V Conclusion
In this paper, we investigate both sides of debate around The Bell Curve by replicating
the statistical analyses of the wage equation by Herrnstein and Murray and their critics. The
cross-sectional regression model used by Herrnstein and Murray (1994) is wrongly specified and
the marginal effect of AFQT and g on the wage rate is over-estimated relative to that due to other
factors. They neglect in their analysis many explanatory factors, such as the human capital
measures including education, job tenure, and labor market experience, which contribute
significantly to the wage differentials. Moreover, the cross-sectional regression model suffers
from omitted variable bias because unobservable individual characteristics, such as quality of
education, cannot be included. Cawley et al. (1996, 1999) employ a random effects panel
regression to demonstrate that the wage return to intelligence is not uniform across demographic
subgroups. However, the strict exogeneity assumption for the random effects model is rejected in
our analysis and only the fixed effects model can be applied to the data set. Since intelligence is
assumed to be time-invariant and thus drops out of the model, the fixed effects regressions
provide little relevant information on the effect of g on the wage differentials.
Hausman and Taylor (1981) propose a variation of the random effects model that
provides consistent estimators despite the violation of the strict exogeneity condition. Applying
the Hausman-Taylor model to the data set, we show that the original random effects model
employed by Cawley et al. under-estimates the predictive power of g on the wage rate. Our
analysis does confirm Fisher et al. (1996) and Cawley et al. (1996, 1999)’s claim that socio-
economic background variables and human capital measures are important factors in explaining
the wage differentials: education, job tenure, marital status, number of children, residential
22