with some countries getting significantly better outcomes for the same (or lower) levels of spending. For example, in the right-hand panel of Figure 10, Malawi and Chad spent a similar amount per child but in Malawi this translates into five learning-adjusted years of schooling, while in Chad, it translates into only 2.6 years.16 These differences between a country’s actual and potential outcomes are usually referred to as efficiency, and this can be measured in different ways, one of which is by comparing a country’s position to the stochastic frontier line in Figure 10. We will come back to the discussion on efficiency in the next section. 14 We use Stata’s Frontier command to estimate the stochastic frontier and assume that the production frontier is linear in logs. 15 We also conducted Ordinary Least Squares regressions of LAYS on public education spending per child and found a positive and statistically significant relationship, both with and without a control for gross domestic product per capita. 16 In 2015, spending per child in Malawi was $176 and in Chad $229 (constant 2015 PPP dollars). 27 Mapping Pathways of Spending and Outcomes The ranges of outcomes that a country is likely to achieve as a result of changes in spending can be mapped to the four pathways shown in Figure 10.17 Each path is not equally desirable. Completely vertical moves would be desirable in countries with limited fiscal space as they would represent large improvements in outcomes resulting from small (or zero) changes in resources. Movement towards the upper left corner of the figure depicts examples of improvements in outcomes accompanied by decreases in spending, which may seem ideal from an efficiency point of view. However, as we saw in the previous section, reductions in real spending over a five-year period only happened in 17 percent of cases. In most cases, countries will move to the right on the spending axis (increases in resources), and in those cases, the steeper the line is, the better, as this shows a fast increase in outcomes. Completely lateral movements would represent increases in resources with no improvements in outcomes. The most undesirable scenarios are those where the movement in the figure is downward (regardless of resources), as these reflect deteriorating outcomes. In Figure 10, it is the change in outcomes relative to the change in spending that reflects the direction of a given path. In this paper we use elasticities to quantify how countries translate additional spending into outcomes, measured by the change in learning-adjusted years of schooling associated with a 1 percent change in spending. We use changes in both measures over a five-year period (with changes in spending lagged by five years), which is enough time to observe any significant changes in outcomes: % Spending elasticity of outcomes = % While elasticities are a useful way to measure the relationship between changes in resources and outcomes, their interpretation is not straightforward. Changes in both resources and outcomes can have positive or negative signs, which makes the elasticity hard to interpret on its own. In addition, when the change in the denominator is very small (or zero), the resulting elasticity is extremely large (or infinite), which can bias average results if not interpreted carefully, especially with such small samples. To facilitate the interpretation of elasticities, we separated the analysis of episodes with positive changes in spending from those with negative changes, which made the sign of the elasticity easier to interpret in each case. In addition, we focused on the desirability of different ranges of elasticities. The most desirable scenarios were those where outcomes improve more than spending increases (or outcomes improve and spending decreases). Less desirable scenarios are those where outcomes improve by less than resources increase (or outcomes deteriorate less than spending decreases). The least desirable scenarios are those where resources increase but outcomes deteriorate (or outcomes deteriorate by more than resources decrease). The interpretation of elasticities is presented in Figure 11, using LAYS as the education outcome. The color coding corresponds to the desirability of the scenarios. Episodes of spending decreases are green if outcomes improved (resulting in an elasticity smaller than 0, as in the left-hand graph). They are yellow if outcomes deteriorated by less than spending decreased (elasticity between 0 and 1, as in the left-hand graph), and red if outcomes deteriorated even more than spending decreased (elasticity larger than 1, as in the left-hand graph). In episodes of spending increases, episodes are coded as green if outcomes improved by more than spending increased (elasticity larger than 1, in green in the right-hand graph). They are yellow if outcomes improved by less than spending increased (elasticity between 0 and 1 in the right-hand graph) 17 We selected 5-year episodes to demonstrate the four different directions countries can take but pathways over longer periods of time for selected countries are described in Box 7. France is an example of a country where spending and LAYS declined over the five-year period selected. 28 and red if outcomes decreased (negative elasticity, in red in the right-hand graph). We next categorized all the changes across five-year episodes into the six categories depicted in Figure 11. Figure 11: Elasticities Provide a Way of Summarizing the Effect of Spending Changes on Outcomes Values of Spending Elasticities Depending on Direction of Spending Changes and Magnitude of Associated Outcome Changes Decreases in Spending Increases in Spending ϵ>1 ϵ<0 LAYS improved LAYS Δ in education outcomes Δ in education outcomes 0<ϵ<1 improved more than spending 0,0 LAYS deteriorated less 0,0 LAYS than spending improved less 0<ϵ<1 than spending ϵ<0 LAYS LAYS ϵ>1 deteriorated more deteriorated than spending Δ in Spending Δ in Spending How Have Countries Managed to Translate Changes in Spending into Changes in Outcomes? Looking at all the episodes in our data, we found that countries did not transform resources into outcomes very effectively. When we explored each country’s changes in spending and outcomes from one five-year period to the next, we found that green scenarios represented only 18 percent of episodes (Figure 12). Most of these episodes also resulted from reductions in spending with very small changes in LAYS (10 percent) as opposed to large improvements in outcomes resulting from increases in spending. In episodes where resources increased–which was the case for most of the episodes in our data set–outcomes improved more than resources in only 7 percent of cases. Outcomes improved by less than resources in more than half of these episodes, and in roughly a third, learning outcomes actually deteriorated. The low elasticities that we observed reflect the prevalence of less desirable scenarios. For episodes of spending increases, the average five-year elasticity of learning-adjusted years of schooling was 0.08, meaning that, for every 10 percent increase in resources, outcomes improved by only 0.8 percent. 18 This is a very low elasticity but similar to those that have been found in the health sector (Gallet and Doucouliagos 2017). However, the elasticity at the 75th percentile of the distribution was 0.20. The elasticity is also different when we break down the LAYS into its two component parts: expected years of schooling and harmonized test scores. A country could achieve the same improvement in LAYS either by improving 18 We excluded outliers in this calculation so that they did not distort the averages. We included episodes with changes in resources smaller than 0.01 percent in Figure 2, but we did not use them in our calculations of average elasticities. We also replicated this analysis on only those country episodes where the PISA learning assessment was used, and the results were not significantly different. 