Institute of public finance
Public Sector Economics conference
Call for papers
LOGIN / REGISTER
Public Sector Economics
HOME
ABOUT
FOR AUTHORS
SUBMIT YOUR PAPER
ARCHIVE
CONTACT
Print page
Search box

822
Views



472
Downloads

Education spending, economic development, and the size of government



Mark Millin*
   
David Fielding*
   
P. Dorian Owen*
Article   |   Year:  2023   |   Pages:  285 - 333   |   Volume:  47   |   Issue:  3
Received:  September 29, 2022   |   Accepted:  June 21, 2023   |   Published online:  September 4, 2023
Download citation        https://doi.org/10.3326/pse.47.3.1       


 

Abstract


We examine the association between economic development and two measures of public spending on education: the “national effort” (public spending on education as a proportion of GDP) and “budget share” (public spending on education as a proportion of total government spending). Using panel data for a large sample of countries from 1989 to 2015, we compare mean levels of national effort and budget share measures for economically and politically distinct groups of countries. We find that economically more developed (richer) countries are characterised by a higher national effort and a lower budget share than less economically developed countries. This implies that richer countries, on average, have larger public sectors than poorer countries, consistent with Wagner’s law and Baumol’s “cost disease” hypothesis.

Keywords:  education spending; Wagner’s law; Baumol’s cost disease; economic development; democracy

JEL:  H52, I22, I25

1 Introduction


Public spending accounts for the lion’s share of the financing of education in most countries; indeed, the “massification” of education is made possible through public provision. Two headline measures of public education spending, namely the “national effort” (total public spending on education as a proportion of GDP) and the “budget share” (total public spending on education as a proportion of total government spending), are commonly used to compare the financing of public education across countries. Whether or not richer (developed) countries spend more on public education than do poorer (developing) countries, regardless of which of the two ratio measures is used, is a matter for empirical inquiry. In this paper, we therefore aim to test whether, on average, richer (developed) economies have larger public education sectors than poorer (developing) economies, in both national effort and budget share terms. As well as providing a global comparative view of education spending patterns, this offers a novel perspective on the implications for the size of the public sector (total government spending as a percentage of GDP) as income per capita increases.

The idea that the size of the public sector is positively related to the level of economic development is not new. Wagner’s “law of increasing state activity”, for instance, points to an apparent empirical regularity whereby an increasing share of overall government expenditure in the national economy is associated with rising income per capita. Wagner (1892; 1958) attributed expansion of the public sector to continued cultural and economic progress, which has associated social, welfare, regulatory and infrastructural requirements that necessitate a growing role for government spending in the economy (Kuckuck, 2014). 

Baumol’s “cost disease” hypothesis (Baumol, 1967; Baumol and Bowen, 1966) also predicts a growing public sector as a proportion of the economy. This is attributed to higher labour intensity and lower productivity growth in the public sector (for example, in education and medical care) than in the private sector. Technological advancement, innovation and substitution of capital for labour lead to increases in wages in the private sector, which are mirrored as cost increases in the public sector. Under this explanation, public sector expansion is largely cost-driven.

The aim of this paper is to examine whether there exist differences in the mean levels of the national effort or budget share measures of education spending for economically (and politically) distinct groups of countries. By exploring the patterns of public education spending, we provide, as a by-product, insights into differences in the size of government for richer versus poorer countries. If the national effort and budget share measures are both larger for richer than for poorer countries, differences in the size of government are indeterminate in the absence of additional information. However, if richer countries have a larger national effort, but a smaller budget share than poorer countries, then this necessarily implies that richer countries, on average, have larger public sectors than poorer countries.

Several hypotheses can be formulated from the relevant empirical literature. The ability of publicly provided education to reach all parts of society (the massification of education), makes education a useful conduit through which social, cultural and economic progress (for example, human capital development) can be advanced. Public financing of education is, therefore, expected to expand along with overall public spending as part of governments’ efforts to promote economic growth and development, especially if education is viewed as a merit good and a productive component of public spending.

To measure economic development, income per capita has been widely used as an explanatory variable in studies of education spending (Shin, 2020; Afonso and Alves, 2017; Cockx and Francken, 2016; Garritzmann and Seng, 2016; Dragomirescu-Gaina, 2015). The intuition is that richer, more developed, countries have greater resources with which to fund various social programmes, such as education (Brown and Hunter, 2004). The evidence suggests a positive relationship between the national effort measure and economic development (Shin, 2020; Cockx and Francken, 2016; Akanbi and Schoeman, 2010; Huber, Mustillo and Stephens, 2008; Busemeyer, 2007; Stasavage, 2005; Baqir, 2002; Ram, 1995; Tilak, 1989).

