How robust are empirical results to sample selection? First, results definitely depend on whether all sectors or just tradables are analyzed. Looking at all sectors together, BC and HB find an increase in the price of skill-intensive nontradables during the 1980s. In light of SS versions that do not directly involve international trade, this suggests that one cause of rising inequality was rising relative prices for skill-intensive nontraded products. For the subset of just tradable manufactures, LS, Leamer, and BC all find no strong trend in relative prices during the 1980s. This finding suggests that trade did not contribute to rising inequality. Clearly, in comparing studies one must be careful to identify differences driven by sample selection of tradables versus nontradables.
Conditional on selecting the overall set of industries to analyze, empirical results also depend on whether data on all available industries is included. In some cases results have been somewhat robust to sample selection. For example, both LS and BC use import and export prices for manufacturing industries despite the fact that these data do not exist for every single manufacturing industry. Import prices exist for only 18 of 20 two-digit SIC industries and only about 50 of 143 three-digit SIC industries. Despite this, both LS and BC find no strong trend in relative prices during the 1980s for these smaller samples, matching the results of Leamer and BC using the full set of 450 four-digit manufacturing industries.
But in other cases the issue of inadequate sample selection appears to be very crucial. For example, Krueger’s analysis of the 1990s uses only 150 of the 450 four-digit SIC manufacturing industries. He acknowledges that his sample of finished-processor industries is incomplete and comments that in “a later draft of this paper, I hope to obtain data for non-finished goods industries” (fn. 8, p. 8). It seems reasonable to wonder whether his analysis is representative of manufacturing overall during this period.