including each of the components separately. In all cases, the effects are negative, except for
family allowances. Columns (7) and (8) in Table 5 report the single two cases where the
separate effects are significant at least at 10 percent level. The effects of spending on active
labor market policies and pension payments are both performance lowering. Since the former
constitutes only a small part of total social expenditures, the negative effect of social
expenditures in columns (1) to (4) seem to a large part be driven by pension spending.18
Taken all together, we identify a negative effect of social spending in OECD countries for
student achievement, in support of our hypothesis. Among the different types of welfare
transfers, it is pension benefits that contribute most to this effect. This result is not surprising
given that in most developed countries pensions systems are highly redistributive in nature: on
the one hand, they guarantee an income-independent minimum rent to every contributor, on
the other hand, the place a cap on the maximum rent, equalizing rent incomes in the non-
active elderly population.
5.3. Tax progressivity
Table 6 uses the same models as Table 4, but replaces the government size variable with a 10-
point scale index of the top marginal income tax rate, adjusted for the income bracket, our
measure of progressivity of the income tax system. For this welfare state generosity measure,
there is a negative correlation with student achievement for the whole sample (column (1)),
but the effect disappears when we include the control variables (column (2)). However, in the
models with country fixed effects in columns (3) and (4), the effect is significant at five and
10 percent level, respectively. When the index increases by one standard deviation, which is
about 2.5 point on the index, student achievement is reduced by 0.21 adjusted “core” country
standard deviations.
How the effect of the variable of interest changes when we alter the model specification varies
greatly between the government size, social transfers and tax progressivity models, which
indicates that the variables have very different features.19 Nevertheless the main result for
both variables is that they tend to reduce student achievement. The quantitative effects of the
adjusted top tax rate are difficult to gauge as using an index variable makes quantitative
predictions difficult.
18 We are unable to exclude the possibility that more public expenditures on pension may equally proxy for a
large body of civil servants. In this case, the prospects of becoming a civil servant with high job security and
generous retirement options may equally lower effort in mandatory schooling.
19 The correlation coefficient between government consumption spending (log) and the top marginal tax rate
index is -0.44, and for the social spending (log) in OECD countries -0.30. Please note that financing of
government activities also occurs through corporate taxation and indirect taxes on e.g. consumption goods.
20