Robustness to Data Aggregation
The SS theory is largely silent on this point. In theory different industries are distinguished by their different relative employment of factors as dictated by their different production technologies. Empirically, it is generally assumed that more-disaggregated data is better.
The results appear to be quite robust to data aggregation. Many studies of the manufacturing sector use four-digit SIC industries (Leamer, Krueger, FH). Studies using three-digit data (LS, SS, BC) and/or two-digit data (LS, BC) obtain qualitatively similar results to the more-disaggregated studies. For example, the finding of constant relative manufacturing prices during the 1980s is obtained at the four-digit level (Leamer), the three-digit level (LS and BC) and the two-digit level (LS and BC).
Robustness to Measurement of Skills
Trade theory is largely silent on this point of how to measure skills. It is generally accepted that the nonproduction-production classification for manufacturing workers suffers more misclassification of skills than a categorization based on education. However, this is claim is a statement about noisiness of data, not necessarily bias.
In fact, the nonproduction-production classification does not appear to be a biased measure of skills. Studies using this measure tend to obtain similar results to other measures such as educational attainment. Again, for the 1980s the conclusion of relatively stable relative product prices and thus little mandated change in wage inequality is obtained from studies using the nonproduction-production classification (LS, Leamer for the case of zero pass-through), using educational attainment to identify two labor groups (BC with two different cutoff points between the two groups), and using skills inferred form actual wages paid (Leamer). Similarly, for the 1990s for the sample of finished-processor industries Krueger obtains the same result for equation (K-1) using both the nonproduction-production classification and average years of education. The results for equation (K-2) are qualitatively the same for the nonproduction-productionclassification (reported in Table 1 of this paper) as for Krueger’s skill measure using minimum wages combined with total wage bills.