EDUCATIONAL OUTCOMES IN OECD COUNTRIES
16
run growth among rich countries. Instead, cognitive skills emerge as the one strong
policy factor underlying growth differences across OECD countries.
4. DIFFERENT LEVELS OF SKILLS AND EDUCATION
An important and recurring policy question is which level of skills and education is
most decisive for OECD growth. We analyze several dimensions of this: whether there
are differential returns across countries to average skills; whether basic skills are more or
less important than top levels of skills; and whether there is a specific role of tertiary
attainment for OECD growth.
4.1. Differential returns to average skills
As a first dimension of heterogeneity, we analyze whether the effect of average
educational outcomes differs along specific dimensions. First, the graphical plot of
Figure 2 suggests that there is no obvious non-linearity in the test score-growth
association across OECD countries. Thus, a squared test-score term does not enter the
model significantly, and an exponential test-score specification does not improve the fit
of the model (not shown).
Second, there is no obvious difference in the effect of average test scores between
countries with initially low vs. high income. While the small sample size does not allow
for extensive interaction models, an interaction of test scores with an indicator for above-
median initial GDP per capita does not enter the model significantly (not shown).
Third, Figure 3 plots quantile regression estimates in 5 percent steps of the effect of
average test scores for percentiles of the growth distribution. It is evident that the effect
is relatively constant across the whole distribution of growth residuals. In fact, all
quantile regression point estimates fall within standard confidence intervals around the
OLS estimate. The estimated average effect thus provides a good representation of the
effect across the full range of quantiles of the conditional growth distribution and is not
driven by specifics in certain parts of the conditional distribution.
The combined results of non-linear specifications, skill interactions, and quantile
regressions suggest no obvious patterns of differential returns to average cognitive skills in
the sample of OECD countries.
4.2. Basic vs. top skills
A leading policy question refers to the effects of different ranges of the skill
distribution. Should developed countries implement relatively egalitarian education
policies or rather relatively elitist policies? Should they focus on decent basic skills for
the whole population or on nurturing top scientists and engineers? To address such
questions, we make use of the micro data of each of the international achievement tests
to calculate the population shares in each country that reach a threshold of basic skills