A Simple Model of Public Education Spending

This section outlines a simple formal model of how democracy and globalization affect total education spending and its composition (the Appendix provides a more detailed derivation). The model demonstrates two key components of education spending. First, although public education provision necessitates fiscal redistribution, this redistribution can be either progressive or regressive depending on the ability to differentially target education spending to particular groups. Second, education spending is inseparable from the labor market. The returns to education, and thus the political incentives to provide it publicly, are determined by supply and demand in the market for educated labor. These market forces are, in turn, dependent on a state's integration into the global economy. Building on these two components, I model political preferences over education under different political regimes and varying labor market conditions.

I consider a country with three groups of individuals: a high-income group, a middle-income group, and a poor group. Individuals earn income from two sources. First, they receive group wealth q_i, where i ∈ {H,M,P} and q_H > q_M > q_P. This income is not derived from the labor market: it can be considered as unequally distributed wealth. Individuals also receive income w_j from their factor endowments of either skilled or unskilled labor.⁴

The returns to skilled and unskilled labor, w_s and w_u, are defined by their relative productivity and their relative abundance. Productivity is determined by the technology available to states with skilled productivity denoted σ_s and unskilled productivity σ_u. The relative abundance of skilled labor is determined by the proportion of the population that is skilled: S ∈ [0,1]. Education spending increases S, which in turn reduces skilled wages by the skill scarcity factor b. This relative abundance of skilled labor implies relative scarcity of unskilled labor, increasing unskilled wages by a factor a.

When education spending is increased, the provision of education expands from the richest to the poorest person. This assumption mirrors the historical expansion of education in both the developing and developed world.⁵

I define an individual's inverse skill index as s_i ∈ [0,1], which is inversely related to the wealth parameter q_i. Individuals with s_i below the level of education provision S receive education; those with higher levels of s_i do not. Thus, as S increases, the number of individuals earning skilled wages also increases—from the richest to the poorest member of society. These assumptions produce the following set of equations for income y_i.

The income equations show how the relative scarcity effects, a and b, and the productivity parameters, σ_s and σ_u, affect the preferences of individuals over education spending. For those who already possess education, any expansion reduces their wages by a factor b. Those who receive education for the first time will receive a jump in their income through the skill premium σ_s − σ_u. Finally, if a is greater than 0, the unskilled individuals benefit indirectly through skill provision as unskilled labor becomes scarcer.

I assume there are two generations where parents, who live in period 0, decide how much to tax themselves to provide public education for their children, who live in period 1.⁶

Parents value their children's welfare altruistically. I assume education is provided with constant marginal cost c. Thus the total cost of education will simply be cS₁. A flat tax on income t is used to fund education spending. The population of the country is normalized to one, thus total tax take is t multiplied by average parental income y₀. Education provision is limited to the public sector so even the rich must pay some positive level of taxation if they wish to see their children educated.⁷ This leads to the budget constraint:

Finally, individual utility is comprised of round 0 income net of taxes and round 1 income (discounted by δ), which is determined by whether children receive skilled or unskilled wages w_j1(S₁) and by positive externalities produced by education g(S₁).⁸

I begin analyzing the effect of democratization on education spending by examining the distribution of education under autocracy, where the high-income group receives education but the middle-income and poor group remain uneducated. Following democratization, the locus of political power switches from the elite to the democratic median voter, who is a member of the middle-income group. Will this voter choose to expand education spending and if so, how far will they do so? To answer these questions, I examine how increasing education to incorporate the middle class affects each group.

All groups suffer a strictly negative impact from the tax cost of providing education, albeit weighted by relative income.⁹

All groups also receive positive externalities. However, the impact on round 1 (or children's) wages varies across groups. The high-income group not only pays the highest absolute amount in taxes but is also negatively impacted, via the skill scarcity parameter b, by the increased relative abundance of educated people.¹⁰ Thus, if autocracies represent the preferences of the high-income group they will choose not to expand education spending. The middle-income group, conversely, is the chief beneficiary of increased education provision, because their income jumps up by the skill premium σ_s − σ_u. However, they may wish to limit further expansion, since, as with the elite, they are also harmed by expansion to the poor.¹¹ Finally, the poor will benefit from education provision, even though they do not receive it, if the unskilled scarcity parameter a or externalities are large enough. Thus both the middle-income group and the poor prefer increased education spending. Assuming that democratization enfranchises at least the middle class, I obtain proposition (1a):

Proposition 1a: The expansion of democracy increases public education spending.

