time. The index ranges from 0 to 10, with higher values representing a higher (income bracket
adjusted) top tax rate, and, thus, more redistribution of income. Between 1970 and 2000, data
have been collected every five years, and from 2000 on, annually. We have linearly
interpolated missing values to maximize the number of observations in our regressions.
GDP and population are taken from the WDI 2007 database, while adult education attainment
is taken from Barro and Lee (2001b). We use the percentage of the population over age 25
with secondary school attained. Up to 2000, the data are available every five year, and we
have linearly interpolated missing values. Since the last year with information is 2000, we lag
the variable by three years in order to include all the international tests in the analysis. In
some robustness analyses, we also use current data on primary school educational
expenditures (% GDP) and pupil-teacher ratios from the World Bank. Again, we linearly
interpolate the variables when there are missing values for at most five years.15
5. Empirical Results
5.1. Government consumption
Table 4 presents results for government consumption spending including up to 72 countries
and 232 observations. The first column simply presents the correlation between student
achievement and government size, measured by government consumption spending as share
of GDP. There is no unconditional correlation. Column (2) includes the control variables GDP
per capita, adult educational attainment w.r.t. secondary education, population size, and time-
specific fixed effects. The number of observations drops due to missing observations for adult
education for some countries. As expected, we find strong positive effects of GDP per capita
and adult education. The positive income effect mirrors Hanushek and Kimko (2000) who
report a positive effect of student achievement on economic growth, and is in accordance with
micro evidence on the effect of parental income on student performance. However, the result
for GDP is not in accordance with our theoretical model where the utility function is concave.
The positive effect of adult education mirrors results from previous micro-econometric
studies.
The conditional effect of government consumption appears large and is significant at the 1
percent level. Conditional on income, a small public sector is favorable, and conditional on
public sector size, students in rich and well-educated countries perform better than those
living in poor and low-educated countries. When the share of government consumption
15 From 1990 on, these data have been roughly collected on an annual basis.
15