EDUCATIONAL OUTCOMES IN OECD COUNTRIES
10
As a starting point for our analyses, we replicate the basic analysis, only replacing the
extended version of the Cohen and Soto (2007) data on years of schooling by the newly
available latest version of the Barro and Lee (2010) database on years of schooling. Our
sample contains the 24 OECD countries with available data. From the total of 30 OECD
countries, the sample misses four countries - the Czech Republic, Hungary, Poland, and
the Slovak Republic - because their communist history prevents them from having
internationally comparable economic data during the period of analysis in the underlying
database. In addition, Germany drops out because of missing economic and test score
data for the Eastern parts before 1990, and Luxembourg is left out as a small country
with a population of less than one million, as is common in growth analysis (see
Mankiw, Romer, and Weil (1992)).
Table 2 presents the basic results on the association between educational outcomes and
long-run economic growth in the sample of OECD countries. The inclusion of initial
GDP per capita in all specifications simply reflects the fact that it is easier to grow when
one is farther from the technology frontier, because one just must imitate others rather
than invent new things. (This “convergence term” enters into the subsequent projections
of economic impacts from school reform, because it suggests that any differences in
growth rates from changes in cognitive skills will eventually die out; see Section 5
below).
When the cognitive-skill data are ignored (column (1)), years of schooling in 1960 are
significantly associated with average annual growth rates in real GDP per capita in 1960-
2000, after controlling for the initial level of GDP per capita. However, once cognitive
skills are included in the model (column (2)), the whole explanatory power is taken over
by cognitive skills. Cognitive skills are highly significantly associated with economic
growth. At the same time, the association between years of schooling and economic
growth becomes statistically insignificant and drops to close to zero. Furthermore, the
OECD-sample growth variance explained by the model increases from 56 percent to 83
percent when measuring human capital by cognitive skills rather than years of schooling.
Note that in the OECD sample, the bivariate association with initial per-capita GDP
already accounts for 49 percent of the variance in subsequent growth, making the
relative increase in understanding non-convergence growth through cognitive skills
substantial.
The estimated coefficient on cognitive skills implies that an increase of one standard
deviation in educational achievement (i.e., 100 test-score points on the PISA scale)
yields an average annual growth rate over 40 years that is 1.86 percentage points higher.
This historical experience suggests a very powerful response to improvements in
educational outcomes, particularly when compared to the average 2.2 percent annual
growth within the OECD over the past two decades.
Figure 2 depicts the fundamental association graphically, plotting growth in real per-
capita GDP between 1960 and 2000 against average test scores after allowing for
differences in initial GDP per capita and initial average years of schooling. With the
slight exceptions of New Zealand (below the regression line) and the United States
(above) - to which we return below - the OECD countries align closely along the