Measuring the extent of earnings discrimination: an update

Ordinary least squares (OLS) regression and data from the March 1990 Current Population Survey are used to estimate yearly earnings averages for white males, white females, black males and black females. In order to test for the possible existence or gender and/or race discrimination, earnings estimates are decomposed, using the tradional Oaxaca (1973) and Blinder (1973) methodology. In comparing these results with those from Carlson and Swartz (1988, using 1979 data), we find that overall and unexplained residual differences have declined quite substantially, but that they still favour white males. We then incorporate an extension of the traditional methodology (Jackson and Lindley, 1989) which allows for testing for the significance of enexplained residuals, and for jointly testing for the significance of residual effect components, the constant and coefficient effects. These results call into question the unambiguous conclusion that earnings differentials uniformly favour white males. Strong support is found for the Jackson–Lindley contention that significant and positive residuals may have have offsetting signs. More specifically, where males appear to receive an entry level premium, as measured by constant effect differentials, but other groups appear to receive favourable treatment as measured by response rates to changes in independent variables (the coefficient effect). This result suggests that efforts may be underway to ‘make up’ for the entry level white male premium.