Public Spending and Education Attainment in the Middle East and North Africa

1 Introduction

It is generally acknowledged that education attainment has advanced significantly in the countries of the Middle East and North Africa region (hereafter referred to as MENA) over the years (Rugh 2002; Iqbal 2006). This progress is documented in the first two columns of Table 1 which shows how the stock of education, as measured by the average years of schooling completed per person over the age of 15, has evolved for fifteen MENA countries over the period 1970–2010. ^[1] Each and every country in the table shows a significant increase in education stock. One could show similar progress as well using other measures of education attainment such as improvements in primary and secondary school enrolment rates.

What is less appreciated is that MENA has achieved the best performance in education attainment in the world over the last four decades. A hint of this is provided in the fifth column of Table 1 which shows that all but two MENA countries (Qatar and Kuwait) feature in the top half of the global league when ranked by percentage change in education stock between 1970 and 2010.

A clearer sense of the relative performance of MENA can be obtained from looking at trends in regionally aggregated data. Table 2 provides population weighted averages for education stock for each of the developing country regions (see Annex 1 for list of countries included in each region) and the high-income OECD group of countries for the period 1970–2010. The table shows that the greatest percentage increase in education stock (239%) during the past 40 years occurred in MENA.

Figure 1 graphs regional data for education stocks at 5-year intervals between 1970 and 2010. It shows that the MENA region was clustered at the low-stock end with South Asia and Sub-Saharan Africa in 1970 but diverged from this cluster over the next 40 years. By 2010, it had converged with the East Asia region and narrowed the gap with the Latin America and Caribbean region. In doing so, it experienced the fastest rate of stock growth among all regions over the past four decades. ^[2]

Figure 1:

Evolution of education stocks across regions, 1970–2010.

Our purpose in this paper is to determine what has accounted for the better comparative performance of MENA. We first check what the literature has to say about the determinants of educational attainment in Section 2. Then we develop an econometric model in Section 3 to investigate this question more rigorously. In Section 4, we report empirical results and a few robustness checks. These show that public spending on education was the policy lever with the strongest effect on education outcomes. In Section 5, we provide some additional observations on the comparative efficiency of public spending in MENA and on its link with education quality.

2 Determinants of Education Attainment: A Brief Literature Survey

An overview of the literature shows that public spending on education (on schools, teachers, administrators, and ancillary facilities and equipment) is considered the key determinant of education attainment in developing countries. In addition, a variety of socioeconomic variables are typically included as independent determinants. Some of these are thought to have a bearing on the effectiveness with which public spending is transformed into successful education performance. Others are thought to have a direct influence on education attainment. Below, we discuss public spending and the following additional variables that often feature in the literature: income; child health; child employment; and urbanization.

3 Empirical Model and Data

The empirical model that is typically found in the literature is of the following general form:

where the subscript i denotes country, E refers to the measure of education adopted as the dependent variable, PBSE refers to public spending on education, and Z refers to a vector of other socioeconomic control variables.

An empirical model to explain changes in education attainment over time may be written in the following general form:

where the left-hand side refers to the change in E over n year interval from year t to year t+n and, on the right-hand side, Et refers to the initial value of E at time t; and the subscript avg refers to the average value of the independent variables over the n year interval.

This empirical model is similar to that found in the available literature in that it considers education attainment as being determined primarily by public expenditures on schools, teachers, and related inputs as well as on factors that have a bearing on the effectiveness of such spending and on the opportunity costs of education. Where it differs is in (a) the definition of the dependent variable as the change in education stocks over time rather than the level at a point in time; (b) the use of the initial stock of education as an independent variable; and (c) the use of average values over 20- to 40-year intervals for most independent variables rather than contemporary values. These three aspects are discussed further below.

4 Regression Results and Robustness Checks

The empirical model specified in eq. [3] is estimated using ordinary least squares (OLS) over a cross-sectional sample of between 72 and 76 developing countries. The regression results are reported in Table 4 in five columns. The first column contains our two main variables of interest, the initial stock of education in 1970, and the average level of public spending on education. The next four columns add one further independent variable at a time to capture the effects of different socioeconomic factors highlighted in the literature.

