Comments and Discussion

J. David Richardson: This paper is a treasure trove of fascinating patterns. It charts the correlates of answers to a question about whether a government "should limit the import of foreign products in order to protect its national economy" across a diverse array of individuals and countries of residence. Robust and memorable patterns from ordered probit regressions with elaborate and rich controls include the ways that:

--individual skills correlate negatively with protectionist sentiment, but the correlation weakens and even reverses as a country's average standard of living falls,

--extreme, but not more moderate nationalism, correlates positively with protectionist sentiment, and

--union status hardly correlates at all with protectionist sentiment.

But I have two broad complaints. The first is about the paper's repeated claim that its empirical patterns correspond to Heckscher-Ohlin theory. The second is about the authors' seeming omission of sensible econometric approaches to interpreting their patterns and perhaps to uncovering new ones.

I find the authors' attempt to root their specification and interpret their results using Heckscher-Ohlin trade theory to be intriguing, but not very compelling. They use the theory rhetorically--heuristically at best. There are two problems in how they argue. The more severe problem is that in a world with more than two inputs, Heckscher-Ohlin theory loses its sharp intuitive predictions about what happens to the returns to any single input when border liberalization occurs. It is no longer intuitively clear whether skilled or unskilled labor gains when other inputs (for example, physical capital) are in the background. A worker's stake in liberalization is determined by "friendship/rival" relations between the multiple inputs--and these are in turn determined by input substitution and complementarity relationships. ¹ The second and less severe problem is that there are alternative trade theories to Heckscher-Ohlin that would explain why skilled workers are less protectionist in richer countries. One recent example (associated with Marc Melitz) is the theory that border liberalization increases the implicit scope of input markets and amplifies the dispersion of the rewards to every input group's heterogeneous membership. A more traditional example is one in which Ricardian technology differences across countries are associated with hard-to-measure quality-and-productivity differences in inputs that go by the same name (with "skilled labor" in rich countries being more productive than in poor countries).

In a similar vein, I applaud the authors' wide spectrum of econometric approaches, and suspect indeed that their empirical results are robust. But I think they neglected some potentially illuminating variations. For one example, though they report separate regressions by country, gender, and labor-force participation, they do not report separate regressions by age group (old-young, say, or old-young women, then men). Since age is important as a freestanding variable in determining Eastern European import-protection preferences, it seems natural to me to ask whether the skill coefficients or the patriotism-chauvinism coefficients differ significantly between old and young. Alas, I can't tell. Or for a second example, the estimated properties of the disturbance term in such split-sample regressions are intrinsically interesting. The estimated error variance is a measure of how diffuse the unexplained preferences are--are western preferences more "inexplicable" than eastern? Alas, I can't tell. More generally, the authors' assumption about the stochastic properties of the disturbance term seems always at the natural extreme--the stochastic properties are always common across observations, both in the full sample and in the various split samples. However I would like to know the effect of maintaining that assumption, but allowing the coefficients to differ across sample splits, testing for significant differences between the entire regressions over the various subsamples using something like standard F tests for probit regressions. Do Canadians determine protectionist sentiment "differently" than the British? Alas, I can't tell.

But I can tell a lot more at the end of the paper than when I started. And...