region, and possibly many more other variables are statistically significant, and cannot be
neglected in the analysis of wage equations.
Moreover, although we find no significant evidence for different wage returns to
intelligence among racial and gender subgroups, the returns to education do differ across
race/ethnicity in the time period from 1979 to 1994. Our analysis indicates structural changes in
the wage equation from the first time period to the second one in the coefficients of marriage,
number of children, current residents, and race-intelligence cross-terms become insignificant in
the second period.25
Finally, we acknowledge other issues due to the limitations of the data set. Some
important causal variables that might affect one’s wage rate are missing in the regression, such as
an individual’s education quality, his/her adolescent neighborhood conditions (crime rate,
average annual income of the community, etc.), and his/her work attitude (work ethics, effort
level, etc.). Designing a legitimate way to approximate these missing variables from the NLSY79
is difficult. Also, the NLSY79 contains only one set of mental aptitude measures, namely, the
ASVAB Test administered in 1980. As a result, we cannot test an important and fundamental
characteristic of intelligence, i.e., that g is time-invariant and cannot be altered easily by
environmental factors. Instead, we must maintain this property as an assumption throughout the
paper. If other sets of the ASVAB scores were available in the later survey years, we could test
for any systematic changes in the test scores over time and investigate whether IQ scores can be
changed through education or external environment variables. Thus, further research, possibly
using a new and more comprehensive data set, is needed to derive more definitive conclusions.
25 The sepearation of the NLSY79 data into two time periods at this particular breakpoint is dictated by the fact that the NLSY is
conducted yearly until 1994 and biennially afterwards; re run two sets of regressions to maintain evenly spaced time periods
within each regression. This division point need not capture the right timing of structural changes in the wage equation.
23