Regression 3E shows that differences between the results from the two periods are likely
due to structural changes in the sample population over time. Being married, having fewer kids,
and living in an urban area are all significantly correlated with higher wage rate in the first
period; however, as the respondents grow older, these factors all become less significant in
explaining wage differentials in the second period. Although the regression results indicate this
structural change, we do not have sufficient information to suggest a testable hypothesis for this
temporal structure change from the first period to the second period in the sample. Such
information would include how much time one spends on child-caring per child or whether
living in the urban area offers a young adult more opportunities to find a well paid job. Moreover,
it remains unclear why there is significant differences in the return to education across
racial/ethnic subgroups in the first period, but not in the second period.
Overall, the Hausman-Taylor estimators are useful in obtaining consistent coefficient
estimates on the time-invariant variable g and its cross terms with the racial and gender dummy
variables without violating the strict exogeneity assumption. With the Hausman specification test
between the fixed effects model and the Hausman-Taylor model, we verify that these estimators
are consistent in our analysis. Hence, we provide evidence that Cawley et al. (1996, 1999) may
have under-estimated the marginal effect of g on wage differentials by employing the random
effects specifications. More surprisingly, the wage returns to intelligence appear to be similar
across racial and gender groups, while the wage returns to education differ significantly between
whites and minorities in the first time period.
21