clear negative association appears between the two variables (the correlation coefficient is -0.634):
that is, the richer the country, the more negative (less positive) is the association between high skills
and protectionist attitudes. To the left of the picture, there is a cluster of poorer countries for which
the coefficient on high skills is either close to zero or positive, while to the right of the picture are the
rich countries in which high skills are clearly associated with a preference for free trade, rather than
protection.24
Again, we take Figure 1 to be strong evidence in favor of the Heckscher-Ohlin perspective.
Of course, it might be objected that high skills are only associated with protectionism in 3 countries,
and that the t-statistics fall below conventional levels in all three cases (see Table 6); but we think
that there is a convincing counter-argument (already suggested in Section 2), which, however, we are
unable to provide evidence for given our current dataset. Our sample of countries, while 20 times
bigger than the samples used in previous studies, only contains 20 countries. There are many, poorer
countries in which the high-skilled might be even more protectionist than in Bulgaria and Slovakia:
one could imagine the negatively sloped relationship in Figure 1 extending further to the left, with the
countries of the world as a whole more evenly divided between those where skills are associated with
protectionism, and those where skills are associated with liberal attitudes towards trade. Of course,
this is pure speculation on our part: nonetheless, the results we are able to obtain from these data
seem entirely consistent with the insights ofEli Heckscher and Bertil Ohlin.
The evidence in Figure 2, which plots the same coefficients against the average years of
schooling in each country,25 is less compelling. The correlation is much weaker (-0.0993), and this is
disturbing, since theory says that attitudes should be related to factor endowments, rather than income
per capita. (The rationale for using income per capita is thus that it is correlated with the skill level of
24 We also ran a regression over the entire sample which included country dummies, the high-skill
variable, and the high-skill variable interacted with country dummies (one country was of course
omitted). The interaction terms werejointly highly statistically significant (and several individual
interaction terms were also statistically significant), indicating that the skill coefficients are indeed
significantly different across countries.
25 Taken from Barro and Lee (2000).
18