29 expected years of schooling while test scores remain constant or vice versa, so it is useful to break down these overall elasticities to illuminate which component is most responsible for gains in LAYS due to increased spending. The elasticity of spending on expected years of schooling was higher (0.04) than the elasticity of increasing resources on improvements in harmonized learning outcomes (0.02). This shows that countries have found it somewhat easier to transform increases in resources into higher educational attainment rates than into better learning outcomes. Figure 12: Outcomes Increased by a Greater Amount than Spending in Only 18 Percent of Cases a. Country episodes with spending decreases b. Country episodes with spending increases 36 17 7 2 5 0 LAYS improved LAYS LAYS deteriorated less deteriorated LAYS LAYS LAYS than spending more than deteriorated improved less improved more spending than spending than spending Source: World Bank calculations based on HCI, UIS, and IMF data. How Do a Country’s Initial Conditions Affect Its Ability to Transform Resources into Outcomes? While intuition suggests that a country’s initial conditions determine the most likely paths for education outcomes, their relationship is not straightforward. First, it depends on which outcome is in question. Transforming resources into wider access has proven to be easier to achieve than improving learning outcomes. Second, starting from a low spending point may result in larger or smaller changes in outcomes depending on how access and learning are determined. If inputs have decreasing returns (that is, the provision of additional inputs has a larger effect on outcomes when existing inputs are low), we would expect that low-spending countries would have larger expected returns. If, on the other hand, it takes a combination of various good-quality inputs to generate learning, and a single input has a very small impact, we would expect low-spending countries to have low returns. As countries spend more and cover more than the most basic needs, we would expect the marginal return to additional resources to decline. Third, the initial level of outcomes is also likely to drive future outcomes, but the effect is also ambiguous. While improving from a lower starting point may seem easier–suggesting that low-income countries should register larger gains in outcomes–existing inefficiencies in a country can also impact results and contribute to poor outcomes. Thus, pouring more resources into a system with inefficiencies may not yield any improvements in outcomes. We explore this question empirically later in this section. Four approaches are commonly used in the literature to measure the distance between the current level of outcomes and the potential level a country may be able to achieve. This is usually referred to as efficiency, and it is a measure of the ability of countries to transform resources into improved education outcomes.19 In the first approach, the residuals from an ordinary least squares (OLS) regression of education outcomes 19 For a fuller discussion of the advantages and disadvantages of the approaches outlined, see Grigoli (2014), Ravallion (2005), and Wagstaff and Wang (2011). 30 on levels of public spending provide a simple measure of the difference between a country’s actual and expected level of education outcomes at its current level of spending. Second, data envelopment analysis (DEA) estimates a non-parametric frontier that maps out the best possible education outcomes for a given level of spending.20 A country’s efficiency level is then measured by the vertical distance between its current position and the frontier. Third, in a similar way to data envelopment analysis, stochastic frontier analysis (SFA) estimates a frontier using a regression model but uses a specified functional form and assumptions about the distribution of the error term (see for example, Jayasuriya and Wodon 2003). The fourth and final approach, based on (Wagstaff and Wang 2011), aims to address a number of limitations inherent in the DEA and SFA efficiency measures while at the same time incorporating the best elements of each approach. It uses a grid search to identify the most efficient countries within specific ranges of spending. It measures a country’s efficiency by the distance between its own outcomes and the closest “efficient” country in terms of spending. These different measures of spending efficiency are all highly correlated and tend to rank countries in similar ways. In the main text of the paper, we used the efficiency measures derived from the stochastic frontier analysis approach as this is most highly correlated with other efficiency measures.21 To check that these estimates are effectively measuring efficiency, we looked at their correlations with some common proxies for efficiency (see Annex Table A9). These generally confirm that the estimates are on the whole measuring efficiency appropriately. Patterns and Trends of Spending Changes and Outcomes We used each country’s level of spending and efficiency scores to create four categories for countries based on where they were in the distribution of spending and efficiency.22 The categories were: (i) high-spending, high efficiency, (ii) high spending, low efficiency, (iii) low spending, high efficiency, and (iv) low spending, low efficiency.23 We then calculated elasticities for each category. The elasticities are presented in Table 7. The elasticities varied significantly by country category. The elasticity of learning-adjusted years of schooling is highest for low-spending countries that are far from the efficiency frontier. Empirically, those are the countries that have experienced the fastest increases in learning-adjusted years of schooling in the past. As countries get closer to the efficiency frontier, their elasticity decreases, indicating that they tend to follow the frontier (that is, the maximum level of outcomes countries at the same level of expending have been able to achieve). For high-spending countries, the trends are similar. For those that started at a high efficiency point, increases in spending have been accompanied by small increases in learning-adjusted years of schooling, whereas high-spending countries that are far from the efficiency frontier have experienced higher elasticities of learning-adjusted years of schooling. 20 In this paper, we used output-based measures of efficiency since the primary policy focus of most countries was on improving outcomes from existing levels of resources rather than looking at ways to reduce funding while maintaining existing levels of education outcomes. 21 Results using the other efficiency measures are available from the authors on request. 22 Median values are used to categorize countries. 23 We define low (high) spending and low (high) efficiency as below (above) average spending and efficiency respectively. 31 Table 7: Median Elasticities by Initial Spending and Efficiency Levels Country episodes with spending increases Low-Spending High-Spending Total Learning-adjusted years of schooling High-efficiency 0.01 0.04 0.02 Low-efficiency 0.19 0.14 0.15 Total 0.10 0.07 0.08 Learning High-efficiency 0.01 -0.001 0.0001 Low-efficiency 0.05 0.04 0.05 Total 0.04 0.001 0.02 Expected years of schooling High-efficiency 0.02 0.04 0.03 Low-efficiency 0.07 0.09 0.08 Total 0.04 0.05 0.04 Source: World Bank calculations based on HCI, UIS, and IMF data. The elasticities of learning outcomes and expected years of schooling also vary by category but follow similar trends. Low-efficiency countries have higher elasticities on both indicators, with elasticities declining as countries approach the frontier. For high-efficiency countries, the elasticities of both indicators are small, indicating that, in those countries, spending more resources on education has led to minimal improvement in outcomes. The differences in elasticities based on high or low spending are less pronounced for these two indicators than they are for LAYS. This analysis suggests that while countries are on average not very effective in transforming resources into outcomes, they are more effective