The Role of Selection Out of the Labor Market in Estimates of Racial Wage Differences

After accounting for selection and including these two important omitted variables, we fi nd no evidence of black wage premiums for women. We then test the data further by examining three settings in which we would be most likely to observe a black- wage premium, if it exists. First, in the previous literature, the most consistent evidence of a black wage premium is among highly educated women for example, Fisher and Houseworth 2012; Black et al. 2008. We estimate wage gaps separately by education level and fi nd no evidence of black wage premiums, even among highly educated women. Second, we add controls for occupation to wage regressions. To the extent that occupational sorting differs by race due to discrimination against black women, wage gaps conditional on occupation understate discrimination against black women Blau and Ferber 1987. Equivalently, black women’s conditional wages should be high relative to white women’s wages, but our estimates that control for occupation again show little signifi cant difference between black and white women’s conditional wages. Third, researchers typically control for age or potential experience, but workers with the same education and age may have very different work experience histories, which imply very different labor market productivities. Because, on average, white women have more years of work experience than black women, we expect blacks will appear to perform even better, relative to whites, when we control for actual labor market experience. However, when we control for actual labor market experience, we fi nd no statistically signifi cant wage premium. These fi ndings provide further evidence of no wage premium for black women.

II. Related Literature

A. The Role of Selection Out of the Labor Market in Estimates of Racial Wage Differences

Fryer 2011 cites selection out of the labor force as a potential explanation for the esti- mated black wage premium among women page 859. Wage gaps that are computed from observed wages by defi nition exclude nonworkers, yet nonworkers might be an important group to consider in computing black- white wage differentials, especially if there are racial differences in patterns of work. Failing to account for selection out of work would result in blacks appearing to perform better than whites if black women with low potential wages are more likely to select out of work than white women and if white women with high potential wages choose not to work more frequently than black women. 3 Empirical evidence tends to confi rm these expectations see, for ex- ample, Neal 2004; and for a similar approach to accounting for selection out of work by males, see Johnson, Kitamura, and Neal 2000; Chandra 1999; Chandra 2003. Neal 2004 shows the importance of accounting for both types of selection out of work when estimating the wage gap for women. With 1990 data, he imputes a low po- tential wage for low- educated women who did not work, received no fi nancial support from a spouse, but received Aid to Families with Dependent Children, Supplemental Security Income SSI, or Food Stamps between 1988 and 1992. Considering selec- 3. Neal and Johnson 1996 hypothesize that highly skilled black women may have less unearned income than highly skilled white women for example, their husbands and fi nancial investments earn less, making highly skilled black women more likely to work. tion out of work for nonworkers with low potential wages increases the wage gap between 2.8 and 3.7 percentage points. Some women choose not to work even with a high potential wage. 4 When Neal 2004 also considers selection out of work by this group, the gap increases another 1 to 1.2 percentage points. Following Neal 2004, we account for differential selection out of the labor force for women with high and low potential wages.

B. The Role of Cost of Living in Estimates of Racial Wage Differences