The model above precluded citizens from purchasing private education, although private education is a sizeable sector in many states. However, adapting the model to include private education does not substantially alter any of its propositions. The rich strictly prefer private to public education, because they can access education without paying for others to receive it. The masses, conversely, prefer public education, because at least some proportion of this education is subsidized for any individual with income below the mean. In the presence of a sizeable public education sector, private education is even less attractive for most citizens since they would have to pay for education twice—once privately and once through the tax system. Consequently, democratization will lead to a shift away from private to public education, as noted in proposition (1b):

Proposition 1b: The expansion of democracy reduces private education spending.

I now turn to examine the effects of globalization on education spending, starting with its supply-side effects on the labor market. The “law of one price,” as developed by Samuelson, presumes that as countries become fully integrated with the global economy, the returns to factors become increasingly determined by global rather than domestic supply.¹²

While this theorem requires an array of stringent assumptions that are poorly met by the realities of modern trade—prices and wages clearly do not fully converge—the more general implication of the theory, that trade reduces the response of factor returns to purely domestic market forces, is intuitive.¹³ The ability of states to specialize in the global economy in itself suggests that states can sustain production mixes internationally that would be impossible under autarky. Consequently, under pure autarky, an increase in the supply of education will be met either by a reduced skill premium, educated unemployment, or both. As states move from autarky to openness, the responsiveness of relative wages to changes in the supply of education will be reduced.

Applying this logic to the model, this implies that the parameters a and b will be smaller in open states. Reexamining equations (4) through (6) shows that integrating into the global economy increases the attractiveness of education provision for both the elite and middle class. Since the expansion of education in an open economy has a smaller impact on skilled wages than in an autarky, the high-income group is less harmed by education spending and may favor increased spending if externalities outweigh tax costs. For the middle-income group there is a similar, albeit smaller positive impact. However, for the poor, globalization actually weakens the positive effects of education expansion since they benefit less from the relative scarcity of unskilled labor. These results imply two propositions. First, if, as in most countries, either the high- or middle-income group control government, education spending should expand when states globalize. Second, since the positive effect of globalization on preferences over education is greatest for the high-income group and smallest for the poor, I expect the positive impact of globalization on education to be stronger in autocracies than democracies.

Proposition 2a: Increased integration with the global economy will lead to increased education spending in both democracies and autocracies.

Proposition 2b: Increased integration with the global economy has a larger positive supply-side effect on education spending in autocracies than in democracies.

Globalization also affects labor market demand. Acemoglu argues that contemporary globalization produces skill-biased technology transfer from developed to developing states.¹⁴

While originally extractive in nature and hence complementary with natural resources and unskilled labor, technology transfer has tended over the past half-century toward skill-biased technologies, such as computing, which require an educated workforce to operate.¹⁵ These technologies may be introduced by entering multinational firms following trade liberalization. However, once exposed to the international market domestic firms will also require a skilled workforce to implement this technology or they risk being driven out of business. Hence, globalization raises the demand for education, by raising the productivity of educated workers.

Turning to the model, skilled wages are now w_s = γ(σ − σ_s) + σ_s − bS₁, where σ is the level of technological productivity in the most developed state, and γ ∈ [0, 1] is the degree to which globalization leads a country to close the gap between domestic productivity and that of the most developed state. Where there is a large potential jump in the skill premium from technology transfer, or where technology transfer γ and skill provision S are complementary, globalization will increase overall education spending. Furthermore, the effects of increasing technology transfer γ on education spending will be stronger for states that are further away from the level of productivity in the most advanced state σ. Thus, developing states will see a greater demand-side impact on education spending than developed states. These results provide proposition (2c):

Proposition 2c: Increased integration with the global economy has a greater demand-side effect on education spending in developing states than in developed states.

Greater openness is not predicted to have a significant impact on private spending. On the demand side, the effect of openness on private spending might be positive, if technology transfer increases private incentives to invest in education that are not met by public provision. On the supply side, however, the predicted increase in public education spending is likely to substitute for private education, thereby reducing private spending. These two countervailing effects provide proposition (2d):

Proposition 2d: Increased integration with the global economy will have a negligible effect on private education spending.

I conclude the analysis by examining the impact of “targeted spending,” where increased education spending is spent on greater “depth” (more per student funding) rather than greater “breadth” (increasing student enrollment). Since the baseline model assumed that education enrollment moves from rich to poor, increased “depth” implies targeting resources toward the well-off. This kind of regressive targeted spending captures the funding of tertiary education, which is de facto limited to upper-income individuals in most states. Where the elite group controls political decision making, its members will choose to increase the depth of education spending rather than its breadth, since this focuses resources on their own children. The key implication is that if education can be targeted to groups with higher levels of income, as with university education as opposed to primary education, its provision will be fiscally regressive and thus favored in more autocratic regimes. This produces proposition (3a):

Proposition 3a: Autocracies will favor a higher tertiary to primary education spending ratio than democracies.