if they start from a low-spending and low-efficiency point. The average elasticity for LAYS is only 0.08, but for low-spending, low-efficiency countries, the average elasticity is 0.19 (Table 7). For those countries, increases in spending have historically resulted in significant improvements in outcomes. For example, with a median LAYS of 6.3 years and a median elasticity of 0.19, these low-spending, low-efficiency countries would need to increase spending by 159 percent to achieve the upper-middle-income country median of 8.2 years–a large but achievable spending increase. On the other hand, the median low-income or lower-middle-income country, which also has a median LAYS of 6.3 years but an elasticity of 0.10, would have to increase spending by more than 300 percent to achieve the same level of outcomes. Similarly, high-spending, high-efficiency countries have experienced small improvements, or in some cases even declines in outcomes, following increases in spending. The results of this analysis could have important implications for policy. First, they imply that boosting public education spending in countries with low spending and poor outcomes is likely to yield significant improvements in outcomes, even if the country is inefficient. This would imply that such countries that have the fiscal space to increase their public education spending should do so as their outcomes are likely to improve. Second, as countries get closer to the efficiency frontier, it becomes increasingly difficult to transform resources into outcomes and thus these countries will tend to follow a flatter path in Figure 10. While this does not mean that such countries should not spend more on education, it does signal the need for policy makers to make greater efforts to increase efficiency if they want to improve outcomes significantly. Even if these countries have the ability to increase spending, any growth in resources should be accompanied by explicit efforts to increase the efficiency of education spending. Based on countries’ previous experiences, these efforts are only likely to be effective if they fundamentally change the way that 32 education resources are spent and therefore push the efficiency frontier outward, rather than continuing with “business as usual.” Third, countries that are at low-efficiency levels may also be able to increase the efficiency of their education spending, and in some cases this approach has allowed countries to achieve vertical movement, that is, improvements in education outcomes without increasing spending. Because of our data limitations, these results need to be interpreted cautiously.24 First, our sample is small. The episodes in our data set include only 67 country episodes that have complete educational-attainment, learning, and spending data in consecutive five-year periods. Second, those countries with complete data on the three indicators also tend to have better outcomes on average, so our analysis may be biased because we focused on relatively high-spending and high-performing countries. Thus, it will be important to expand this analysis to include more countries at the lower end of the spending and education outcomes distributions. 24 We also carried out the same analysis on an alternative measure of education outcomes that enabled us to look at 130 country episodes. The measure multiplied a combined primary and secondary gross enrollment ratio by a country’s harmonized learning score. The magnitude of the elasticities was similar, and the overall patterns described in the paper are not materially affected by using this alternative measure of education outcomes. 33 Box 7: Spending and Outcome Pathways for Countries with Different Spending, Efficiency Levels Our analysis of elasticities and pathways can be illustrated by six country examples: two relatively low-spending countries with different levels of efficiency (Colombia being efficient and Peru inefficient), two high-spending countries with different levels of efficiency (France being efficient and Kuwait inefficient), and two countries that decreased their spending (Argentina and Burkina Faso). As the figure below shows, Peru managed to move almost vertically between 2000 and 2010. Spending increased relatively little (from PPP$630 per child to PPP$916 per child), but outcomes improved significantly, moving Peru closer to the efficiency frontier. However, as Peru reached that frontier between 2010 and 2015, spending increased significantly but outcomes did not improve commensurately. For Colombia, the starting point was $1,250 per child in 2000, and the country was closer to the frontier than Peru. While outcomes did improve in Colombia, they did so at a slower pace than in Peru during the same period. In fact, Colombia seems to have followed the frontier trend line in its path over the past 15 years. Low-Spending, Low-Efficiency Countries Can Improve Their Education Outcomes Significantly Expenditure Per Child and Learning-Adjusted Years of Schooling (LAYS), 1991–2015 (constant 2015 PPP dollars) a. Peru b. Colombia 14 14 12 12 Learning Adjusted Years of Schooling Learning Adjusted Years of Schooling 10 10 2015 8 2015 8 2005 2010 2010 6 6 2000 4 4 2 2 0 0 0 5,000 10,000 15,000 0 5,000 10,000 15,000 Spending per child ($) Spending per child ($) In the case of high-spending countries, the elasticities show that some countries, including France, have not experienced significant improvements in outcomes as a result of increasing their spending (see figure below). While France increased its per child spending from $8,300 to $9,200 between 2000 and 2015, its outcomes deteriorated. Meanwhile, Kuwait increased its per child spending from $12,000 to $14,000 which resulted in only a small improvement in outcomes, leaving Kuwait still far below the efficiency frontier. 34 Increases in Spending for High-Spending Countries Has Limited Impacts on Education Outcomes Expenditure Per Child and Learning-Adjusted Years of Schooling (LAYS), 1991–2015 (constant 2015 PPP dollars) c. France d. Kuwait 14 14 12 2010 12 Learning Adjusted Years of Learning Adjusted Years of 10 2000 2005 2015 10 8 8 Schooling Schooling 2015 6 6 2010 4 4 2 2 0 0 0 5,000 10,000 15,000 0 5,000 10,000 15,000 Spending per child ($) Spending per child ($) In the case of countries that decreased their public education spending, the elasticities show that some countries are able to increase education outcomes even while reducing spending, while others are able to maintain high levels of outcomes with lower spending. Argentina reduced its spending per child from $2,300 in 2000 to $1,900 in 2005, a decrease of 15 percent, while its outcomes increased slightly, by 1 percent. In contrast, Burkina Faso decreased its spending by 17 percent from 2010 to 2015, while its outcomes improved by 26 percent. In both cases, these countries increased their spending efficiency, although the impact on outcomes was very different. Decreases in Spending Have Varied Effects on Education Outcomes Expenditure Per Child and Learning-Adjusted Years of Schooling (LAYS), 1991–2015 (constant 2015 PPP dollars) e. Argentina f. Burkina Faso 14 14 12 12 Learning Adjusted Years of Learning Adjusted Years of 10 10 8 2000 8 2005 Schooling Schooling 6 6 4 4 2015 2 2 2010 0 0 0 2,000 4,000 0 2,000 4,000 Spending per child ($) Spending per child ($) These six examples illustrate the possible pathways that countries can take in funding education. The figures also show the benchmarking that is possible with these elasticities. The benchmarking of resource mobilization can be mapped to the x-axis in the figures above, while the expected pathways for outcomes can be mapped to the y-axis. If data were available for all countries and all years, this would allow countries to compare where they stand in relation to other countries and to chart a pathway to achieve their desired goals in terms of spending and outcomes. This would enable policy makers to have much more informed discussions about what their countries can and cannot achieve given their starting points and would guide them in making policy decisions on education spending. 