EDUCATIONAL OUTCOMES IN OECD COUNTRIES
11
regression line that depicts the positive association between cognitive skills and
economic growth.
Column (3) of Table 2 reports the same model excluding years of schooling, whose
effect could not be significantly differentiated from zero. The point estimate on
cognitive skills, as well as the adjusted R2, increases slightly in this reduced model.
Columns (4) and (5) break down the analysis into the 20-year sub-periods of 1960-
1980 and 1980-2000. The positive association of growth with cognitive skills is clearly
visible in both sub-periods, with the point estimate slightly larger in the later period.
To reduce concerns of reverse causality between economic growth and quantitative
schooling investments, the basic model uses school attainment in 1960 before the growth
period, but combines that with average test scores over the entire period. However, the
results are hardly affected by using average years of attainment across the period
(column (6)).
Because of the limited sample size, we want to ensure that results are not driven by
individual outliers. Columns (7) and (8) show the robustness of results to excluding
specific countries. Column (7) excludes Mexico and Turkey, two countries at the
bottom of the sample of OECD countries today in terms of measures of GDP per capita,
socioeconomic background, and educational spending. While the point estimate on
cognitive skills is slightly reduced, the association remains strong and statistically highly
significant. The same is true when excluding Korea (column (8)), a country with
extraordinary conditional test scores and growth experience (see Figure 2).
Finally, columns (9) and (10) provide tests for differences in the education-growth
nexus between OECD and non-OECD countries. The sample now is the full sample of
50 countries with data on test scores and economic growth. In the model without
cognitive skills (column (9)), the significant association between years of schooling and
economic growth does not differ significantly between OECD and non-OECD countries.
However, the OECD dummy is marginally significant at the 10-percent level and
positive, indicating a remaining growth advantage of OECD countries unexplained by
the model. But once cognitive skills are included in column (10), neither the OECD
dummy nor its interaction with cognitive skills are statistically significant, indicating that
the OECD countries actually fit well within the rest of the world on this association.10
We return to this issue when considering the separate impacts of economic institutions.
3. CAN INSTITUTIONS EXPLAIN DIFFERENCES IN RICH-COUNTRY
GROWTH?
To understand to what extent consideration of institutional differences across OECD
countries contributes to long-run growth differences and alters the assessment of the
importance of educational outcomes, this section focuses on fundamental institutional
measures of property rights and openness as well as measures of the regulation of
product and labor markets in our OECD-country growth models.
10 The OECD dummy and the interactions are also not significant if both interactions - with years of schooling and with
cognitive skills - are included together.