Finally, I derive a proposition from the Stolper-Samuelson theorem about the impact of globalization on the composition of education spending in developed versus developing countries.¹⁶

Developing states have an abundance of labor with fairly low education and a scarcity of university-educated individuals. Hence, when the economy opens, production will shift to sectors with high demand for workers with primary education not university education. Conversely, developed states have a relative abundance of university-trained workers. Thus, global integration will lead to a shift toward high-skill production demanding college education. Provided government spending tracks changes in demand, this produces proposition (3b):

Proposition 3b: Following globalization, the tertiary to primary education spending ratio should decrease in developing states and increase in developed states.

Empirical Analysis

This section tests the several propositions derived from the formal model on a data set of 113 countries between 1960 and 2000. I begin by considering the effect of regime type and openness on aggregate education spending before examining how these effects play out in different subsamples (autocracies versus democracies; developed versus developing states). I finish the empirical analysis by examining the impact of regime type and openness on the composition of education spending. Before turning to the statistical analysis, I briefly discuss the variables I use. The formal model of education spending focuses on tax revenues spent on moving individuals from an unskilled group to a skilled group. Since this is a simplification of the process of education provision, it is impossible to find a precise analog. However, in terms of aggregate national spending, the most tractable measure is public expenditure on education as a percentage of GDP. This measure comes from the World Development Indicators and is available for 113 states from 1960 to 2002, with an average of 15.4 observations for each state. The variable covers all state expenditures on primary, secondary, and tertiary education and at both subnational and national levels. Across the sample this variable has a mean of 4.2 and a standard deviation of 1.9. Private education spending is also measured as a percentage of national income. Finally, the composition of education spending is measured by using the ratio of per student funding on tertiary (university) education versus that on primary education.¹⁷

This measure operationalizes the “depth” of education spending: that is, the proportion targeted to high-income groups.

The model developed in the last section implied that whether the high- or middle-income group controls political decision making affects the level of education spending. To capture this mechanism I use the polity index.¹⁸

This variable ranges from −10 (the most autocratic state) to +10 (the most democratic state) by 1-point increments. While the polity score is comprised of a variety of scores of political competitiveness and restrictions on executive participation and recruitment, which add a layer of complexity not fully represented by the model of the previous section, the overall −10 to 10 range corresponds fairly effectively to the formal model's depiction of extending political representation from an autocratic elite to a democratic mass.¹⁹

To operationalize openness, I first use the log of exports plus imports over GDP. While this is a commonly used measure, it is highly affected by state size.²⁰

Accordingly, I incorporate country fixed effects in some regressions, thus averaging out such idiosyncrasies and focusing on dynamic changes. I also employ a variable created by Hiscox and Kastner that measures the distance of a state from its gravity-model-predicted optimal level of trade.²¹ This distance is then compared to the “ideal” standard of the Netherlands in 1964, with the scale representing the percentage reduction in imports that arises from deviating from this free-trade benchmark. Greater values on the index imply higher levels of de facto protectionism. Critically, this measure examines the outcome of trade policies, including both tariffs and nontariff barriers.

The log openness and hiscox-kastner variables pick up slightly different aspects of economic integration. log openness includes export levels, which indicate the ability of citizens to sell their skills internationally, thereby reducing the dependence of wage returns on domestic skill supply. The hiscox-kastner variable focuses on the deviation of imports from expected levels, indicating the degree to which domestic firms are protected from international competition, and presumably also from technology transfer. Hence the log openness variable more closely resembles the supply-side mechanism of propositions (2a) and (2b), whereas the hiscox-kastner variable resembles the demand-side mechanism of proposition (2c), although this distinction is not sharp.

I also include a number of control variables in the data analysis to capture key factors that may affect educational expenditure. I include government consumption (excluding education) as a percentage of GDP to control for the size of the public sector, which is known to be larger in democracies.²²

I control for the proportion of population under age fifteen to capture demographic pressures on education spending. I also control for a state's logged GDP (and its square) to account for possible “Wagner's Law” effects, and for logged population, to account for scale economies in the provision of public goods.²³ Finally, a linear time trend is also used to capture the secular increase in educational expenditure.²⁴

Table 1 presents the first set of tests, which analyze the effect of regime type and openness on aggregate education spending. The first test in Model A uses the Beck-Katz panel-corrected standard error (PCSE) technique.²⁵

The PCSE technique adapts ordinary least squares (OLS) regression to take account of the temporal autocorrelation and cross-national contemporary correlation that bedevil panel data. The model is Y_it = β₀ + β₁Y_it−1 + β₂ X_it + γZ_it + e_it, where Y_it is the dependent variable, Y_it−1 is a one-period lag of the dependent variable, X_it includes the key independent variables, Z_it is a vector of control variables including regional dummies, and e_it is an AR1 error term with composition e_it = ρe_it−1 + ε_it. The PCSE method then adjusts the resulting standard errors for contemporaneous correlation. The second test, in Model B, uses the fixed-effects panel data method, which gives each state a different time-invariant error term. This method controls for the fact that states may have idiosyncratic education spending, perhaps for cultural or religious reasons not accounted for by the formal model. The model used in the fixed-effects analysis is: Y_it = β₀ + β₁Y_it−1 + β₂ X_it + γZ_it + u_i + ε_it, where u_i is a country dummy. Although the fixed-effects model removes cross-sectional differences from the analysis, it more closely corresponds to the mechanism developed in the formal model, which analyzed changes within states.

