Differences in returns to education: an analysis by race

American Journal of Economics and Sociology, The,  July, 1994  by Javed Ashraf

• E-mail
• Print
• Link

The inclusion of the selectivity variable in the wage equation represents a facet of the model rarely encountered in such studies. Since possibility exists that the data drew a sample of workers who were not a randomly drawn group of individuals from the population at large, not correcting for selectivity bias in the model would lead to inconsistent estimates of the coefficients in the wage equation. Heckman's correction for such bias has been widely used in the literature on union-nonunion wage differentials.(2) While such corrections have been extremely rare in studies examining racial discrimination,(3) the use of selectivity variables in the model used in this study is appropriate. These variables are defined as [-f([EMP.sub.i])/F([EMP.sub.i])] and f([EMP.sub.i])/{1 - F([EMP.sub.i])} where F and f are the cumulative and density functions of a standard normal variable. [EMP.sub.i] is the predicted employment status of an individual, obtained from probit estimates of the reduced-form equation determining employment status. Given the possibility of different factors not influencing the work decision of whites and blacks in the same manner, the selectivity variables were computed from separate probits for each race. Given the issue examined in this study, the second of these expressions namely f([Emp.sub.i])/[1 - F([Emp.sub.i])] is redundant, since it would pertain to individuals who were not working, and thus for whom market wages were not observed. Thus the variable used to correct for selectivity bias was [-f([Emp.sub.i])/F([Emp.sub.i])] in both the white and black samples.

IV

Empirical Results

TABLES 1A and 1B list coefficient estimates of the wage equation for white and black workers separately for five different years.(4) The SOUTH variable (representing workers residing in the south) exhibited an interesting trend. Although SOUTH was statistically significant throughout the twenty year period, the absolute value of the coefficient estimate dropped steadily over the period surveyed. Thus, while white workers in the south earned 16.47% less than those in other parts of the country in 1967, this differential had narrowed to only 4.88% by 1986.(5) A similar trend was observed in the sample of black workers, with workers in the south earning 21.33% less than their counterparts in the rest of the country in 1967, but only 3.92% less in 1986. White collar workers, both black and white, earned significantly more than their counterparts in blue-collar occupations, and the earnings gap remained quite stable over the twenty years.

Union membership was a strong determinant of earnings across both races during the entire survey period. However, the gains from collective bargaining declined for whites until the early 1980s at which point they appeared to have started moving up again. For blacks, the decline in gains from unionization persisted till the end of the survey period. It is interesting to note that the gains from unionization remained considerably higher for blacks than for whites during the entire period. This appears to be a testimony to the fact that unions are color-blind in their membership practices. The earnings of blacks and whites tend to be much closer to each other in unionized environments than in those where collective bargaining is not practiced. As a result, the gains to union membership can reasonably be expected to be larger for blacks than for whites, since nonunionized blacks had more "catching-up" to do.(6)