Evidence concerning the budget share is more limited. The few studies employing this measure mostly report a positive association between budget share and economic development (Fosu, 2010; Stasavage, 2005; Baqir, 2002), although the relationship is not always significant, and the studies by Fosu and Stasavage are concerned only with African countries. Angelov (2019) provides an example of a more recent study that employs a budget share measure of education spending to compare European Union countries’ education spending but does not investigate the relationship between education spending and economic development. However, it is reasonable to suppose that, as countries grow and develop, the size and complexity of their respective public sectors (the variety of public goods to be financed by government) will grow, so education could end up constituting a reducing share of the total budget allocation, ceteris paribus. A negative association between the budget share and the level of development would be likely if education is a “necessity” with respect to total government spending.

The type of political regime is also relevant in an analysis of education spending. Regardless of the outcome measure (national effort or budget share), democratic countries are expected to spend more on education, ceteris paribus. It is well documented in the political economy literature that democracy is positively associated with the public provision of basic services, such as education (Baum and Lake, 2003; Lake and Baum, 2001), although there are different views about the exact mechanisms underpinning this association (Harding and Stasavage, 2014). On the one hand, spending more on socially productive public goods, such as education, provides a politically popular way for governments to demonstrate accountability and broaden their voter pool. Brown and Hunter (2004), for example, make this point with respect to spending on primary education in Latin America. On the other hand, evidence also exists for democratic developing countries (e.g., Brazil) that poorer electorates prefer government to allocate spending to areas other than education (Bursztyn, 2016); hence, a negative association between democracy and education spending is possible. However, overall, many empirical studies find evidence that public education spending is higher in democracies (Murshed et al., 2022; Shin, 2020; Garritzmann and Seng, 2016; Avelino, Brown and Hunter, 2015; Stasavage, 2005; Baqir, 2002). Consequently, in our analysis of public education spending, we categorise countries by political regime (democratic versus nondemocratic) as well as by levels of income, while controlling for other social and economic factors.

The rest of this paper proceeds as follows. In section 2, we describe the data and outline the empirical method to be applied. In section 3, we report the empirical results, including checks for robustness. The main findings are discussed in section 4, and section 5 concludes.


2 Data and empirical methods


We use annual panel data from 1989 to 2015 for up to 193 countries, although the number of available observations depends on the variables being considered. Table 1 presents details of the data collected. Two different continuous outcome measures for public education spending are examined, namely the national effort (total public spending on education as a proportion of GDP, pse/gdp) and budget share (total public spending on education as a proportion of total government spending, pse/gov). Three key categorical explanatory measures are used because our aim is to compare education spending for economically and politically distinct groupings of countries. The level of economic development (ypc2015) is represented by a set of dummy variables, categorising countries into five groups adapted from the World Bank’s Country and Lending Groups as at 2015. These are based on gross national income (GNI) per capita in US dollars using the World Bank’s Atlas method, which smooths exchange rate fluctuations and provides a comparable crosscountry measure for grouping countries by income per capita. The sample contains representation across the full range of income levels. The richest group consists of the 21 wealthiest, long-standing Organisation for Economic Co-operation and Development (OECD) “core” countries; these constitute the same set of countries examined by Busemeyer (2007). The other four groups are high-income (mostly non-OECD), upper-middle-income, lower-middle-income, and low-income countries. Appendix table A1 gives a list of countries included in each group.

Table 1
Data definitions and sources
DISPLAY Table

An alternative classification of countries by development status is based on a binary richer-country/poorer-country split, defined in terms of regional country groupings (region). Appendix table A2 provides a list of countries included in each group. A binary perspective on education spending patterns can be explored by using a pair of regional dummy variables representing rich versus poor countries.

A classification of countries depending on whether they are democratic or nondemocratic (poldemoc) is used to represent different political regime types. A classification of countries by regime type (democratic versus non-democratic) is not listed because this can vary over time. For each of the key categorical explanatory measures, sample selection bias is mitigated because the economic groupings of countries are invariant over the study period, and the political regime type (democratic versus non-democratic) typically varies only very slowly over time in most countries.

Several potentially important control variables are included in the analyses. The size of the school-going population up to age 24 (pop024) captures the positive demographic effect of the proportion of young people on education spending (Busemeyer, 2007, 2008; Brown and Hunter, 2004; Castles, 1989). The urbanisation ratio (urban) captures the positive effect of a greater concentration of the total population in urban areas on a government’s propensity to act in favour of fundamental social needs, such as education (Akanbi and Schoeman, 2010; Huber, Mustillo and Stephens, 2008; Avelino, Brown and Hunter, 2005; Baqir, 2002; Schultz, 1988). Total international trade (trade) is often included in empirical analyses of education spending (Ozkok, 2017; Busemeyer, 2009; Huber, Mustillo and Stephens, 2008; Iversen and Stephens, 2008; Kaufman and Segura-Ubiergo, 2001). This allows for two possible effects: a positive compensation effect, in which government “compensates” society for the adverse effects of globalisation through greater social and welfare spending, and a negative efficiency effect, in which government sees increased globalisation as a mechanism to promote competitiveness, reducing the need for social and welfare spending.1 Which trade effect dominates is an empirical question.