Models C and D deal with a potential problem in the analysis: the possibility of reverse causation. It is conceivable that education may itself lead to democratization, a theme touched on in many works of modernization theory.²⁶

If this effect is real, then the resulting endogeneity of the polity score will seriously bias the coefficients because it will correlate with the error term. Consequently, I develop two instruments for democracy, used in Models C and D. Developing instruments for usage in fixed-effects models is challenging because it requires finding measures that covary with the potentially endogenous variable across time rather than space. Thus, common cross-sectional instruments for democracy, as with colonial origins or religion, are not appropriate. The first instrument I employ is the five-year lag of the polity score, used in Model C, which has a high correlation with the polity score (0.86).²⁷ However, there are a number of serious problems with using lags as instruments, the chief being that any serial correlation in the error term will retain the bias. Consequently, Model D uses the average polity score in a country's region as the instrument (correlation of 0.74). This variable is substantively exogenous: there is no reason to expect that the level of democracy in a region is caused by any one state's education policy.²⁸ Both Models C and D employ two-stage least squares regression using these instruments and incorporate country fixed effects.

Models E and F repeat the analysis from Models A and B but replace log openness with the hiscox-kastner measure, which reduces the data set from 113 to sixty-five countries.²⁹

Finally, Models G and H test propositions (1b) and (2d), examining the effects of democracy and openness on private spending as a percentage of GDP. As dynamic data availability of this measure is poor, I employ a cross-sectional regression, using country-means for the 1990s, which reduces the data set to forty-four states.

Across Models A through F in Table 1 the coefficient relating the polity score to public education expenditure is statistically significant at the p < .05 level, except in Model D where it is significant at the p < .10 level. The effects of increased openness are also all statistically significant—in both log openness and hiscox-kastner forms—at the p < .05 level, except for Model C, which is significant at the p < .10 level. In order to interpret these coefficients, it is most useful to examine long-run first differences. To calculate these effects I use the following formula: β/(1 − γ), where β is the coefficient on the independent variable and γ is the coefficient on the lagged dependent variable. This index is then multiplied by two standard deviations of the independent variable (15 points on the polity index; 1.06 on log openness; and 30 on the hiscox-kastner index).³⁰

The estimated long-run effect of this change in the polity score is an increase in educational expenditure of between 0.46 and 1.54 percent of GDP. This would amount to between a 10 and 30 percent increase in the education budget of the average state in the data set. A two-standard deviation change in openness, as measured by the log openness and hiscox-kastner variables, is estimated to lead to a change in education spending of between 0.63 and 1.44 percent of GDP (excluding the large but statistically weaker result in Model C). The similar magnitude of these effects raises the question of whether democracy and openness might be substitutes in terms of increasing education spending, as suggested in proposition (2b) and tested in Table 2. These predicted effects of changes in democracy and openness are substantively large—amounting to between one half and two within-country standard deviations on the dependent variable.³¹

I conclude by examining Models G and H, which test the effects of democracy and openness on private education spending on a cross-sectional data set of country means. Model H differs from Model G by adding a control for public education spending. In both models I find a robust and substantial negative effect of democracy on private spending, with a fifteen-point shift on the polity score associated with a reduction in private spending of 1.8 percent of GDP. This support for proposition (1b) is amplified by the fact that no other variable comes close to statistical significance. Even in unpromising data conditions, democracy appears to have a strong impact on private education spending. However, openness has a negligible effect on private education, as predicted by proposition (2d).

