capture how a country’s income level is related to the structure of education expenditures; and
(iv) a measure of the government’s ideology in order to control for systematic partisan biases
in education expenditures. The ideology variable is an index that is 1 when the government
is left-wing with respect to economic policy, and else 0.10
One important variable for which do not explicitly control is the amount private education
expenditures. Unfortunately, private education expenditures are dicult to measure for the
poorer countries in our sample. However, the propensity of a country to nance education
privately can be understood as a relatively time-constant country-specic characteristic. For
example, private education expenditures have traditionally been and continue to be more
important in Anglo-Saxon than in Continental European countries. We therefore use the
country xed eects to control for the importance of private education expenditures.
Apart from these control variables, we use several additional variables either as controls in
the robustness checks, or as instruments, or both. These are, rst, the gross enrollment rates
for primary, secondary, or tertiary educational programs and, second, the population shares
of the age groups relevant for each of the three types of educational programs. In addition,
we also consider the amount of development aid paid to certain countries. We explain the
rationale for using these variables in the respective sections; they are concisely described in
table 1.
Our unbalanced dataset covers 86 countries, both developing and developed, over the 1999-
2006 period. Summary statistics for all variables used in the subsequent regressions are
collected in table 5; a list of the countries that are considered in this study can be found in
table 7. Both tables are in the appendix.
Inter alia, the summary statistics in table 5 reveal that the average expenditures per student
as percent of GDP per capita are much higher for tertiary (64.48 %) than for either primary
(16.29%) or secondary programs (22.73%). The same holds with respect to the variability in
the three expenditure types as revealed by the standard deviations.
3.2 Baseline models
We collect the baseline results in table 2. We estimate a set of four models for each of
the dierent types of education expenditures. All models include the full set of economic
and political control variables, and country and year xed eects (results for these are not
reported). For hypothesis tests, we always use robust standard errors. At the bottom of the
table (in the row labeled Glob. at Dev=1 ), we report the marginal eect of globalization in
developing countries (i. e., the estimate of δ + 7) for those models in which we include the
interaction between globalization and the developing countries dummy.
10Note that this ideology variable is derived from the DPI dataset. Whereas this dataset distinguishes
between right, center, left, and other governments, we use, for compactness, a 0-1 classication. We code
left-wing governments as 1 and all other governments as 0.
10