A number of other control variables are used for robustness checking. The size of the population aged 65 and above (pop65) represents a demographic cohort that competes for education spending in the form of transfer payments to the elderly population (Shin, 2020; Busemeyer, 2008; Iversen and Stephens, 2008; Avelino, Brown and Hunter, 2005; Brown and Hunter, 1999). Military spending (military) is also expected to compete for education’s share of public resources, especially in countries with a large military presence (Shin, 2020; Baqir, 2002). The fiscal balance (fiscbal) and gross public debt stock (debt) are both expected to have implications for how much of the public purse is allocated to education (Busemeyer, 2009; Huber, Mustillo and Stephens, 2008; Tilak, 1989, 1990). Human capital development, as measured by the Penn World Table (Feenstra, Inklaar and Timmer, 2015) human capital index (hci), is not typically used in this empirical literature, but is included to control for the current-period stock of human capital as a proxy for the quality of education in a country.

Pooled descriptive statistics for each variable are reported in appendix table A3. Data availability is a pervasive problem in the literature on education spending. The two measures of education spending are available for fewer countries (N) and a smaller average number of time-series observations than are any of the explanatory variables: the sample is roughly half as large in most cases.

The approach we adopt – one-way or two-way ANOVA and ANCOVA, with the focus being a two-way factorial analysis of covariance – aims for a descriptive characterisation of average differences between broad groupings of countries, rather than implying specific causal linkages. The method is a variant of fixed effects estimation, but instead of estimating country fixed effects, more highly aggregated group effects are estimated. An advantage of this method is that it is possible to estimate mean differences in the groups of interest while controlling for other relevant variables. The regression equations include interactions of political and economic dummy variables, allowing for different intercepts in each political-economic group. However, no other interaction terms are included, and the parameters for the controls are assumed to be constant across all countries. Allowing for heterogeneous group parameters would mean having to interact all the group dummies with the control variables, leading to a proliferation of explanatory variables and excessive multicollinearity

The models in equations (1) and (2) represent the empirical specifications to be tested. Separate single-equation models are estimated for national effort and budget share. In the model in equation (1), we include interaction terms between categorical variables for five economic groups and two political groups (democratic, non-democratic), yielding 10 categories. In the model in equation (2), we include interaction terms between categorical variables for two regional groups (richer, poorer) and the two political groups, yielding four categories.

(1)


(2)

Here, Y is either the national effort or budget share measure of total education spending; Ej (j = 1, …, 5) constitutes a set of five (1/0) dummy variables, one for each of the five GNI per capita country groups; Pm (m = 0, 1) is a set of two (1/0) dummy variables, one for each of the political groupings, i.e., democratic, (m = 1) or non-democratic (m = 0); Rr (r = 0, 1) is a set of two (1/0) dummy variables, one for each of the two regional country groups (poorer or richer); Xn (n = 1, …, N) is a set of continuous control variables comprising a minimum of three or a maximum of eight controls; and ε is a generic random error term. Subscripts i and t denote observations for country i and time t, respectively, and ajm, arm and βn are parameters.

In order to focus on differences in national effort and budget share across groups, we reparameterise equations (1) and (2). We include an intercept term and, if there are k distinct economic/political categories, k-1 dummies are included, to avoid perfect multicollinearity. The base category is then represented by the intercept. For equation (1), the base category is the group of 21 OECD countries that are democratic. For equation (2), the base category is richer countries (or, more accurately, regions comprising the richest countries of the world) that are democratic. In the reparameterised model, the coefficients on the interactions between the dummy variables represent mean differences in the education spending measure for the relevant composite economic/political category relative to the base category. So, for example, for comparisons of different economic groups with a common political categorisation, a series of positive (negative) mean differences indicates that poorer countries have, on average, higher (lower) levels of the education spending measure relative to the relevant base category.

The least-squares dummy-variable (LSDV) estimator with heteroskedasticity-robust standard errors is used to obtain the baseline set of results. We undertake several types of robustness check. First, we report quantile (median) regression and robust regression estimates of the parameters to check for sensitivity to outlier observations.2 Second, we examine a number of different estimators of the standard errors for the LSDV results.3 These include one-way (country or year) and two-way (country and year) clustering, Newey-West heteroskedasticity and autocorrelation consistent (HAC) standard errors (Newey and West, 1987,  1994), and Driscoll and Kraay’s (1998) standard errors, which are robust to heteroskedastic, autocorrelated and cross-sectionally dependent errors. Third, we examine the effects of including time dummies to control for year effects and adding additional control variables (hci, pop65, military, fiscbal, debt). Fourth, we examine the effects of using a continuous measure of GDP per capita (gdppc) as a way to check whether the substantive pattern of results is noticeably different from using our preferred GNI per capita categorisation of countries. Finally, we explore the implications of including a Gini index of income inequality (gini) and hci lagged by one period, and examining different quantiles (0.2, 0.25, 0.4, 0.5, 0.6, 0.75, 0.8) for the quantile estimator; with all these additional specifications we incorporate the main controls (pop024, urban, and trade).4


3 Results


Table 2 reports the main empirical estimates for the national effort and budget share, for the model with 10 economic/political categories; the corresponding results for the model with four categories are reported in table 3. In the tables of results, the coefficient estimates are labelled “j#m” (j = 1, …, 5; m = 0, 1) for equation (1) and “r#m” (r = 0, 1; m = 0, 1) for equation (2). “BASE” represents the intercept estimate. Each estimation method (LSDV, quantile, and robust) is applied to a model with no controls (A), and with three controls (B). Note that there are no non-democratic OECD or richer countries, so there are no results for these combinations.

