relationships (although the latter finding is not robust across specifications). There is a statistically
weak but positive link between trade union membership and public sector employment and
protectionism, while the self-employed are more likely to be free traders, as are those who can be
identified as right wing on the basis of the political party they support.20 The last three variables
measure familiarity with, and attitudes towards, relevant regional associations (e.g. the European
Union in Ireland’s case). There was a negative but statistically insignificant relationship between
familiarity with the association and protectionism; but those who felt that their country did not benefit
from the association, as well as those who felt that their country should be protecting its
independence, were more likely to be protectionist.21
Finally, we experimented with our specification by including country GDP per capita, with
and without country dummies.22 With country dummies omitted, richer countries tend to be associated
with a higher preference for free trade, but this relationship goes away (and indeed the sign of the
coefficient is reversed) when country dummies are introduced. More importantly for our purposes,
the sign of the interaction term between high skills and GDP per capita remains negative and
significant in both these specifications. Another noteworthy feature of these regressions is that when
country dummies are excluded, the coefficient on Roman Catholicism become much larger and more
statistically significant, suggesting that variation in religious beliefs might be explaining cross-
country differences in protectionist attitudes, as well as within-country differences. Roman Catholic
social teaching has traditionally been more skeptical of the market than some brands of Protestantism,
so these correlations make intuitive sense; on the other hand, Catholicism could be proxying across
our set of countries forthe Mediterranean region, which might have more protectionist attitudes for
reasons unrelated to religion.
20 Data on these variables are not available for all countries.
21 Again, these regional integration questions were only asked in a subset of our countries.
22 An additional country dummy is of course dropped in the specification involving both country
dummies and GDP per capita.
16