35 4. Benchmarking Changes in Spending and Outcomes: A Guide for Policy Making Determining how much to spend on education is in part a political decision. While education experts and policy makers might always advocate for more resources to be spent on education, finance ministers must balance competing fiscal demands and priorities for the country. Roads, health, sanitation, and infrastructure all require resources, and countries with low educational outcomes generally suffer from deficits in many of these sectors. Thus, when governments are deciding whether to give more funding to education as opposed to other sectors, they need to consider both the country’s capacity to mobilize these additional resources and the potential effect of such a move on learning outcomes. Whether the education sector ultimately gets the funding can also depend on the education minister’s ability to lobby effectively for these resources, which in turn depends on his or her ability to commit to specific results from a spending increase. The objective of this benchmarking exercise is to establish the basis for a tool that allows countries to guide these spending decisions. It seeks to help countries compare themselves to reasonable benchmarks of changes in spending and educational outcomes based on their initial conditions. In order to benchmark spending increases, this section uses a simplified fiscal space framework to guide these types of decisions and provide country-specific benchmarks for changes in spending and changes in outcomes. These references for potential increases in spending and outcomes are based on similar country episodes and are then used for (i) benchmarking education spending for a specific country, both in terms of level of, and potential to increase, resource mobilization; and (ii) benchmarking changes in outcomes associated with increases in resources based on the country’s initial conditions. These benchmarking approaches are then used to assess the need and ability of specific countries to increase public financing for education in a financially sustainable, efficient, and equitable way (Figure 13). Figure 13: Main Components of Fiscal Space for Education and Link to Outcomes Education specific gov. revenue General Government Revenues Spending Efficiency and Effectiveness ↓ ↓ ↓ ↓ Education Outputs Public expenditure X Education Share of GDP per Public Education → share of GDP Public Expenditure X Capita = Spending per Capita and Outcomes (e.g., LAYS) ↑ ↑ ↑ ↑ External Assistance Deficit Financing Other determinants (e.g., income, health) Source: World Bank adaptation of diagram in Tandon (2017). Note: Shaded boxes highlight the key components of fiscal space used in the approach outlined in this section. In practice, this approach is designed to provide information that can help policy makers answer a series of questions to guide their decisions on whether to mobilize additional public resources for education. These questions include: - How does the level of spending per student in my country compare with spending in similar countries? o Is my country spending too much or too little? 36 o Is my country’s level of spending explained by:  Size of the economy or economic growth?  Size of government (share of government revenues in GDP)?  Prioritization of education in the government budget? - How fast can my country mobilize additional resources: o If I follow the average country? o If I follow countries with the fastest spending growth? - What are the policy levers that I can move to increase spending? o Is prioritization of education in the budget low, thus requiring only a change in priorities? o Does the increase in spending depend on the overall government budget and thus needs a fiscal reform? - What is my country likely to achieve in terms of outcomes from this increase in spending: o If I compare myself to the countries with average returns? o If I compare myself to the countries with the highest returns? o If I compare myself to the countries with the lowest returns? Jointly, all these questions guide the decision of how much education spending should increase. They also require accountability from the education sector by asking: What should I expect from this increase in resources? The approach outlined in this section is not expected to provide a complete country-level education financing strategy. It uses broad public finance indicators to identify whether a country has the fiscal space to mobilize additional public resources for education. Ideally, the results from this exercise would serve as a starting point for a more detailed country-level assessment of whether and how additional funds should be diverted to the education sector. For example, the approach outlined will identify whether a country has the potential to increase public spending as a share of national income, based on a comparison with similar countries. However, a more detailed country assessment would be needed to determine whether increases in public spending are feasible and whether they could be financed through a growth in tax revenues, borrowing, or increased external assistance (see Figure 13). The approach has some limitations due to the poor quality of existing spending data. If we had complete data on spending and outcomes for all countries, this approach would be able to provide much more realistic scenarios on the feasibility of adjusting resource levels and improving outcomes in all countries. As highlighted in Section 2, there is only limited information on private education spending and this can affect the conclusions that can be drawn. For example, trends in outcomes will be affected not only by changes in public spending but also by changes in private education spending. Excluding these changes in private spending has the potential to affect elasticities and the conclusions that can be drawn. Moreover, the quality and coverage of public education spending data also limited the number of countries that could be included in our analysis as well as the precision of the benchmarks that could be estimated. It is expected that as more and better data become available, the benchmarking approach outlined in this section can be further refined and improved. Despite these limitations, the analysis performed on existing data highlights how the proposed approach can help guide spending decisions on education. 37 Benchmarking Spending It is not difficult for finance ministers to determine whether their country is spending too much or too little on education relative to comparable countries. The UNESCO Institute of Statistics has information for most countries about their total public education spending, the share of GDP or government budget spent on education, and their education spending per student. Choose comparable countries and the minister will have the answer. For countries to establish the path forward, on the other hand, is not so straightforward. If a country is spending below what comparable countries are spending, how fast can that country expect to increase its resources? What are the policy levers that need to be pulled to achieve that level of spending? The experience of countries that have registered the fastest growth in resources, as well as those that have invested the most in education, can provide additional useful information for other countries assessing the fiscal space they have to mobilize further public resources for education. The former countries’ experiences provide a benchmark for how much a country with similar conditions can expect to grow its resources and how quickly, while the latter countries’ experiences provide a benchmark for what level of spending a country with similar conditions may expect to achieve. In this exercise, we use the 75th percentile of both the level of spending and the change in spending as benchmarks for high spending and fast growth. For example, Table 8 shows that over the last 20 years, only a quarter of all low- and middle-income countries worldwide have devoted more