Table 2 splits the sample into two sets of subsamples—first, autocracies and democracies in Models A through D; and second, Organization for Economic Cooperation and Development (OECD) and non-OECD states in Models E through H. All models use fixed-effects regression and are thus directly comparable to Models B and F in Table 1. Models A and B compare the effects of log openness on education spending in autocracies and democracies. In both cases, the coefficient on openness is positive but it is only statistically significant in the autocracy subsample. This result is likely to reflect the greater impact of supply-side forces on education in autocracies, as implied by proposition (2a). However in Models C and D, which compare the effect of the hiscox-kastner variable in autocracies and democracies, I find coefficients of similar magnitude and statistical significance. Here the models show that deviations from “free trade” levels of imports have a similarly negative impact on education spending in both autocracies and democracies. Though the analogy is imperfect, if the hiscox-kastner variable better picks up demand-side forces such as technology transfer, with log openness picking up supply-side forces, this finding suggests that demand-side forces have equal impact on education across autocracies and democracies, whereas supply-side forces have a different impact, as implied by propositions (2b) and (2c).³²

I directly test proposition (2c) in Models E through H, which break the sample into OECD and non-OECD subsamples.³³

Models E and F, using the log openness measure, show a striking pattern. In developing countries, the impact of openness is strongly positive, whereas in developed states it is strongly negative. For developing countries, this is likely a result of the demand-side impact of technology transfer combined with their greater likelihood of being autocratic. Why, however, is the impact of changes in log openness negative in developed states? These states are already democratic; hence the supply effects are minimal. Furthermore they are the originators not recipients of technology transfer, thus weakening the demand-side effects. Moreover, most of these states have been open to trade since 1960. Consequently, the main causes of changes in log openness are likely to be income or trade shocks, which cause generalized contractions in government spending or shift resources from education to public spending that buffers against economic volatility, such as social insurance.³⁴ Models G and H, using the hiscox-kastner measure, find a positive and significant effect of opening up in developing states and an insignificant effect in developed states. Since this variable controls for GDP, economic shocks are removed from the measure. Hence, the negative impact of log openness from Models E and F is replaced by the insignificant impact predicted by propositions (2b) and (2c).

I conclude the empirical analysis with Table 3, which tests propositions (3a) and (3b) on a data set of country mean spending on primary and tertiary education during the 1990s. Because time series evidence on the composition of education spending is fragmentary, it is necessary to conduct a between effects analysis on the data rather than analyzing dynamic patterns. Models A through C use the standard set of controls, whereas Models D through F also include measures of overall enrollment levels in tertiary and primary education, in order to isolate the impact of increased per student spending from changes in enrollment. Each set of tests examines three subsamples: first, all the cases; second, non-OECD states; and third, OECD states. Models A through C show the expected effect of democracy on the tertiary to primary spending ratio as presented in proposition (3a). The ratio variable has significant variability in the data set, ranging from 0.4 to 85. A two standard deviation shift in polity is predicted to lead to a reduction in the ratio variable of fifteen points in the full sample and twenty-three points in developing states. For OECD states the negative effect remains but is much smaller (the highest level of the ratio is 4.6 in the OECD) and statistically insignificant. These patterns are robust to inclusion of the enrollment variables in Models D through F. Turning to the effect of openness, I find ample support for proposition (3b). In the full sample there is a statistically weak negative effect of openness on the ratio variable. However, in developing states there is a robust and sizeable negative effect of openness on the ratio variable, as states move resources to sectors of comparative advantage, which demand lower-skilled labor. Conversely, Model C shows a robust positive effect of openness on the ratio in OECD states, whose comparative advantage lies in university-educated labor. Models D through F replicate this pattern, controlling for enrollment. However, Model F shows that the statistical significance of openness on the ratio is weakened in the OECD once I control for enrollment levels. Overall, however, propositions (3a) and (3b) appear strongly supported by this cross-sectional analysis.

Bringing together the statistical results from this section, there are robust and substantive effects of both regime type and openness on a variety of forms of education spending. Aggregate public education spending is estimated to increase dramatically, by up to 30 percent, following shifts of two standard deviations in democracy or openness. Private spending is also strongly affected by democracy, this time negatively, although openness appears to have a negligible impact. The estimated impact of openness on public education spending varies significantly across states, with autocracies and developing states seeing stronger effects. Finally, I found that democracy is everywhere associated with a shift in spending from tertiary to primary education but that the effect of openness reverses when one moves from developing to developed states.

The Philippines: Examining Democratic and Trade Volatility

The previous section demonstrated a robust relationship between democracy, openness, and education spending over a broad data set. But the aggregate nature of statistical analysis necessarily glosses over how the casual mechanisms developed in the formal model actually play out in historical practice. Accordingly, this section traces these mechanisms in the Philippines, which has experienced significant volatility since 1946 in regime type, economic openness, and education spending, thus providing a fertile test of the propositions outlined in the formal model. Furthermore, the Philippines is a strong test case because it is commonly portrayed as having an idiosyncratically pro-education culture, where one might expect uniformly high spending.³⁵

Yet this cultural affinity for education has proved a weak bulwark against the tides of politics, because education spending in the Philippines has varied widely since independence. As I shall show, changes in regime type and openness constitute the most telling explanations of this volatility.

