hypothesis of non-correlation is not rejected,23 indicating that the Hausman-Taylor estimators are
consistent.
Regressions 3C and 3D add three interaction terms of education with the race/gender
dummy variables to the regression of each time period to investigate the possibility of different
returns to education across different demographic groups. The coefficient of g decreases to 0.243
in both periods (and is statistically insignificant in the later period), probably due to the positive
correlation between g and individual education level. Similar to 3A and 3B, none of the slope
terms of g with race/gender dummy variables are statistically significant. Hence, our previous
finding of no systematic difference in the marginal wage return to intelligence across the
race/gender groups is corroborated. In contrast, the interaction terms of education with both the
black and Hispanic race dummy variables in regression 3C are significant at the 5 percent
significance level. All things equal, the marginal wage return to education is 0.031 and 0.046
lower for blacks and Hispanics respectively compared to whites. In other words, although one
extra year of education increases hourly wage by 10.5 percent for whites, it increases the hourly
wage by only 7.4 percent for blacks and 5.9 percent for Hispanics. Furthermore, the cross term of
education with the gender dummy is not significant in regression 3C, indicating that males and
females earn about the same return to education. The coefficients of other control factors, such as
tenure and experience, in regressions 3C and 3D are all identical with those in 3A and 3B.
Comparing regressions 3C and 3D yields some interesting observations. Both the
Hispanic-education and black-education cross terms become insignificant after 1994. In addition,
the coefficient of the education-gender interaction term is not significant in 3D. Moreover, the
coefficients on number of children, marital status, and residence area all become insignificant in
3D. Two possibilities may account for the discrepancies between the two periods. First, a sample
23 Before 1994: Chi square = 1.01, p>Chi square = 0.999; after 1994: Chi square = 1.10, p>Chi square = 0.999.
19
More intriguing information
1. THE USE OF EXTRANEOUS INFORMATION IN THE DEVELOPMENT OF A POLICY SIMULATION MODEL2. The name is absent
3. An alternative way to model merit good arguments
4. The effect of classroom diversity on tolerance and participation in England, Sweden and Germany
5. The WTO and the Cartagena Protocol: International Policy Coordination or Conflict?
6. Regional science policy and the growth of knowledge megacentres in bioscience clusters
7. Fighting windmills? EU industrial interests and global climate negotiations
8. Firm Closure, Financial Losses and the Consequences for an Entrepreneurial Restart
9. The name is absent
10. Wirkung einer Feiertagsbereinigung des Länderfinanzausgleichs: eine empirische Analyse des deutschen Finanzausgleichs