Having surveyed each of the nine product-price studies individually, I now try to synthesize their similarities and differences. I do this in three steps. First, I discuss how research has refined a set empirical strategies for applying the SS theorem to the data. Second, I comment on some important methodological issues regarding the robustness of results. “The facts” about product prices and their warranted wage changes are relatively sensitive to the selection and weighting of industries sampled and to the decade considered. In contrast, “the facts” are relatively insensitive to the extent of data aggregation and the measurement of skills. Third, in light of these methodological issues I summarize what “the facts” seem to be about product-price changes and their mandated wage changes.
A Methodological Progression
Here is a summary of the various empirical methodologies surveyed in the previous section. Bhagwati (1991): Discussion of U.S. terms of trade Lawrence and Slaughter (1993): Regression analysis (LS) Pj^sos = a+ b (NPW/PW)j1980 + ej Sachs and Shatz (1994): Regression analysis (SS) Pj*1980s = a+ b (PW/(PW+NPW))j1980 + bc(Dcomputers) + ej Feenstra and Hanson (1995): Descriptive comparison of domestic and import prices Leamer (1996): Discussion of industry relative prices plus regression analysis (L-T) [(1 – l) x TFPj*] = (0 ij)bit+ ej (L-G) [Pj* + (l x TFPj*)] = (0 ij)big + ej Baldwin & Cain (1997): Discussion of industry relative prices plus regression analysis (BC) Pj* = a + (0 ij)bi + ej Krueger (1997): Regression analysis (K-1) Pj*1990s = a+ b (PW/(PW+NPW))j + bc(Dcomputers) + ej (K-2) Pj* = a + (0ij)bi + ej Feenstra and Hanson (1997): Regression analysis (FH) “Outsourcing” TFPj* = (0 ij)bi+ ej Harrigan and Balaban (1997): Regression analysis of cost-share equations from translog representation of U.S. revenue function How do these different approaches relate to each other? By listing them chronologically, I think one can identify a progressive refinement of how to apply the SS theorem in the data.