Following independence in 1946 the Philippines had a moderately democratic political system, which soon responded to pent-up popular demand for increased education spending. Consequently, under the Liberal government of Diosdado Macapagal, enrollment in primary education expanded from 56 percent to 97 percent between 1960 and 1965.³⁶

In 1966, Ferdinand Marcos was democratically elected president and one of his earliest acts of legislation was the 1966 “Magna Carta for School Teachers.” This act represented the culmination of popular demands for an increasingly professionalized education system in the Philippines and led to a 25 percent increase in education spending between 1960 and 1970. Teachers were granted academic freedom, the ability to unionize and, most critically in terms of education funding, a guarantee that salaries would be cost-indexed.³⁷ This spending on both the quantity and the quality of teachers was a necessary counterpart to the broader increases in enrollment that accompanied bringing education to the masses.

In 1972, however, Marcos declared martial law, forcing a transition to personal rule. The authoritarian turn precipitated a sharp decline in education from 2.8 percent to below 2 percent of GDP. Marcos's policies increasingly reflected the education spending preferences of his elite client base. As Kang notes, “under martial law in the 1970s the priorities of the government, as reflected in the budgetary allocations to defense and industry, shifted away from education,” with education spending declining from 29.1 percent of government spending in 1965 to just 11.4 percent by 1975.³⁸

Consequently, salaries for teachers dropped well below the poverty line. While resources exited the public education system, autocracy led to an increase in private spending. Enrollments in private primary schools increased 2.5 times faster than in public schools between 1965 and 1987.³⁹

It was not until the late 1980s, after Marcos was forced to relinquish power, that educational expenditure recovered to its pre-1972 level, under the post-Marcos government of Corazon Aquino. One of Aquino's first acts as president was the convocation of a Congressional Commission study on education (EDCOM), which highlighted the collapse in education funding relative to comparator nations and recommended massive reinvestment, particularly in teachers' salaries. Consequently, by 1998, educational expenditure had tripled its late-Marcos level and climbed to over 4 percent of GDP, with poorer Filipinos the chief beneficiaries of expanded coverage, thanks to the stipulation that education be made available free at the primary and secondary levels, stated in Article XIV of the Philippine constitution of 1987.⁴⁰

An additional constitutional stipulation stated that education should be the largest proportion of the state's budget and amount to at least 6 percent of GDP. Although this target remains unmet, it demonstrates the determination of the new democratic government to target education as a key policy. It is also a useful confirmation that public education expenditure as a percentage of GDP is an appropriate dependent variable for the empirical analysis in this article, given that this variable was formally incorporated into the Philippine constitution. Together, these constitutional requirements amply demonstrate the close relationship between the transition to democracy and the preferences of the masses over education spending, given the new emphasis on increased spending and enrollment entitlements. The interlinked pattern of regime type and education spending can be seen in Figure 1, which displays the polity score and public spending on education in the Philippines from 1960 to 2000.

Education funding in the Philippines has not solely been driven by political regime change. The integration of the Philippine economy with the global economy has varied considerably over the past fifty years. Figure 2 shows the change in the inverted hiscox-kastner variable over this time period.⁴¹

Although there has been an upward trend in openness throughout the period, fluctuations off this trend appear to covary with those in education spending. Particularly noticeable is the decline in openness during the latter period of the Marcos dictatorship from 1974 to 1986. Following the 1986 return to democracy, openness shoots upward, essentially doubling by the mid-1990s, with a commensurate increase in education spending. The relationship between openness and education spending appears tight, but democracy and openness were also closely related during this period in the Philippines. In order to separate out the effects of openness from political change, I turn to a brief examination of postwar Philippine trade policy.

Political independence in 1946 did not foster economic independence since the Philippines signed the Bell Trade Act with the United States, which established a series of asymmetric tariffs and U.S. quotas on imports from the Philippines: as Hawes notes, “it was a splendid example of neocolonialism.”⁴²

The restrictions on trade limited the development of an industrial export orientation. Instead, Philippine industrialization turned inward, focusing on a heavily protected internal market.⁴³ As in India—discussed in the next section—control over the export sector was seized by a few well-connected families, who ran conglomerates combining agricultural and industrial businesses. Since industry was heavily protected, there was little technology transfer from abroad and hence demand for education failed to spike upward. Instead the Philippine economy began to mimic Latin American, rather than East Asian, development with little emphasis placed on educated labor.⁴⁴

Marcos's declaration of martial law in 1972 was followed by a rhetorical shift toward export-oriented trade policy. Rhetoric was, however, largely unmatched by action, since Marcos, once keen to remove industrial firms from elite family control, set about protecting industry in the name of his cronies. Export zones were established but remained niches. As McKay explains:

