In columns (4) to (9) of Table 8 we add to our model educational expenditures per pupil in
primary schools as a percentage of GDP and pupil-teacher ratios in primary schools from the
World Bank education database.24 The effect of educational expenditures and pupil-teacher
ratio are insignificant in all but one specification. Most important, the effects of government
consumption, social expenditures, and the progressivity of the income tax system remain
qualitatively unchanged when these measures of resource use and school quality in primary
education are accounted for. Overall, we find no indication that the generosity of the welfare
system and government public goods’ creation proxies previously unobserved educational
expenses or school quality.
Lastly, we investigate whether the choice of functional form of the model is important. One
may argue that it is not short-term fluctuations in the independent variables that are important,
but the development in the medium or long term. We have carried out identical regressions as
reported in Tables 4-6 using 5-year moving averages of the independent variable in place of
current values. The findings for government size in Table 4 appear partly sensitive to the
choice of time window, although a robust and large performance lowering effect at the 1
percent level remains if country-specific time trends are included. The effect of social
expenditures analyzed for the OECD countries (analogously to Table 5) appears insignificant
throughout, albeit their coefficients prevail in size and direction. In contrast, the negative
association between active labor market policies spending and pension benefit spending is
strongly corroborated.25 Estimation of 5-year moving averages corroborates the results for
progressivity of the tax system for OECD countries (analogously to Table 6), while the
coefficients for the whole country sample are now smaller and insignificant, albeit all with
negative signs. We have also investigated whether the results for government spending are
sensitive to using the log of the spending shares. The analogous results for Table 4 are similar
and show, again, the importance of country-specific time trends to identify the effect of
government size in the world sample. In contrast, the coefficients for social spending in the
OECD become now insignificant, suggesting a model misspecification. Results for social
spending components are, again, comparable to the original Table 5, while we now also
observe family allowances to be significantly and positively contributing.
24 For secondary education, the number of observations was insufficient. The inclusion of interpolated values lets
the samples in Table 8 increase by 19-31 percent.
25 Significant at least at the 5 percent level. In addition, housing subsidies appear now conducive to student
performance, (at the 5 percent level) in a similar manner as family allowances in the current value model.
25