The most important finding from table 2 (equation (1)), and table 3 (equation (2)) is a reversal in the pattern of mean differences for the levels of the national effort compared to the budget share. Interaction of the economic and political dummies (table 2), or regional and political dummies (table 3), reveals a pattern of significant negative mean differences (compared to the base category) for the national effort but positive mean differences for the budget share. These patterns are similar regardless of whether no controls or three controls are used. When we control for political categorisation, richer (poorer) countries tend to spend more, on average, in national effort (budget share) terms, although the association is not always monotonic.

Whether a country has a democratic political system is associated with its education spending patterns, with significant mean differences within the same economic or regional group. For example, regardless of the spending measure (national effort or budget share), when we control for economic or regional group, democratic countries tend to spend more on average than their non-democratic counterparts. Table 4 reports a summary of the results from a series of pairwise Wald tests, conducted on the robust regression estimates obtained from tables 2 and 3, for the null hypothesis of parameter equality (i.e., no difference in the mean levels of education spending for countries with democratic versus non-democratic systems, within the same economic or regional group). For example, we can test whether the mean level of education spending in low-income democratic countries differs significantly from that of low-income countries that are not democratic. Because the intercept term is the common base category for all economic/political groups, we can ignore that and focus on the differences in the relevant coefficient estimates. We are conducting multiple hypothesis tests, which inflates the overall “familywise” Type I error rate, so we apply a Bonferroni correction to the level of significance used for each individual test by dividing the familywise error rate (set at 0.05) by the number of tests (for example, 0.05/4 tests = 0.0125). Even with such a correction, most pairwise comparisons still reveal statistically significant differences.

Table 2
Mean differences in the national effort and budget share by income group and regime type
DISPLAY Table

Table 3
Mean differences in the national effort and budget share by country region and regime type
DISPLAY Table

Table 4
Wald tests for parameter equality of the factor-variable interactions
DISPLAY Table

Estimated coefficients on the control variables have the expected signs. Both the youth population and urbanisation variables have positive coefficients. The coefficient on the trade variable is positive in most cases, which supports the compensation hypothesis.

The empirical patterns are generally robust to the use of two alternative estimation methods (quantile and robust, reported in tables 2 and 3) and to the use of alternative standard errors for the LSDV estimation (reported in appendix tables A4-A7). The largest standard errors are those clustered by country (as opposed to by year or by country and year). This is not surprising, because there are many countries for which very few observations are available for the dependent variable, and this makes it more difficult to estimate coefficients precisely when clustering by country.

Robustness checks considering differences in model specification (including year dummies and employing more than three controls) are reported in appendix tables A8 and A9 (using LSDV estimation), and tables A10 and A11 (using robust estimation); for these, only the more parsimonious regional and political specification (in equation (2)) is used, because a richer versus poorer interpretation is the key focus of our study. We make three observations about these additional robustness results. Firstly, including year dummies leaves the substantive patterns of mean differences unchanged; signs of the estimated coefficients are unaffected in all cases, although there are some changes in marginal levels of statistical significance for some of the budget share results. Secondly, if a robust estimator is used to deal with outliers, the empirical patterns are exhibited more clearly regardless of the specification used. Thirdly, the signs of the coefficients on the various additional controls (hci, pop65, military, fiscbal and debt) are as expected in most cases. Introducing an additional control each time entails an increasingly more complex specification that either does not confound or only partially confounds the empirical patterns.5 The most comprehensive specification (using eight controls) provides additional support for the empirical patterns in the baseline results. Overall, the observed empirical patterns of negative (positive) mean differences for the national effort (budget share), compared to the base category, are robust to the use of different estimators for the coefficients and standard errors, and to plausible changes to the specification.

Several additional robustness checks use the robust estimator (or, where applicable, the quantile estimator), including the main controls (pop024, urban, and trade) in all cases. These results are reported in appendix tables A12 and A13. Firstly, to check that the general patterns for both measures of education spending are maintained when using a continuous measure of income per capita, the robust estimator is used with GDP per capita (gdppc) and political democracy (poldemoc) as explanatory variables. This also checks whether using GDP per capita (instead of our preferred World Bank Atlas method of GNI per capita country groupings) reveals anything noticeably different about the data patterns. The results are reported in column I of tables A12 and A13. These specifications are consistent with the empirical patterns observed in the main results, with a significant positive coefficient on GDP per capita for national effort and a significant negative coefficient for budget share.6

Secondly, the effect of including a measure of income inequality (gini) (column II of tables A12 and A13) is explored because within-country disparities in income are likely to influence education attainments and, hence, the political motives behind the funding of education. However, poor data coverage plagues the use of a Gini measure (or any other measure) of income inequality, limiting the extent to which meaningful inferences can be made. Nonetheless, the general patterns are maintained, albeit with some inconclusive effects; the latter is not surprising given the considerably reduced number of observations available when introducing a Gini measure. The Gini coefficient itself is not statistically significant in the national effort regression but has a statistically significant positive sign for the budget share measure.