The results are very clear. All the action is in the first column. The initial stock and public spending explain 88% of the variation in the change of education stocks across developing countries over the past 40 years. Including other variables in subsequent columns adds only 1% in explanatory value.

The initial level of education in 1970 has by far the strongest effect on the subsequent attainment of more education over the next four decades. The standardized coefficient for this variable is consistent in sign, size, and value across all five columns. The coefficient value of –0.73 implies that countries with an education stock that was 1 standard deviation less than the sample average in 1970 increased their stocks by 73% more than the average by 2010. The negative sign on the initial level of education shows that convergence or catch-up has occurred: countries with low initial education stocks have increased their stocks faster than those with higher initial stocks. Thus convergence not only plays a role in income growth (as shown in Barro (1991) and subsequent studies) but also in the growth of education stocks.

The results also show that public spending on education is a significant and positive determinant of education attainment. The higher the public spending on education per capita, the higher the change in education attainment. This is consistent with results reported for public spending on education by Gupta et al. (2002) and Baldacci et al. (2008) among others.

Caloric intake (which we take to reflect child health status) also matters: the worse the undernourishment of the population, the smaller the improvement in education attainment. Similarly significant results have been reported using alternative measures of child health such as the child mortality rate (see Gupta et al. 2002; Baldacci et al. 2008).

The income growth variable is not significant in our regressions. While we cannot compare this result directly with other studies in the literature since they use an income-level variable rather than the growth variable, even the level variable produces mixed results. For example, Gupta et al. (2002) report a significant positive association between gross secondary enrollment rates and per capita income but not for gross primary and secondary rates taken together. Baldacci et al. (2008) find current per capita income to be a significant and positive determinant of primary and secondary gross enrollment rates in some but not in all of the models they run.

With regard to urbanization, Gupta et al. (2002) report a significant result for this variable but Baldacci et al. (2008) and Rajkumar and Swaroop (2008) do not. Finally, the opportunity cost of schooling as measured by the child employment rate is not a significant variable in our regression, in contrast to the results reported by Goldin and Katz (1997) for the case of the United States.

We subjected our results to four robustness checks: using alternate proxies for the public spending and income variables; adding other independent variables; adding high-income countries to the sample; and using a two-stage least squares (TSLS) estimator. Our results are robust to each of these checks in the sense that the signs and significance of initial stock and public spending do not change. Table 5 shows the results for the first three robustness checks and Table 6 reports the results for the TSLS run.

The first two columns of Table 6 shows the results of substituting education spending as a share of GDP for education spending per capita and income per capita in place of income growth. Most of the studies cited earlier use these proxies for public spending and income. But the results do not change. Public spending on education is still significant when entered as a ratio to GDP and income is still insignificant when entered as the average level rather than the average growth over time. ^[6]

The third and fourth columns show the results of adding two new independent variables, control of corruption and ethnic tensions. Such variables have been used in some earlier studies (see, for example, Gupta et al. 2002; Rajkumar and Swaroop 2008) to investigate the effects of institutional quality and political context on human capital attainment. In keeping with earlier studies, we have interacted corruption with public spending. These variables are not significant in our regressions and do not change the significance of initial stock or public spending, our two main variables of interest.

The fifth column shows the results of running our basic model over a larger sample of countries, including high-income countries. Once again, no notable change occurs in the results.

Table 6 shows the results of running the model using a TSLS procedure to account for omitted variable bias. We use the initial value (in 1970) of public spending on education as an instrument for its average value over the period 1970–2010. Here, the idea is that the initial value of public spending on education influences its average value but is not directly correlated with education attainment growth. We use the Hansen J-test of overidentifying restrictions to test the validity of this excluded instrument. We present the OLS estimates because the Hausman specification test suggests that there is no difference between the coefficients of the OLS and TSLS estimators. In such a case, the OLS estimator is more efficient than the TSLS estimator given that the latter uses less information than the former. Once again, initial education stock, public spending, and undernourishment remain significant and with the right signs. All other independent variables remain statistically insignificant.

