This preview shows page 1. Sign up to view the full content.
Unformatted text preview: s of statistical discrimination,
that is, whether because Blacks expect lower labour market returns, they would invest
less in education. While they do not have data on expectations, they find mixed
results on the interactions between AFQT scores and the black dummy (positive for
men, negative for women) Neal and Johnson (1996) also incorporate a methodological innovation to deal with
potential selection bias into the labor market due to the lower participation rate of
Blacks, that is, they consider the wage offer curve.
o They focus on median (rather than mean) wage regression and assuming that those
who are not employed would have lower wage offers than the median offer of
those who are employed and are otherwise observationally equivalent.
o They assign log wages of zero (hourly wages of one cent) to all male
nonparticipants But note that the same applies to Hispanics and that O’Neill and O’Neill (2005)
report similar results (qualitatively) with the more recent data (when NLSY
respondents are older). Source: Neal and Johnson (1996)
BLACK-WHITE WAGE DIFFERENCES 8 75 TABLE 1 Black
*ge A FQT - .244 (.026)
(.014) ... NOTE.-The dependent variable is the log of hourly wages. T h e wage observations come from 1990 and 1991.
All wages are measured in 1991 dollars. If a person works in both years, the wage is measured as the average of
the two wage observations. Wage observations below $1.00 p er hour or above $75 a re eliminated from the data.
T h e sample consists of the NLSY cross-section sample plus the supplemental samples of blacks and Hispanics.
Respondents who did not take the ASVAB test are eliminated from the sample. Further, 163 respondents are
eliminated because the records document a problem with their test. All respondents were born after 1961. Standard
errors are in parentheses. skill-adjusted gaps among 24-year-old women.1° Second, it is possible
that selection effects contaminate the estimates of racial wage gaps
for women. For all women, the mean of observed wages likely overstates the mean of the wage offer distribution. If this selection effect
is most acute in the minority samples, the results in table 1 will understate the wage costs of racial discrimination suffered by women. Such
a result seems likely if highly skilled minority women have less nonearned income than their white counterparts.
However, since we have no direct evidence concerning the extent
of selection bias in the three samples of women, we focus most of
our attention on men. We present parallel results for women, but a
complete analysis of the racial wage gaps observed among women
remains a topic for further research.
The usual approach is to control for skill with a schooling variable.
When years of schooling (in 1991 when wages are observed) is used
instead of AFQT as the measure of skill (as shown in cols. 2 and 5
of table l), it reduces the unadjusted wage gap by only one-fifth for
men and only one-sixth for women.
l o M urnane et al. report a Hispanic-white wage gap of .lo5 among women. Their
approach differs from ours...
View Full Document