Thirdly, we control for the effect on current education spending of the lagged level of education by including the human capital index (hci) variable lagged one period. Results are reported in column III in tables A12 and A13. For the most part, the general patterns noted previously are maintained.

Fourthly, in addition to the benchmark median or 0.5 quantile regression (estimates at the 50th percentile for the sample), estimates are also produced for other quantiles (20th, 25th, 40th, 60th, 75th, and 80th percentiles). Results are reported in columns IV to X in tables A12 and A13. Overall, the general patterns of predominantly positive (negative) association between the level of economic development and national effort (budget share) in education spending are maintained.


4 Discussion


From the perspective of the 2 × 2 categorisation in equation (2), richer (developed) countries tend to make a greater national effort with respect to education (they spend more on average on education as a share of GDP). In contrast, they tend to have lower budget shares (they spend less on average on education as a share of total government spending) relative to poorer (less-developed) countries.

In terms of national effort, richer country governments do not necessarily value education more highly than poorer country governments, but they have greater capacity to generate income from taxes. They can raise more income from taxes because they have larger formal private-sector economies. They are therefore less fiscally constrained and can spend more on areas such as education. The inability of poorer-country governments to extract revenue from a relatively small tax base constrains not only the growth of these countries’ public sectors – a point noted by Holcombe (2005), albeit in more general terms – but also their national effort with respect to education. Poorer countries tend to have greater informal-sector, cashbased economic activity relative to the size of the formal private-sector economy (Schneider and Enste, 2000), which makes it more difficult for governments in such countries to extract the tax revenue necessary to finance public education.

From a budget share perspective, poorer countries tend to spend more on education as a share of total government spending because they generally have smaller public sectors, which means education tends to comprise a larger share of the total public sector budget. However, richer countries are more likely to have large, complex public sectors with a greater variety of fiscal components to be financed from tax revenue. For example, a larger role of the state in providing various kinds of welfare support in richer countries could lead to other forms of public spending, such as education, being assigned a lower priority. An implication of this reasoning is that publicly provided education, as a whole, might take on the characteristics of a necessity with respect to public-sector spending in richer countries. Consequently, from a fiscal varieties perspective, education’s share of the total “fiscal pie” tends to be smaller in richer countries with larger public sectors and a greater variety of fiscal components to be paid for from the public purse, explaining why the budget share allocation to education spending is lower (higher) in richer (poorer) countries. There is also a political dimension to this explanation. The priorities for education spending differ among poorer countries with contrasting levels (or states) of democracy. Political pressures compel governments in poorer, democratic countries to spend more on areas such as education, and when poorer democratic countries grow, they can more easily generate income from taxes to satisfy political pressures to spend more on education.

For comparable levels of economic development, democratic governments tend to spend more on education. On the other hand, our empirical results for the robust estimator with controls (table 3 and table A11) show that poorer, non-democratic countries have low budget shares that are not necessarily much different from those of richer (democratic) countries. This suggests that the former not only have smaller public sectors, but also have lower allocations to education from the public purse. This might partly explain why such countries remain poor and less developed.

The observation that richer (developed) countries, on average, tend to spend more on education as a share of GDP and less on education as a share of total government spending than poorer (less-developed) countries, implies that richer countries on average have larger public sectors (total government spending as a share of GDP) than poorer countries. This follows from the identity (E/Y)/(E/G) ≡ G/Y, where E is public education spending; Y is GDP and G is total government spending. If the national effort, E/Y, and budget share, E/G, are both larger for richer than for poorer countries, then differences in the size of government, G/Y, between richer and poorer countries will depend on the relative size of the increases. However, if, as our results suggest, richer countries have a larger national effort, but smaller budget share than poorer countries, then the identity necessarily implies that richer countries have larger public sectors than poorer countries.

Table 5
Three inequality propositions
DISPLAY Table

Table 5 summarises the key empirical findings in this study in the form of three inequality propositions representing the characteristics of richer compared to poorer countries. To the best of our knowledge, such a characterisation of education spending (Propositions 1 and 2) and, by implication, the size of the public sector (Proposition 3) has not been presented in this form before. Because the inequalities in Propositions 1 and 2, based on our empirical results, are different for national effort compared to budget share, they imply that richer (poorer) countries have larger (smaller) public sectors.7 Proposition 3 logically follows as a consequence of Propositions 1 and 2; however, if empirical analysis of education spending had revealed the same direction of association for both measures, then Proposition 3 would not necessarily result. The same could be said for any other national effort or budget share measure of fiscal expenditure. Therefore, our analysis provides a novel way to characterise differences in the size of government at different levels of income.