than 19 percent of their overall government budgets to education, and it seems unlikely that most countries would be able to devote a larger share over the long term. A 19 percent share of education spending in the government budget may therefore provide a benchmark for what above-average, education-prioritizing countries have spent in the past, and against which other countries that wish to increase their prioritization of education could set their own efforts. Similarly, only a quarter of low- and middle-income countries have government budgets that account for more than 28 percent of gross domestic product, so governments that wish to increase their public spending might consider this as a benchmark to aim for. Table 8: Spending Growth of High-Spending Countries Real Public education Share of education in Share of total education spending as a % of total government government budget spending GDP budget as % of GDP growth (%) 75th 75th 75th 75th 75th 75th 75th percentile percentile percentile percentile percentile percentile percentile of five- of five- of five- of five- year year year year average percentage percentage percentage growth point point point growth growth growth Low- and middle- 43 5.5 0.6 19 1.6 28 3.2 income total UMICS 32 5.3 0.6 19 1.4 30 3.0 LMICs 41 6.3 0.6 20 1.6 26 3.8 LICs 60 4.8 0.8 19 2.4 22 2.4 Source: World Bank calculations based on UIS and IMF data. Note: The 75th percentile is defined separately for each indicator. The averages for the share of education and the total government budget are based on the country period averages for 2010–13 and 2014–17. LIC = low-income country, LMIC = lower-middle-income country, and UMIC = upper-middle-income country. 38 Table 8 also shows that the growth rate of public education spending for low-income and middle-income countries at the 75th percentile of the five-year growth episodes was 43 percent. The specific growth rates for the 75th percentile countries varied by income group, with the fastest rates recorded in low-income countries. These growth rates can provide one potential benchmark for countries to consider when developing targets for their resource-mobilization efforts. While the 75th percentile’s growth rate may provide a reasonable benchmark for an optimistic scenario for resource mobilization, just considering this one figure is not enough. Increases in public education spending typically occur as a result of a variety of factors, including faster economic growth, changes in the size of the public sector as part of GDP, and the priority policy makers give to education in the government budgets (see Figure 13).25 Breaking down the sources that fuel episodes of growth into these component parts and identifying the countries with the largest contributions from each component can provide more specific information for policy makers to answer the question: How did fast spending growth countries mobilize additional resources for education? However, a country’s capacity to mobilize further resources also depends on its initial fiscal situation. Education often must compete against other sectors for scarce resources, and the overall share of education in government budgets cannot rise indefinitely. Similarly, the share of government spending as a percentage of GDP cannot rise without increasing tax revenues, external grant assistance, government borrowing, or a combination of these. How much resource mobilization can be achieved over a five-year period also depends on the rate at which additional resources can be raised and absorbed into the education sector. For countries that are starting significantly below the 75th percentile benchmarks above, it is unlikely that they would be able to achieve these benchmarks all at once. Table 8 shows the percentage point increases in the overall size of the government sector as a percentage of GDP and in the share of education in the government budget for 75th percentile countries over a five-year period. For example, low-income and middle-income countries that ranked in the 75th percentile in terms of prioritizing education increased education spending as a share of the government budget by 1.6 percentage points over a five-year period, and those ranking in the 75th percentile in terms of increasing government spending as a share of GDP saw an average 3.2 percentage points rise in education spending over a five-year period. These could serve as additional benchmarks for the progress that countries can expect to achieve in a single five-year period. The combination of a country’s initial fiscal situation, the levels of spending of high-spending countries, and the rate of improvement registered by fast-spending growth countries can provide a useful benchmark for a country gauging its potential for increasing public education spending. The first step for policy makers seeking to develop a realistic plan for future growth of public education spending is to compare their country’s initial fiscal situation with those of high-spending countries. If the size of the government sector for their country and the share of their budget earmarked for education are smaller compared to those of the high-spending countries, then the country may have space to increase spending. The second step is to establish the magnitude of any possible increase by looking at how quickly fast-spending growth countries were able to expand their overall government spending as a share of GDP and the share going specifically to education. Besides setting such benchmark targets, countries will also need to consider their initial conditions, their own unique situation and priorities, as well as the experience of high-spending and fast- spending growth countries. Another option would be to set a less ambitious benchmark target by looking at the levels of spending and spending growth rates of the average country over all five-year growth episodes.26 These two benchmarks can provide policy makers with a useful range to assess their country’s prospects for mobilizing future resources. However, these are just one set of potential benchmarks. Policy 25 We do not include economic growth in this section since our focus is on exploring the potential for governments to make fiscal adjustments to mobilize additional resources for education. 26 Annex Table A7 includes the same information as in Table 8 for the median country. 39 makers could choose other country groupings (e.g., regional competitors or resource-rich countries) that may provide more appropriate benchmarks for future resource-mobilization efforts. Benchmarking Expected Outcomes from Spending Changes These spending benchmarks would provide the basis for how much a country could expect to mobilize in a reasonable period and what a reasonable target would be for its main policy options (the share of government expenditure over GDP and the share of the budget going to education). The next step is to establish whether the country should indeed increase its education funding and how quickly, by looking at the expected return from that investment in terms of outcomes. We use a similar approach as the one proposed above to benchmark the way countries translate an increase in resources into outcomes. We propose using outcome elasticities of spending increases to benchmark these changes. In particular, we propose using ranges of elasticities of country-periods with comparable initial conditions to estimate the expected range of outcomes for specific countries. Given our current data limitations, we only attempt to use four categories of countries for comparison based on their level of spending and their efficiency: (i) high spending, high efficiency, (ii) high spending, low efficiency, (iii) low spending, high efficiency, and (iv) low spending, low efficiency. We do this to illustrate the differences in elasticities. With much more complete and accurate data, we could estimate elasticities for other subgroups much more precisely. Having these ranges of potential outcomes based on real country examples would be helpful for both education and finance