Revisiting The Bell Curve Debate Regarding the Effects of
Cognitive Ability on Wages
Abstract
In The Bell Curve, Herrnstein and Murray (1994) claim, based on evidence from cross-
sectional regressions, that differences in wages in the U.S. labor market are predominantly
explained by general intelligence. Cawley, Heckman, and Vytlacil (1999), using evidence from
random effects panel regressions, reject this claim, in part because returns to general intelligence
vary by racial and gender subgroups in their results. In this article, we examine the regression
methods used by both sides of the debate and conclude that neither is the appropriate method to
analyze the NLSY data that both use. We introduce the Hausman-Taylor estimator to obtain
consistent estimated coefficients on the time-invariant general intelligence-related variables and
also extend the analysis up through 2002. While many additional socio-economic factors are
important explanatory variables in determining the wage rate, the effect of general intelligence
on wages is larger in the Hausman-Taylor specification for the 1979-1994 panel than in either
the cross-sectional or random effects models, though it becomes statistically insignificant for the
1994-2002 panel. The Hausman-Taylor analysis also indicates no significantly different returns
to intelligence by race or gender group.
Keywords: wages; cognitive ability; education
JEL Codes: J24; J31