5 Conclusion


We examine whether there are mean differences in the levels of public spending on education for two widely used national-level measures (national effort and budget share) for different economic (or regional) and political groupings of countries. Controlling for the state of democracy, we find that richer (poorer) countries tend to spend, on average, a larger (smaller) share of GDP on education, but a smaller (larger) share of total government spending on education. Richer countries, on average, make a greater national educational effort, whereas poorer countries allocate a greater budget share to education. By implication, richer countries, on average, have larger public sectors than poorer countries. In addition, for comparable levels of income, democratic countries tend to spend more on education than is the case for their non-democratic counterparts.

The findings with respect to levels of income can be summarised in the form of three inequality propositions. Examination of education spending patterns with respect to the national effort and budget share measures provides indirect support for a positive association between the size of government and income, consistent with Wagner’s law and Baumol’s “cost disease” hypothesis. Peacock and Scott (2000) note that different components of government expenditure might grow at different rates. Therefore, from the perspective of public policy analysis, future research might focus on testing the inequality propositions identified in this study with respect to other components of the government’s budget allocation (for example, the national effort and budget share of health, military, or welfare spending).


Appendix


Table A1
List of countries and territories by GNI per capita group in 2015 (ypc2015)
DISPLAY Table
Table A2
List of countries by two regional country groups (region)
DISPLAY Table
Table A3
Descriptive statistics
DISPLAY Table
Table A4
Mean differences in the national effort by income group and regime type – alternative standard error estimates
DISPLAY Table
Table A5
Mean differences in the budget share by income group and regime type – alternative standard error estimates
DISPLAY Table
Table A6
Mean differences in the national effort by country region and regime type – alternative standard error estimates
DISPLAY Table
Table A7
Mean differences in the budget share by country region and regime type – alternative standard error estimates
DISPLAY Table
Table A8
A summary of changes to the model specification (national effort and LSDV estimator)
DISPLAY Table
Table A9
A summary of changes to the model specification (budget share and LSDV estimator)
DISPLAY Table
Table A10
A summary of changes to the model specification (national effort and robust estimator)
DISPLAY Table
Table A11
A summary of changes to the model specification (budget share and robust estimator)
DISPLAY Table
Table A12
A summary of additional robustness checks (national effort with robust and quantile estimators)
DISPLAY Table
Table A13
A summary of additional robustness checks (budget share with robust and quantile estimators)
DISPLAY Table


Funding


Mark Millin wishes to acknowledge financial support in the form of doctoral scholarship funding from the National Research Foundation (NRF) of South Africa and the University of Otago, and a publication bursary from the University of Otago.

Notes


* We thank two anonymous reviewers, and participants at the 59th Annual Conference of the New Zealand Association of Economists and at the University of Otago, Department of Economics, Brown Bag seminar for helpful comments and suggestions.

1 More detailed explanations of the compensation and efficiency hypotheses are provided by Walter (2010), Adserà and Boix (2002), Garrett (1998a, 1998b, 2001), Rodrik ( 1998), Katzenstein (1985), Ruggie (1982) and Cameron (1978).

2 Robust estimation uses the “rreg” routine in Stata. An initial screening based on Cook’s distance is used to remove gross outliers. Starting values are then calculated, and Huber iterations performed, followed by biweight iterations, to determine the down-weighting of any outliers; see Hamilton (1991) for further details.

3 Baum, Nichols and Schaffer (2010) and Cameron and Miller (2015) provide a practical discussion of clusterrobust inference. All estimates are obtained using Stata; one-way clustering of standard errors is performed using “cluster(country)” or “cluster(year)”. Two-way clustering is performed with the user-written program “vce2way” (Yoo, 2017).

4 We are grateful to a reviewer for suggesting these additional robustness checks.

5 Partial confounding refers to the case where only poorer countries that are not democratic are shown to have significantly different means from the base group (richer and democratic countries), and with the expected sign. No confounding refers to the case where either poorer country groups (irrespective of the state of democracy) or poorer and democratic countries are shown to have significantly different means from the base group, and with the expected sign.

6 Alternative specifications were also fitted using GDP per capita and its squared and cubed values, along with the political democracy variable and main controls. In all cases considered, the main results are supported, i.e., the coefficient on the linear GDP per capita term maintains the same sign, is not too dissimilar in size, and remains statistically significant. Note that, in the main results, non-linearities are allowed for by estimating piecewise linear effects, i.e., average effects for different income groupings of countries.

7 We note two points relating to these inequalities. First, it does not matter whether E, Y and G are measured in real or nominal terms, provided both the numerator and denominator of the relevant ratio are measured in the same nominal or real terms (using the same deflator). Second, the same estimated size of the public sector in any one country, as given by sources such as the IMF, cannot simply be obtained by taking the quotient of the national effort and budget share for that country because these education spending measures are estimates. The quotient will give only a rough approximation of the size of government, especially for countries that have less accurate education spending data.

Disclosure statement


The authors have no potential conflict of interest to report.