ministers in their budget deliberations. For education ministers of countries with high expected returns (that is, where initial conditions are those where countries in the past have experienced large improvements in outcomes after an increase in resources), this data could be a powerful tool in budget discussions, increasing the likelihood that these countries would increase education funding. In countries with low expected returns, the data could serve as a powerful accountability tool and help spearhead reforms to improve the efficiency of spending, especially in situations where the country is close to the efficiency frontier. In both cases, the resulting allocation of resources is likely to be more efficient than it is now. Putting It All Together: An Example To explain how the proposed approach might work in practice, we explore Sierra Leone’s current spending on education to assess i) what may be possible in terms of mobilizing additional resources for education, and ii) what the government might expect from any increases in public spending on education. Average public spending on education in Sierra Leone was only 1.6 percent of GDP in 2015 compared to a low-income country average of 4.3 percent. Sierra Leone’s low education spending was a result of the relatively small percentage of government spending in the country’s GDP as well as the low priority given to education in the government budget. Specifically, Sierra Leone devoted 13.8 percent of government spending to education while the low-income country median was 16.2 percent and countries at the 75th percentile in terms of the share of their budgets devoted to education was 18.7 percent (Table 9). Similarly, government spending in Sierra Leone was 11.8 percent of GDP, compared to a low-income country median of 15.6 percent and 21.9 percent for countries at the 75th percentile. This suggests that Sierra Leone has space to increase its education spending by expanding government revenue and making education a higher priority in the budget. 40 Table 9: Financing Parameters in Sierra Leone, 2014–17 Averages Share of total Share of education in government spending as total government % of GDP spending Sierra Leone 11.8 13.8 Median for LICs 15.6 16.2 75th percentile for LICs 21.9 18.7 Source: World Bank calculations based on UIS and IMF data. Note: LIC = low-income country. How rapidly could Sierra Leone increase its public spending on education? The underlying drivers of public education spending growth can provide useful benchmarks. It is not realistic to expect Sierra Leone to reach the levels of the 75th percentile countries in five years. Table 8 shows that over a given five-year period, fast-spending-growth low-income countries expanded the overall size of their government sector as a share of GDP by 2.4 percentage points and increased education spending’s share of the government budget by 2.4 percentage points. These figures show that these rates of growth are feasible within five years and suggest that Sierra Leone should be able to increase its public education spending from 1.6 to 2.3 percent of GDP over a five-year period. A less ambitious expansion path would be to try to emulate median countries rather than 75th percentile countries. In this case, Sierra Leone would aim for the average growth of spending in low-income countries by increasing its public education spending as a share of GDP from 1.6 to 1.7 percent over five years (Figure 14a). These two potential pathways provide a plausible range of targets within which Sierra Leone could increase its public education spending over the next five years. Using school-age population projections, these possible spending paths translate into a range of per child funding of between $166 and $218 in 2020/21 compared to $126 per child in 2015/16 (Figure 14b). Are these spending increases worth pursuing for Sierra Leone? Based on the past record of low spending in inefficient countries (see Table 8, Share of total government budget as % of GDP), spending increases in this range have on average resulted in an increase of approximately one-third to two-thirds of one year of learning-adjusted years of schooling (Figure 14c). This is not to say that just by increasing spending these improvements in education outcomes are automatic, only that this is the average conversion of spending into resources for countries like Sierra Leone. 41 Figure 14: Sierra Leone Has Space to Expand Education Spending and Improve Outcomes Potential Financing Pathways for Sierra Leone a. Total education spending, b. Education spending per child, c. Learning-adjusted years of $ constant PPP (millions) $ constant PPP schooling 650 250 5 2.3% 218 4.7 200 4.4 550 166 4 4.1 150 450 1.7% 100 126 3 350 50 1.6% 250 0 2 2015/16 2020/21 2015/16 2020/21 2015/16 2020/21 Average growth (median) Average growth (median) Average growth (median) Ambitious growth (75th percentile) Ambitious growth (75th percentile) Ambitious growth (75th percentile) Source: World Bank calculations based on HCI, UIS, and IMF data. Note: The data labels in panel (a) show education spending as a percentage of GDP. In the Sierra Leone example, we have used high-spending and fast-spending-growth low-income countries as benchmarks for gauging the scope for increasing public education spending over five years. However, other benchmarks may be more relevant for specific countries and for more detailed analysis. For example, Sierra Leone has recently come out of a protracted civil war, so benchmarks associated with similar post- conflict countries might be more appropriate in this case. The World Bank is currently developing a benchmarking tool that includes common comparison groups as well as the possibility for analysts to choose specific groups. As countries run out of fiscal space or approach the efficiency frontier, the room for increasing funding for education will be limited. In the case of inadequate fiscal space, our benchmarking tool would show that a country’s resources could not increase dramatically, and thus improvements in outcomes would need to come from efficiency gains while policy makers increase fiscal space through tax reform or other means. In this case, the tool may help make the case for broader tax reform, for example. Within the education sector, however, the tool would suggest the need for efficiency gains: getting better outcomes out of existing resources. Similarly, and regardless of fiscal space, for countries approaching the efficiency frontier and not expecting large gains in outcomes, the tool would also underscore the need to increase efficiency even if resources are increased. 5. Spending More and Spending Better In this paper, we have shown that government spending on education has increased significantly around the world over the last 20 years, with low-income countries registering the highest rates of growth. Economic growth has been the key driver of these gains in most countries. Other factors that have played a smaller role include the expansion of government spending as a share of GDP and an increase in the share of government budgets being allocated to education spending. Over a single five-year period, all countries with available data between 1998 and 2017 have boosted public spending on education by an average of 21 percent. Episodes of more rapid growth have been associated with rapid increases in the prioritization of education in the budget. These rapid-growth periods have also tended to occur in countries that share a similar set of initial conditions, including budgets that devoted a smaller percentage of funds to education and government budgets that took up a smaller percentage of GDP. 42 With this rapid growth in global public education spending–fueled in part by investments by low-spending countries– education systems around the world today are enjoying greater levels of funding than two decades earlier. While countries have been relatively effective at mobilizing more resources for education, they have been less effective in transforming these funding increases into improved