Thus, Marcos's rhetoric bore little resemblance to the structure of the Philippine economy outside the export zones. With manufacturing protected and uncompetitive, the Philippines did not face the same demand for primary and secondary education felt in the East Asian “Tiger” economies. Contrary to Philippine comparative advantage, it was spending on higher education that grew most dramatically: with an annual growth rate of 5.2 percent during the early 1980s versus just 1.1 percent for primary education.⁴⁶

With local elites and Marcos cronies controlling these protected sectors, the Philippine economy could not absorb these highly educated workers. Surveying the Marcos legacy in higher education, the Education Secretary Lourdes Quisumbing remarked in 1987 that “we have the supreme irony of [higher] education proceeding much faster than economic development, and creating difficult burdens for the country in terms of an educated unemployed.”⁴⁷

Political liberalization in 1986 was swiftly followed by economic liberalization under Aquino—who removed most import controls—and during the administration of Fidel Ramos. Following Ramos's election in 1992, education spending spiked upward to more than 4 percent of national income. Ramos's “Philippines 2000” strategy explicitly connected trade liberalization and education spending, with an emphasis on the following the East Asian model. Under Ramos's administration, manufactured goods became the majority of Philippine exports for the first time, forcing up the demand for an educated labor force to aid in their production.⁴⁸

The post-Ramos era has continued along the same lines with manufactured exports supported by an increasingly educated Philippine workforce. Overall, the dual transition to democracy and export-orientation has strongly supported Philippine education spending, just as the combination of Marcos's autocracy and protectionism undermined it. I now turn to two states where democracy and openness have not coincided but rather the reverse.

Explaining Two Puzzles—India and Malaysia

India and Malaysia provide symmetric, contrasting puzzles. Why has India, perhaps the most successful democratic experiment in the developing world, failed so dramatically in terms of providing education to its citizenry? Why, conversely, has authoritarian Malaysia had higher levels of education spending than any other major Asian state? Examining openness allows one to address these twin puzzles, where regime type appears an insufficient explanation. Furthermore, these cases test whether the casual mechanism developed in the first section works in countries that are “off the diagonal” of autarkic autocracies and open democracies.

India's failure to educate its workforce, particularly in rural areas, is infamous. Half of the Indian population remains illiterate, comprising 40 percent of the world's illiterates.⁴⁹

This outcome seems surprising on a number of levels. First, India has been a well-functioning democracy since 1947. Second, Indian political rhetoric has consistently stressed the importance of education.⁵⁰ However, this rhetoric has rarely been translated into effective action: even today India spends just 4 percent of GDP on education. This disconnect is in part explained by Indian trade policy. India's industrialization strategy of import substitution emerged in the mid-1950s as a compromise between economic nationalism (swadeshi), socialism, and Gandhian village production. This strategy choked off foreign investment and hence foreign technology, reducing the demand for skilled labor. Furthermore, the difficulty of obtaining export licenses in the “permit, license, quota Raj”⁵¹ led to the absence of an external market for skilled labor. This meant that increases in education supply, such that there were, were swiftly followed by unemployment or lowered wages.

The failure to meet government rhetoric about education began in the 1950s, with the composition of India's Five Year Plans for development of infrastructure and social services. During the First Five Year Plan, education spending amounted to a fairly substantial 8 percent of all spending, half of which was devoted to elementary education. While this only amounted to about 15 percent of overall education spending, the share of education spending in the Plans directly reflected government preferences over development, rather than nonpolicy variables such as demographic pressures. After the First Five Year Plan, however, education spending dropped to 5.8 percent of spending in the Second Five Year Plan and to just 2.7 percent by the Sixth Five Year Plan in the early 1980s. During that period the emphasis on elementary education was replaced by investment in higher education, with the former sector's share of spending reduced to 30 percent. Tilak notes that “1956 to 1969 marked the beginning of a drastic decline of resources allocated to elementary education and a doubling or trebling of resources allocated for higher education.”⁵²

Thus, India's developmental strategy was focused away from areas of comparative advantage toward import-substituting high-end manufacturing. The government began to limit investment in education and channel it toward the children of the elite (those most likely to attend higher education) in an ill-fated attempt to jumpstart the Indian industrial sector. However, because Indians were exporting few manufactures abroad, the limited size of the market for such goods led to an oversupply of high-end skills and a consequent collapse in the returns to such education. Tilak notes a “growth of educated unemployment [and] mismatches in the labor market” by the mid 1970s.⁵³

This is the pattern outlined in the first section: autarkies cannot sustain the same proportion of high-skilled workers as open states. The growth of educated unemployment led to a further turn away from education funding in the 1970s and education essentially ceased to be a significant component of the Indian Five Year Plans until the late 1990s. Non-Plan education spending hardly fared much better, hovering at around 2 percent of national income until the late 1980s.

