independent and significant effect on the wage rate and should be included in the set of control
variables.
In their analysis, Fisher et al. (1996) employ a more complete set of control variables,13
and conclude that the wage regressions of Herrnstein and Murray suffer from omitted variable
bias, resulting in an over-estimation of the effect of the AFQT score on the wage rate. Regression
1C is our replication of Fisher et al. (1996), but using the 2002 respondent data; it confirms their
findings. In contrast to equations 1A and 1B, the estimated coefficient for the normalized AFQT
scores drops to 0.027 and becomes statistically insignificant. This result shows that a
considerable amount of the explanatory power of the AFQT scores on the wage rate can be
attributed to the newly included variables. Furthermore, the estimated coefficients on many of
the control variables other than the AFQT scores are statistically significant. Notably, the
coefficient of years of schooling is 0.069 and significant at the 1 percent level.
We also decompose the SES index into its components in regression 1C. Herrnstein and
Murray summed up the normalized father’s education, mother’s education, family income, and
Duncan’s Social Economic Index, and averaged the sum to obtain the SES index, thus effectively
assigning the same weight to all the four components. However, regression 1C does not provide
support for the hypothesis that the four components are equally weighted in their effects on the
wage rate. In fact, an F test on the null hypothesis that the coefficients of the four components
are equal can be rejected at 5 percent significance level (F=4.35, p>F=0.046). Thus, breaking the
SES index up into its components generates a more accurate measure of the individual impacts
of the components. Moreover, we follow Mincer (1974) and define (potential) labor market
13 The modified list of control variables includes parental home environment (family income, mother’s education, father’s
education, parent’s SEI index, and the number of siblings), respondent’s adolescent community environment (residence region of
the country at age 14), respondent’s current family and community background (marital status, number of children, and the local
unemployment rate), human capital measures(education, current/most recent job tenure, and labor market experiences), age, race
and gender.