al. (1996, 1999) respectively employ the raw AFQT scores and g as proxies for measured
cognitive abilities in their re-analyses of The Bell Curve. In our analysis, we extract g from the
full battery of the ASVAB tests through a variation of factor analysis called principal component
analysis; g is formed by taking principal components of the correlation matrix of ASVAB test
scores, and then multiplying the component associated with the largest eigenvalue from the test
scores matrix. Principal component analysis produces essentially the same results as other factor
analysis methods but is affected the least by sampling error.11
III Cross-sectional and Panel Analyses
Cross-sectional Regressions
Herrnstein and Murray’s analysis on the wage rate is cross-sectional in nature. Instead of
utilizing all of the NLSY79 survey years, they focus on the 1989 survey data only. Their measure
of the wage rate, the dependent variable, is the log of the hourly wage rate of the current or most
recent job reported by the NLSY79 respondents in 1989. The independent variables include the
AFQT score, the respondent’s socio-economic index as the measure of family background,12 and
age, all normalized. People who are currently enrolled in college are excluded from the sample
because their wage rates from mostly on-campus jobs are low regardless of their intelligence.
Regression 1A in table 1 is the closest replication to Herrnstein and Murray’s analysis
that we could achieve. The log of the hourly pay rate for the respondent’s current/most recent job
is the dependent variable, while normalized age, parents’ SES index, and the AFQT scores are
11 See Cawley et al. (1996) .
12 The SES index is formed based on father’s education, mother’s education, log of family income, and the Duncan Social-
Economic Index (SEI) score associated with either father’s or mother’s occupation (the higher among the two). The SEI score is
based on the United States Bureau of the Census, 1950 Census of Population. Summary statistics on education and income of all
the occupations are obtained from the Census and adjusted for age differences among occupations. Then SEI score is constructed
for each occupation measured using its age-adjusted median education and income level. For a complete description of the
methodology, see Duncan et al. (1972).