References


  1. Adserà, A. and Boix, C., 2002. Trade, democracy, and the size of the public sector: The political underpinnings of openness. International Organization, 56(2), pp. 229-262 [CrossRef]

  2. Afonso, A. and Alves, J., 2017. Reconsidering Wagner’s Law: Evidence from the functions of the government. Applied Economics Letters, 24(5), pp. 346-350 [CrossRef]

  3. Akanbi, O. A. and Schoeman, N. J., 2010. The determinants of public expenditure and expenditure on education in particular in a selection of African countries. South African Journal of Economic and Management Sciences, 13(1), pp. 50-61 [CrossRef]

  4. Angelov, A. G., 2019. Public expenditure on education in the EU member states: A cluster analysis. Economic Archive, 72(1), pp. 52-64.

  5. Avelino, G., Brown, D. S. and Hunter, W., 2005. The effects of capital mobility, trade openness, and democracy on social spending in Latin America, 1980-1999. American Journal of Political Science, 49(3), pp. 625-641 [CrossRef]

  6. Baqir, R., 2002. Social sector spending in a panel of countries. IMF Working Paper, No. 02/35 [CrossRef]

  7. Baum, C. F., Nichols, A. and Schaffer, M. E., 2010. Evaluating one-way and two-way cluster-robust covariance matrix estimates. BOS’10 Stata Conference, July 2010.

  8. Baum, M. A. and Lake, D. A., 2003. The political economy of growth: Democracy and human capital. American Journal of Political Science, 47(2), pp. 333-347 [CrossRef]

  9. Baumol, W. J. and Bowen, W. G., 1966. Performing arts – The economic dilemma. A study of problems common to theater, opera, music and dance. New York, NY: The Twentieth Century Fund.

  10. Baumol, W. J., 1967. Macroeconomics of unbalanced growth: The anatomy of urban crisis. American Economic Review, 57(3), pp. 415-426.

  11. Brown, D. S. and Hunter, W., 1999. Democracy and social spending in Latin America, 1980-92. American Political Science Review, 93(4), pp. 779-790 [CrossRef]

  12. Brown, D. S. and Hunter, W., 2004. Democracy and human capital formation: Education spending in Latin America, 1980 to 1997. Comparative Political Studies, 37(7), pp. 842-864 [CrossRef]

  13. Bursztyn, L., 2016. Poverty and the political economy of public education spending: Evidence from Brazil. Journal of the European Economic Association, 14(5), pp. 1101-1128 [CrossRef]

  14. Busemeyer, M. R., 2007. Determinants of public education spending in 21 OECD democracies, 1980-2001. Journal of European Public Policy, 14(4), pp. 582-610 [CrossRef]

  15. Busemeyer, M. R., 2008. The impact of fiscal decentralisation on education and other types of spending. Swiss Political Science Review, 14(3), pp. 451-481  [CrossRef]

  16. Busemeyer, M. R., 2009. Social democrats and the new partisan politics of public investment in education. Journal of European Public Policy, 16(1), pp. 107-126 [CrossRef]

  17. Cameron, A. C. and Miller, D. L., 2015. A practitioner’s guide to cluster-robust inference. Journal of Human Resources, 50(2), pp. 317-372 [CrossRef]

  18. Cameron, D. R., 1978. The expansion of the public economy: A comparative analysis. American Political Science Review, 72(4), pp. 1243-1261 [CrossRef]

  19. Castles, F. G., 1989. Explaining public education expenditure in OECD nations. European Journal of Political Research, 17(4), pp. 431-448 [CrossRef]

  20. Cockx, L. and Francken, N., 2016. Natural resources: A curse on education spending? Energy Policy, 92, pp. 394-408 [CrossRef]

  21. Dragomirescu-Gaina, C., 2015. An empirical inquiry into the determinants of public education spending in Europe. IZA Journal of European Labor Studies, 4(25), pp. 1-24 [CrossRef]

  22. Driscoll, J. C. and Kraay, A. C., 1998. Consistent covariance matrix estimation with spatially dependent panel data. Review of Economics and Statistics, 80(4), pp. 549-560 [CrossRef]

  23. Feenstra, R. C., Inklaar, R. and Timmer, M. P., 2015. The next generation of the Penn World Table. American Economic Review, 105(10), pp. 3150-3182 [CrossRef]

  24. Fosu, A. K., 2010. The external debt-servicing constraint and public-expenditure composition in sub-Saharan Africa. African Development Review, 22(3), pp. 378-393 [CrossRef]

  25. Freedom House, 2016. Freedom in the world survey data.

  26. Garrett, G., 1998a. Global markets and national politics: Collision course or virtuous circle? International Organization, 52(4), pp. 787-824 [CrossRef]

  27. Garrett, G., 1998b. Partisan politics in the global economy. New York, NY: Cambridge University Press.

  28. Garrett, G., 2001. Globalization and government spending around the world. Studies in Comparative International Development, 35(4), pp. 3-29 [CrossRef]

  29. Garritzmann, J. L. and Seng, K., 2016. Party politics and education spending: Challenging some common wisdom. Journal of European Public Policy, 23(4), pp. 510-530 [CrossRef]

