In this section we consider econometric issues that may cause us to mistakenly reject RFPEQ. Product misclassification can cause us to reject RFPEQ spuriously. To see why, suppose that industry i comprises two goods, j and k, and that good j is non-production worker intense relative to good k by a factor of n*r (e.g. stylish dress shirts versus ordinary dress shirts).15 In addition, to keep things simple, assume there are no factor quality differences either across regions or between factors within a region, so that all 0’s are equal to zero. If region r produces good j and region s produces good k, then instead of equations 11 and 10 we have
In words, while relative wages must still be equal in the two regions, relative intensities differ by n*r. As a result, when we compute relative wage bills for equation 12, we need to account for nir
Clearly, nir = 1 can cause us to reject RFPEQ even if yNP = 1. Indeed, ПгГ = 1 also can cause us to fail to reject RFPEQ even if yNP = 1. To minimize the problem of product misclassification, we use the most detailed industry data we can assemble by region.
As always, measurement error is of concern. However, our two specifications for testing RFPEQ minimize the problems with measurement error. With the US as the base in equation 26, under the null hypothesis measurement error in the regressor should be minimized. In equation 28 estimated over pairs of regions, we can check for problems due to measurement error by looking at the reverse regressions, i.e. switching base regions.
5. Empirical Results
In this section we describe our data set and present results from the tests of relative factor price equality.
The data cover the years 1972-1992 and come from the Longitudinal Research Database of the Bureau of the Census. We use data only from the Census of Manufactures which is conducted every fifth year on all manufacturing plants in the lower 48 states.17 We make use of the information on quantities of and total payments to two types of labor. We exclude plants that have non-positive value-added or any non-positive inputs. In addition, we exclude all Standard Industrial Classification (SIC) codes that represent miscellaneous products within an industry, i.e. SICs 39xx or xxx9, to reduce the possibility that we are comparing different goods within an industry. This leaves us with 385 of the original 458 4-digit SIC industries.