5 Education Spending and Performance: Some Additional Observations

The regression results suggest that we should look for differences among regions in initial stocks of education and rates of public spending on education in order to understand why some regions have done better than others in education performance since these are the two most important variables in terms of economic and statistical significance. We could add undernourishment to this list since it is also consistently significant in our results though of much less explanatory importance.

Table 7 compares regions in terms of these three independent variables. This shows that MENA had the second lowest value for initial education, the highest value for public spending on education and the best (lowest) value for undernourishment. Thus, it is clear that MENA’s comparative edge comes from its low initial education stock and its high rate of public spending on education. The only pairwise comparison where initial education stock does not give MENA an edge is for the case of Sub-Saharan Africa where the latter had a lower stock (of 1.9 years) to begin with. In this case, however, the huge spending gap between MENA and Sub-Saharan Africa ($252 per capita per year versus $37) accounts for the performance edge. The same applies to a pairwise comparison with South Asia: since the two regions started off with approximately the same stocks in 1970, it was the far higher rate of public spending in MENA that accounted for the difference in the change in education stocks over time. MENA’s low rate of undernourishment adds to its edge in all pairwise comparisons but only slightly since this variable has much lower explanatory value and economic significance.

Between initial education stock and public spending, only the latter is of policy relevance. So it is worth investigating other aspects of MENA’s public spending as well, such as its efficiency. We have data on public spending for the period 1980–2010. From this, we can calculate the amount of education achieved per dollar spent as well as the amount of spending needed to obtain an additional year of education stock. These two calculations are shown for all developing country regions in Table 8.

The first column of Table 8 shows that MENA achieved 1.54 years of additional education stock per $100 in public spending per capita per year. This is somewhat better than that achieved by Latin America and Europe/Central Asia but much worse than that achieved by South Asia and Sub-Saharan Africa. A similar ranking may be seen when considering the amount of per capita education spending that was needed to add 1 year to the cumulative education stock. MENA spent close to $65 per capita per year (in 2005 purchasing power parity (PPP) terms) to achieve each additional year of schooling, faring somewhat better in this regard than Latin America and Europe/Central Asia but much worse than South Asia, Sub-Saharan Africa, and East Asia.

The seemingly super-efficient numbers for South Asia need some comment. South Asia spent only $2.89 to obtain an additional year of education stock. This reflects the much lower education input costs (for example, on teacher salaries and school buildings) of the much poorer countries that comprise that region. It should also be noted that, despite its relative efficiency, South Asia did not increase its education stock as much as MENA because it spent far less ($8.4 per capita compared to $252.3 as shown in Table 7).

A second aspect of the link between education spending and performance is that of quality. ^[7] It is generally acknowledged that the quality of education in MENA leaves a lot to be desired. The limited, though growing, body of comparative data on education quality shows most MENA countries clustered in the bottom half of the global distribution. For example, 13 MENA countries participated in one or more segments of the 2007 and 2011 rounds of TIMSS (Trends in International Mathematics and Science Study) but none scored above the global average. While such a result may be expected of less developed countries with few resources to spend on education, it is surprising for MENA which is a middle-income region and has been spending considerable resources on education.

Figure 2 shows a scatter plot of country-level TIMSS scores (for mathematics at the eighth-grade level in 2011) and public spending on education (averaged over 1980–2010). MENA countries are labeled and all but Iran can all be seen to fall below the bivariate regression line. Similar results are obtained for other measures of education quality, such as the TIMSS score for science. This suggests that MENA countries have underperformed in terms of education quality in relation to their levels of public spending.

Figure 2:

Public spending on education and TIMSS score for Mathematics.

Recent research also suggests that student characteristics (such as parent’s education and exposure to books at home) tend to be more important than school characteristics (such as teacher qualifications and class sizes) in explaining test scores in MENA countries (see Badr et al. 2012). All this leads to the conclusion that while the challenge of hiring teachers, building schools and getting children into classrooms has been successfully met in MENA that of converting all this schooling into productive learning remains to be met in the future.

Annex 1: Regional country composition

Annex 2: Variable definitions and sources