  30. Hamilton, L. C., 1991. How robust is robust regression? Stata Technical Bulletin, 2, pp. 21-26.

  31. Harding, R. and Stasavage, D., 2014. What democracy does (and doesn’t do) for basic services: School fees, school inputs, and African elections. Journal of Politics, 76(1), pp. 229-245 [CrossRef]

  32. Holcombe, R. G., 2005. Government growth in the twenty-first century. Public Choice, 124(1-2), pp. 95-114 [CrossRef]

  33. Huber, E., Mustillo, T. and Stephens, J. D., 2008. Politics and social spending in Latin America. Journal of Politics, 70(2), pp. 420-436 [CrossRef]

  34. Iversen, T. and Stephens, J. D., 2008. Partisan politics, the welfare state, and three worlds of human capital formation. Comparative Political Studies, 41(4/5), pp. 600-637 [CrossRef]

  35. Katzenstein, P. J., 1985. Small states in world markets. Ithaca, NY: Cornell University Press.

  36. Kaufman, R. R. and Segura-Ubiergo, A., 2001. Globalization, domestic politics, and social spending in Latin America: A time-series cross-section analysis, 1973-97. World Politics, 53(4), pp. 553-587 [CrossRef]

  37. Kuckuck, J., 2014. Testing Wagner’s law at different stages of economic development. FinanzArchiv/Public Finance Analysis, 70(1), pp. 128-168 [CrossRef]

  38. Lake, D. A. and Baum, M. A., 2001. The invisible hand of democracy: Political control and the provision of public services. Comparative Political Studies, 34(6), pp. 587-621 [CrossRef]

  39. Murshed, S. M., Bergougui, B., Badiuzzaman, M. and Pulok, M. H., 2022. Fiscal capacity, democratic institutions and social welfare outcomes in developing countries. Defence and Peace Economics, 33(3), pp. 280-305 [CrossRef]

  40. Newey, W. K. and West, K. D., 1987. A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55(3), pp. 703-708.

  41. Newey, W. K. and West, K. D., 1994. Automatic lag selection in covariance matrix estimation. Review of Economic Studies, 61(4), pp. 631-653 [CrossRef]

  42. Ozkok, Z., 2017. Financing education in Europe: The globalization perspective. Economics & Politics, 29(1), pp. 69-90 [CrossRef]

  43. Peacock, A. and Scott, A., 2000. The curious attraction of Wagner's law. Public Choice, 102(1-2), pp. 1-17 [CrossRef]

  44. Ram, R., 1995. Public educational expenditures in the United States: An analytical comparison with other industrialized countries. Economics of Education Review, 14(1), pp. 53-61 [CrossRef]

  45. Rodrik, D., 1998. Why do more open economies have bigger governments? Journal of Political Economy, 106(5), pp. 997-1032 [CrossRef]

  46. Ruggie, J. G., 1982. International regimes, transactions, and change: Embedded liberalism in the postwar economic order. International Organization, 36(2), pp. 379-415 [CrossRef]

  47. Schneider, F. and Enste, D. H., 2000. Shadow economies: Size, causes, and consequences. Journal of Economic Literature, 38(1), pp. 77-114 [CrossRef]


  48. Schultz, T. P., 1988. Expansion of public school expenditures and enrollments: Intercountry evidence on the effects of income, prices, and population growth. Economics of Education Review, 7(2), pp. 167-183 [CrossRef]

  49. Shin, D., 2020. The military in politics and democracy: Its impact on government spending for education and health. Social Science Quarterly, 101(5), pp. 1810-1826 [CrossRef]

  50. Stasavage, D., 2005. Democracy and education spending in Africa. American Journal of Political Science, 49(2), 343-358 [CrossRef]

  51. Tilak, J. B. G., 1989. The recession and public investment in education in Latin America. Journal of Interamerican Studies and World Affairs, 31(1-2), pp. 125-146 [CrossRef]

  52. Tilak, J. B. G., 1990. External debt and public investment in education in Sub-Saharan Africa. Journal of Education Finance, 15(4), pp. 470-486.

  53. Wagner, A., 1892. Grundlegung der politischen ökonomie (3rd ed.). Leipzig: Winter.

  54. Wagner, A., 1958. Three extracts on public finance. In: R. A. Musgrave and A. T. Peacock, eds. Classics in the theory of public finance. London: Palgrave Macmillan, pp. 1-15.

  55. Walter, S., 2010. Globalization and the welfare state: Testing the microfoundations of the compensation hypothesis. International Studies Quarterly, 54(2), pp. 403-426 [CrossRef]

  56. World Bank, 2017a. Education statistics: Education expenditure. Washington: The World Bank.

  57. World Bank, 2017b. TCdata360 database. Washington: The World Bank.

  58. World Bank, 2017c. World development indicators. Washington: The World Bank.
  September, 2023
III/2023
In order to give you a better user experience, cookies have been stored on your computer.
Accept cookie     More information