education matters in the determination of one's wage rate. Graduates from better colleges have a
better chance to land a higher-paying starting job. As with years of schooling, education quality
also changes systematically with intelligence. Better schools tend to be more selective and
require a higher score on admission tests, such as the SAT or GRE. Due to the limitation of the
NLSY79, these unobservable characteristics cannot be included in the analysis, leading to
potentially biased estimators. Thus, we refrain from further interpretations of the cross-sectional
regression results.
Panel Regression Results and Interpretation
The random effects panel regression model employed by Cawley et al. (1996, 1999)
controls effectively for unobservable factors so that they can obtain consistent estimates. To
investigate wage returns to intelligence across different demographic groups, they divide the
sample into six sub-samples and run separate regressions for each sub-sample.14 The intelligence
measure, g, is derived from principal component analysis. Their control variables include a set of
human capital measures: schooling (measured as grades completed), weeks of tenure in the
current/most recent job, tenure squared (to account for diminishing return to tenure), labor
market experience, and experience squared. Eicker-White standard errors generalized for panel
data are used to correct for heteroskedasticity. Their results show that “ability is rewarded
unequally in the labor market - workers of a given measured ability receive different wages
depending on their race and gender, with these differences being statistically and numerically
significant” (Cawley et al. [1999], p. 251).
14 The six groups are white males, white females, black males, black females, Hispanic males, and Hispanic females.
10