The Indian development strategy from the late 1950s to 1990s, based on autarkic industrialization, artificially deflated education spending and likely hindered Indian economic growth. Frustration with the slow “Hindu rate of growth,” combined with an economic crisis in the early 1990s, led to a breakthrough in trade orientation during Narasimha Rao's administration, elected in 1991. Rao's finance minister Manmohan Singh developed a broad range of liberalization policies, which permitted majority foreign ownership of Indian companies, reduced tariffs, and eliminated many quotas. In doing so, he catalyzed the development of India's rapidly growing high-technology and outsourcing industries, with the growth rate increasing from around 5 to 7 percent. Currently, and in stark contrast to the experience of the 1970s, India cannot educate workers fast enough at present for the skill demands of the newly globalized Indian companies and Western multinationals engaging in outsourcing.⁵⁴

The government has nonetheless been expanding education spending significantly, spurred by business demands, with a surge of 1 percent of GDP during the 1990s.

The Malaysian experience provides a dramatic contrast to India. Since the 1960s, Malaysia has spent more than 5 percent of its national income on public education and at various points, including the early and late 1990s, more than 7 percent of GDP, which placed it firmly in within the top decile of international spenders. However, for most of this period, Malaysia has been an authoritarian state, with a powerful executive that is largely uncontrolled by the legislative and judiciary. Thus, on a political level, Malaysia does not appear a likely candidate to spend large amounts on public education. However, because of Malaysia's early entry into the global economy, these unpromising initial conditions have nonetheless supported major investments in education.

Malaysia's experience with British colonialism was mixed. Malaysia was abundant in natural resources but colonial government was beneficial largely for the British and for the Chinese and Indian immigrant communities. The indigenous Malays fared less well, becoming a numerical and economic minority. As in most colonies, there was little interest in providing public services for the indigenous population: as Snodgrass notes, “colonial governments in Malaya provided little education … most of what happened occurred through private rather than public initiative.”⁵⁵

Consequently, the run-up to independence was accompanied by a new set of demands for education by the indigenous Malays, culminating in the release of the Razak Report in 1956, which recommended that primary education be publicly offered to all children. This goal was achieved with astonishing pace, particularly in comparison to other developing countries with similarly ambitious aims: by 1960, 90 percent of Malaysian children were enrolled in primary schools. This expansion was contemporaneous with Malaysia's program of early industrialization, which led to increasing exports in manufactured goods requiring a more educated workforce. As in the Philippines, these early steps in education spending came at a time of relative democracy for Malaysia. However, the 1969 race riots by the Malays against the wealthier Chinese led to a political tightening, resulting in a semi-autocratic regime. Education spending, however, did not suffer in the turn to autocracy. As argued in proposition (2b), openness proved to be a substitute for democracy in terms of promoting education spending in Malaysia.

The post-1969 shift toward export-oriented industrialization led to further education investment. The New Economic Program (NEP), lasting from 1970 to 1990, had as a key aim “the development of a strong human resource orientation in order to be internationally competitive during the process of industrialization.”⁵⁶

The NEP promoted export-oriented assembly industries, in particular, apparel and the new electronics industry, both of which demanded moderate educational skills, and employed imported foreign technology.⁵⁷ Education provision was expanded to meet these technology-induced demands: secondary enrollment by 1990 was nearly 60 percent. The push from globalization meant that by the late 1970s, skills gaps were emerging in exporting industries, pushing up skilled wages despite the increases in enrollment. This contrasts sharply with the wage decreases that accompanied increased enrollment in autarkic India.

The New Development Program, established in 1991, responded to these export-led demands for skills. The key aim during the 1990s was to supply high-end skills that would facilitate the transition to production higher up the value-chain. Skills shortages were now considered a key threat to further development as international demand for Malaysian goods pushed up demands for skills. As implied by proposition (3b), Malaysia's transition to developed status precipitated a shift toward tertiary education spending during the 1990s. In response to this development, the government has pushed the education budget to more than 7 percent of GDP.

This section has demonstrated the substantial impact of trade policy on education across both democracies and autocracies. Although ostensibly democratic, India lagged in education spending due to its autarkic trade policy. Malaysia, though authoritarian, nonetheless invested heavily in education thanks to its export orientation. Why did regime type not overwhelm the effect of openness? One possible answer is that both India and Malaysia have mixed regime types.⁵⁸

The stultifying impact of one-party rule and the caste system on education could explain India's failure to meet rhetoric with education policy.⁵⁹ Similarly Malaysia's Malay political elite preferred to target education at the poorer Malay population and away from the Chinese economic elite, thus weakening the impact of authoritarianism. Still, these structural explanations varied weakly across time and thus fail to explain the changes in education spending in these states, particularly India, which retains the caste system but has opened its borders. Openness then provides a key insight into variation in education spending across a broad array of regimes.