learning outcomes. Using the limited available data in this area, we found that the magnitude of improvements in education outcomes associated with increased public education spending has been relatively small. On average, a 10 percent increase in spending has led to a 0.8 percent improvement in education outcomes. Our findings suggest that the effectiveness of public education spending depends in part on a country’s initial levels of efficiency and spending, and that countries starting from a low-spending and low-efficiency base have been able to improve outcomes significantly. This implies that efforts by these countries to catch up to other countries in terms of spending have resulted also in their catching up in terms of outcomes. In fact, our findings conclude that countries with fiscal flexibility, low spending, and poor outcomes are the best candidates for increasing education funding, especially if they start from a very inefficient point. We also used our findings on the speed and effectiveness of resource mobilization efforts to benchmark the space that countries have to mobilize further public resources for education as well as the improvements in outcomes they can expect from these spending increases. In particular, we used a combination of current public education spending, the growth and level of the government budget as a percentage of GDP, and the priority that high-spending countries give to education spending in their budgets to suggest a range of realistic pathways for expanding the education budget in any given country. We also showed, by using elasticities, that it is possible to benchmark a feasible pathway for outcomes as a result of different levels of spending increases. These pathways could prove useful for policy making in at least two ways. First, they help to answer the two key questions any finance minister would ask: (i) Do I have the space to grow spending on education? and (ii) What will I get out of this spending? By referring to these benchmarks, decision makers can get basic information about what other countries with conditions similar to their own country have experienced in terms of spending and outcomes, and what they can expect in their own country. Second, our findings on the role that efficiency plays in the elasticity of outcomes can help countries better understand whether and what types of education and financial management reforms they should undertake to accompany an education funding increase. Specifically, we find that countries already at a high level of efficiency must make further, significant improvements in efficiency if funding increases are to generate any noticeable improvements in outcomes. Our general results and the analysis we performed on specific countries have shown that the benchmarking approach can be effective at the country level. However, existing data limitations currently make it difficult for this approach to be used in many countries. There is an urgent need for better quality and coverage of data on education spending and outcomes at the country level. Initiatives in the health sector show that it is possible to collect better data on both public and private spending in all countries. Drawing on the lessons learned from initiatives of this kind, countries should intensify their efforts to gather the data that policy makers need to make better-informed spending decisions and to more effectively monitor learning outcomes. 43 References Altinok, Nadir, Noam Angrist, and Harry Anthony Patrinos. 2018. Global Data Set on Education Quality (1965–2015). Washington, DC: The World Bank. Commins, S. 2017. Fragility, Conflict, and Violence. World Development Report 2018 background paper. Washington, DC: World Bank. Education Commission. 2016. The Learning Generation: Investing in Education for a Changing World. New York: International Commission on Financing Global Education Opportunity. Filmer, Deon, Halsey Rogers, Noam Angrist, and Shwetlena Sabarwal. 2018. Learning-Adjusted Years of Schooling (LAYS): Defining a New Macro Measure of Education. Washington, DC: The World Bank. Gallet, Craig A, and Hristos Doucouliagos. 2017. "The Impact of Healthcare Spending on Health Outcomes: A Meta-regression Analysis." Social Science & Medicine 179:9-17. Grigoli, Francesco. 2014. A Hybrid Approach to Estimating the Efficiency of Public Spending on Education in Emerging and Developing Economies. Washington, DC: International Monetary Fund. Jayasuriya, Ruwan, and Quentin Wodon. 2003. Explaining Country Efficiency in Improving Health and Education Indicators: The Role of Urbanization. World Development Report 2004 Background Paper. Washington, DC. Kim, Gwang-Jo. 2002. Education Policies and Reform in South Korea. Secondary Education in Africa: Strategies for Renewal. Washington, DC: World Bank. Kraay, Aart C. 2018. Methodology for a World Bank Human Capital Index. Washington, DC: World Bank. Patrinos, Harry Anthony, and Noam Angrist. 2018. Global Dataset on Education Quality: A Review and Update (2000–2017). Washington, DC: World Bank. Ravallion, Martin. 2005. "On Measuring Aggregate “Social Efficiency”." Economic Development and Cultural Change 53 (2):273-292. Rolleston, C. 2009. Financing Primary Education for All: Trends Post-Dakar and the Importance of Growth. Background paper for the 2010 UNESCO EFA Global Monitoring Report. Tandon, A. 2017. "Exploring "Fiscal Space" in Social Sector Budgeting." Presentation at Harvard Ministerial Leadership Forum. UIS (UNESCO Institute for Statistics). 2016. Who Pays for What in Education? The Real Costs Revealed Through National Education Accounts. Montreal: UNESCO Institute of Statistics. UNESCO. 2015. Pricing the Right to Education: the Cost of Reaching New Targets by 2030. Paris: UNESCO. 44 UNESCO (United Nations Educational, Scientific, and Cultural Organization),. 2011. EFA Global Monitoring Report. The Hidden Crisis: Armed Conflict and Education. Paris: UNESCO. Van Der Burg, L, and S Whitley. 2016. Unexpected Allies: Fossil Fuel Subsidy Reform and Education Finance. London: Overseas Development Institute. Wagstaff, Adam, and Liang Choon Wang. 2011. "A Hybrid Approach to Efficiency Measurement with Empirical Illustrations from Education and Health." 45 Annex Tables Table A1: Descriptive Statistics of Main Government Spending Variables Ed. GDP, Ed. Spend/ Ed. Spend/ Ed. Spend/ Total Spend, Ed. Ed. PPP Student, Student, Student, Govt. Year (2015 Spend, Spend, (2011 Primary Secondary Tertiary Spend, % $PPP, % GDP % Govt. $PPP, (2015 $PPP) (2015 $PPP) (2017 $PPP) GDP millions) millions) 1999 Mean 20,256 4.6 15.0 3,030 4,360 8,342 325,702 25.4 n 125 135 118 57 59 52 183 96 2000 Mean 20,231 4.4 14.7 2,778 4,265 8,060 334,180 25.0 n 126 135 128 65 64 60 187 95 2001 Mean 22,300 4.7 15.3 3,104 4,725 7,937 340,328 24.9 n 118 127 117 65 62 61 188 103 2002 Mean 21,724 4.6 15.1 3,400 4,688 6,967 347,799 25.0 n 127 135 128 73 71 66 189 109 2003 Mean 23,958 4.7 15.4 3,604 5,223 7,873 360,914 24.7 n 118 125 117 67 67 59 189 116 2004 Mean 24,492 4.4 14.7 3,432 4,778 6,583 380,360 24.2 n 122 132 126 74 76 75 189 120 2005 Mean 28,474 4.5 15.0 3,471 5,082 7,332 398,307 24.6 n 114 123 116 77 73 70 189 121 2006 Mean 29,757 4.5 15.0 3,852 5,225 7,371 419,545 24.6 Nn 113 123 116 67 70 71 189 124 2007 Mean 29,484 4.5 15.0 3,941 5,315 7,504 439,852 24.3 n 111 121 114 74 76 72 190 124 2008 Mean 30,172 4.5 15.0 3,882 4,999 6,897 452,529 25.5 n 127 134 128 90 86 82 190 132 2009 Mean 33,545 4.8 14.5 3,788 4,924 6,879 450,860 26.9 n 122 131 127 92 85 93 190 135 2010 Mean 29,508 4.6 14.4 3,889 4,845 6,724 474,367 26.3 n 133 140 136 99 94 100 190 142 2011 Mean 31,561 4.4 14.3 3,855 4,800 6,153 481,309 26.2 n 130 135 131 106 102 100 195 142 2012 Mean 33,146 4.5 14.6 3,521 4,713 6,025 510,314 26.3 n 127 130 128 96 96 103 190 139 2013 Mean 33,080 4.6 14.4 3,865 4,675 6,557 527,705 26.6 n 131 136 132 101 100 101 190 131 2014 Mean 32,838 4.6 14.4 4,119 5,131 7,524 549,152 27.0 n 126 131 129 93 95 92 189 128 2015 Mean 37,221 4.7 14.6 4,495 5,069 7,237 568,123 27.3 n 119 125 121 83 89 86 188 119 Note: Ed. = Education; PPP = Purchasing Power Parity; GDP = Gross Domestic Product; Govt. = Government; Mean = Average Value; n = Number of Observations. Source: UIS, IMF, and WDI online databases. 