world sample. In columns (7) and (8), however the standard errors are relatively large.
Nevertheless, the variation that drives these results appears to differ between OECD countries
and non-OECD countries. While the variation across country-specific trends aids identifying a
strong effect for the whole sample, inclusion of trends in the OECD sample does not influence
the effect of government size.
In sum, we identify negative correlations between government’s public goods creation,
measured by its consumption spending, and student achievement. This evidence is in
accordance with our hypothesis that a more generous welfare system generates disincentives
for educational investment. The result indicates that when government consumption spending
as a share of GDP increases by 0.1 log-points, student achievement is reduced by about 0.1
“core country” standard deviations.
5.2. Social expenditures
Table 5 presents results for government social expenditures, measured as share of GDP, in 29
OECD countries, resulting in a sample of 124 observations. Column (1) shows that the
unconditional correlation between welfare transfers to households and student achievement is
negative and significant at 5 percent level. Inclusion of co-variates even increases the effect of
social expenditures both in terms of magnitude and statistical significance (column (2)).17
Also within OECD countries, there are positive impacts of GDP and adult education
attainment. Column (3) of Table 5 adds country fixed effects. The effect of social
expenditures is still significant at 5 percent level. The magnitude of the coefficient is larger
than those when country fixed effects are excluded, but of similar size compared to when
government consumption spending in Table 4 is employed. Column (4) shows that the results
are not sensitive to whether the model includes time-specific fixed effects or a time trend
common to all countries.
However, when including country-specific time trends, the effect of social expenditures
completely disappears (column (5) of Table 5). We conclude that it is country-specific trends
that drive the results, which indicates that in the OECD sample there are some systematic
medium-term changes in government policy that students and parents react on.
In the last part of Table 5 we distinguish between different components of social expenditures
relating to specific social policy areas such as health, family care, labor market, pension
system, etc. In column (6) we replace total social expenditures with all its various
components. All components have a negative sign as expected, except ‘health care spending’,
‘other spending’, and ‘family allowances’. Notably, the spending category ‘other spending’ is
17 The sample is smaller in column (2) than in column (1) of Table 5 because adult educational attainment is not
available for Luxembourg.
18