46 Table A2: Patterns and Trends in Government Spending as a Share of GDP, 1998–2017 1998–2001 2002–05 2006–09 2010–13 2014–17 Income Groups LIC 13.3 13.3 15.7 18.2 19.5 LMIC 19.0 19.4 20.9 21.7 23.9 UMIC 22.1 23.2 25.6 25.6 26.3 HIC 31.0 30.9 30.9 31.8 31.7 Regions AFR 19.0 19.1 18.6 19.4 21.9 ECA 31.2 31.5 33.4 33.8 33.2 LCR 19.6 21.0 21.5 22.8 23.8 SAR 17.3 16.4 20.7 20.9 21.1 EAP 19.6 17.8 22.3 23.6 26.0 MNA 30.6 30.0 28.1 27.0 26.9 Source: World Bank calculations based on UIS and IMF data. Note: Income groups are defined by countries’ income group classifications as of 2017. LIC = low- income country, LMIC = lower-middle-income country, UMIC = upper-middle-income country, and HIC = high-income country. AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, SAR = South Asia, EAP = East Asia and the Pacific, and MNA = Middle East and North Africa. Table A3: Public Education Spending Per Student (Constant 2015 PPP $), 1998–01 to 2014–17 1998–2001 2002–05 2010–13 2014–17 Pri. Sec. Tert. Pri. Sec. Tert. Pri. Sec. Tert. Pri. Sec. Tert. Income Groups LIC 136 418 6,611 187 391 3,874 162 307 3,105 207 339 2,010 LMIC 522 978 4,783 570 985 4,199 808 1,087 2,390 913 1,414 2,573 UMIC 1,232 1,876 4,837 1,668 1,899 4,810 1,973 2,540 3,786 2,510 3,183 4,060 HIC 5,319 7,160 11,887 6,231 8,237 10,711 8,401 9,452 12,155 8,205 9,134 13,217 Regions AFR 441 1,121 7,691 563 1,225 6,940 489 724 3,658 664 1,135 3,747 ECA 5,795 7,600 9,329 6,629 8,123 8,902 7,039 7,982 8,643 7,127 7,257 9,085 LCR 1,390 1,537 4,105 1,302 1,447 2,765 1,858 2,122 3,926 2,527 2,749 3,772 SAR - - - 599 1,364 1,229 608 600 2,515 884 898 2,323 EAP 2,193 2,400 7,025 2,510 2,741 6,233 4,028 5,800 8,482 3,990 5,104 10,381 MNA 3,901 4,211 13,525 3,833 4,340 5,756 5,102 6,066 7,109 6,143 7,244 8,955 Source: World Bank calculations based on UIS and IMF data. Note: Income groups are defined by countries’ income group classifications as of 2017. LIC = low-income country, LMIC = lower-middle-income country, UMIC = upper-middle-income country, and HIC = high-income country. AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, SAR = South Asia, EAP = East Asia and the Pacific, and MNA = Middle East and North Africa. Table A4: Public Education Spending Per Child (Constant 2015 PPP $), 1998–01 to 2014–17 1998–2001 2002–05 2010–13 2014–17 Pri. Sec. Tert. Pri. Sec. Tert. Pri. Sec. Tert. Pri. Sec. Tert. Income Groups LIC 109 74 96 148 85 125 153 120 145 188 140 165 LMIC 611 440 483 628 635 635 793 770 786 894 1,118 908 UMIC 1,226 1,312 1,308 1,656 1,442 1,424 1,971 2,323 2,246 2,488 3,021 2,582 HIC 5,276 6,821 5,954 6,198 8,037 7,103 8,305 9,202 8,851 8,130 8,992 8,887 Regions AFR 413 557 528 514 701 669 472 461 466 647 849 811 47 ECA 5,682 7,421 6,676 6,461 7,976 7,107 6,984 7,646 7,841 7,079 7,206 7,469 LCR 1,390 1,202 1,251 1,295 1,241 1,230 1,825 1,912 1,796 2,460 2,550 2,549 SAR - - - 589 864 858 604 408 409 874 682 683 EAP 2,225 2,071 2,285 2,353 1,988 1,796 3,787 5,493 3,939 3,942 4,891 5,280 MNA 3,741 3,278 3,010 3,790 3,879 3,965 5,015 5,556 6,144 6,111 6,713 7,334 Source: World Bank calculations based on UIS and IMF data. Note: Income groups are defined by countries’ income group classifications as of 2017. LIC = low-income country, LMIC = lower-middle-income country, UMIC = upper-middle-income country, and HIC = high-income country. AFR = Africa, ECA = Europe and Central Asia, LCR = Latin America and Caribbean, SAR = South Asia, EAP = East Asia and the Pacific, and MNA = Middle East and North Africa. Table A5: Five-Year Episodes of Public Education Spending Growth by Income Group, 1999–2017 Total Decomposition into sources of five-year growth: five-year Education spending Aggregate prioritization growth public spending (govt. ed. exp. Economic (govt. exp. as as % of govt. growth % of GDP) exp.) Low income 39 27 5 6 Lowest growth quartile -4 25 -10 -19 2nd quartile 20 21 5 -7 3rd quartile 46 29 11 7 Highest growth quartile 92 34 15 44 Lower middle income 24 20 6 -2 Lowest growth quartile -6 12 -2 -16 2nd quartile 15 20 0 -6 3rd quartile 31 23 4 4 Highest growth quartile 57 26 23 8 Upper middle income 18 17 0 1 Lowest growth quartile -15 12 -12 -14 2nd quartile 14 15 1 -2 3rd quartile 25 19 2 4 Highest growth quartile 48 20 10 17 High income 13 11 2 0 Lowest growth quartile -7 6 -5 -8 2nd quartile 7 8 1 -2 3rd quartile 15 9 4 2 Highest growth quartile 36 21 7 7 All low- and middle-income countries 26 21 4 1 Lowest growth quartile -8 15 -8 -16 2nd quartile 16 19 2 -5 3rd quartile 33 23 5 4 Highest growth quartile 63 26 17 20 Source: World Bank calculations based on UIS and IMF data. 48 Table A6: Initial Country Conditions by Quartile of Five-Year Public Education Spending Growth Total Initial conditions (values at start five-year of five-year period) spending Education growth Aggregate prioritization public spending (govt. ed. exp. (govt. exp. as as % of govt. % of GDP) exp.) Low income 39 18 16 Lowest growth quartile -4 20 18 2nd quartile 20 16 17 3rd quartile 46 17 15 Highest growth quartile 92 20 13 Lower middle income 24 24 16 Lowest growth quartile -6 27 16 2nd quartile 15 23 17 3rd quartile 31 24 16 Highest growth quartile 57 22 14 Upper middle income 18 30 13 Lowest growth quartile -15 36 14 2nd quartile 14 29 13 3rd quartile 25 29 13 Highest growth quartile 48 27 13 High income 13 33 13 Lowest growth quartile -7 36 13 2nd quartile 7 35 13 3rd quartile 15 32 13 Highest growth quartile 36 30 13 All low- and middle-income countries 26 25 15 Lowest growth quartile -8 28 16 2nd quartile 16 23 16 3rd quartile 33 24 15 Highest growth quartile 63 23 13 Source: World Bank calculations based on UIS and IMF data. Table A7: Public Education Spending Growth of Median Countries Real Public education Share of education in Share of total education spending as a % of total government government budget spending GDP budget as % of GDP growth (%) 50th 50th 50th 50th 50th 50th 50th percentile percentile percentile percentile percentile percentile percentile of five- of five- of five- of five- year year year year percent percentage percentage percentage growth point point point growth growth growth Low- and middle- 22.2 4.3 0.1 14.6 0.04 22.3 0.9 income total UMICS 19.8 4.5 0.1 12.8 0.3 27.8 0.4 LMICs 22.1 4.5 0.1 14.9 -0.3 23.0 1.2 LICs 29.7 3.7 0.2 16.2 0.1 15.6 0.8 Source: World Bank calculations based on UIS and IMF data. 49 Note: The median is defined separately for each indicator. The averages for the share of education and the total government budget are based on the country period averages for 2010–13 and 2014–17. Table A8: Rank Correlation Coefficients between Efficiency Measures OLS DEA SFA W&W 1. Outcome: expected years of schooling Spending: public education spending per child OLS 1 DEA 0.781*** 1 SFA 0.815*** 0.916*** 1 W&W 0.773*** 0.965*** 0.890*** 1 2. Outcome: harmonized test scores Spending: public education spending per student OLS 1 DEA 0.774*** 1 SFA 0.853*** 0.919*** 1 W&W 0.742*** 0.915*** 0.825*** 1 3. Outcome: learning-adjusted years of schooling Spending: public education spending per student OLS 1 DEA 0.744*** 1 SFA 0.817*** 0.917*** 1 W&W 0.769*** 0.924*** 0.842*** 1 Note: OLS = Ordinary Least Squares; DEA = Data Envelopment Analysis; SFA = Stochastic Frontier Analysis; W&W = Wagstaff and Wang (2011). * p<0.05, ** p<0.01, ***, p<0.001. Table A9: Correlation Coefficients between SFA Efficiency Measure and Proxies of Education Spending Efficiency Efficiency measure based on: Expected years Learning-adjusted Learning per of schooling per years of schooling per student student student Student-teacher ratio (primary) 0.646*** 0.114 0.515*** Student-teacher ratio (secondary) 0.492*** 0.263*** 0.414*** Population density -0.020 -0.128* -0.108 Notes: A positive correlation coefficient shows that efficiency increases as the values of the proxy variable rise. SFA = Stochastic Frontier Analysis. * p<0.05, ** p<0.01, *** p<0.001 50