The measure of cognitive ability is the score on the Armed Forces Qualification Test (AFQT). For a detailed description of the NLSY79, see Herrnstein and Murray (1994): Appendix 2. For a detailed discussion of technical issues regarding the AFQT as a measure of IQ, see Herrnstein and Murray (1994): Appendix 3. The short story is that the AFQT is one of the most highly g-loaded paper-and-pencil tests. By way of comparison, the median factor g-loading for the subtests of the AFQT is .85, compared to a median of .69 for the subtests of the most widely used IQ test for adults, the Wechsler Adult Intelligence Scale (WAIS). The first factor, g, accounts for over 70 percent of the variance in the AFQT compared to 53 percent in the WAIS. Herrnstein and Murray (1994): 607. AFQT scores were normalized by year of age at testing and converted to the IQ metric, with a range of 55–145. They are hereafter referred to as IQ scores.
The analyses adjusting for IQ are ordinary least squares regressions for years of education and earned income, and logit analyses for the probability of getting a college degree. All of the analyses were replicated with and without interaction terms. Few of the interaction terms even approached statistical significance, and the fitted values for the analyses including interaction terms were extremely close to the fitted values for those that were restricted to main effects. I therefore report the analyses based on main effects.
To facilitate comparisons of the results for the two NLSY cohorts, I limited the samples to comparable age ranges at the date of the follow-up I report. At the date of the most recent interview for the NLSY97 in 2015, 99 percent of the cohort were ages 30–35, with a small number (69) who were 36. The sample for analysis was limited to ages 30–35. I selected 1994 as the follow-up survey for the NLSY79 cohort and limited the analysis for both cohorts to those who were ages 30–35 on the date of the interview (the NLSY79 cohort was chosen from a wider range of birth years, 1957–64, than the NLSY97 cohort, who were born from 1980 to 1984). Income figures for both cohorts were converted to 2018 dollars.
The table below shows results for years of education and percentage of persons with a college degree for both of the NLSY cohorts. For both indicators, the dependent variable was regressed on IQ and a binary variable denoting sex. The first table shows the fitted values for years of education when IQ is set to 80, 100, and 120, which correspond approximately to the 10th, 50th, and 90th percentiles of IQ in a normal distribution.
Years of education: NLSY79
IQ: 80
Men: 11.4
Women: 11.7
IQ: 100
Men: 13.3
Women: 13.6
IQ: 120
Men: 15.3
Women: 15.5
Years of education: NLSY97
IQ: 80
Men: 11.6
Women: 12.3
IQ: 100
Men: 13.9
Women: 14.6
IQ: 120
Men: 16.1
Women: 16.8
Females had higher values for years of education than males for every IQ category in both NLSY samples. Consistent with the story for sex differences in educational attainment in chapter 4, the female advantage in fitted values for years of education is greater for the NLSY97, whose members were born in 1980–84, than for the NLSY79, whose members were born in 1957–64.
The next table shows the results for achievement of a college degree. Since college degrees are rare among people with measured IQs of less than 100, I show the results for fitted values of IQ set to 100, 110, and 120.
College degree: NLSY79
IQ: 100
Men: 15%
Women: 19%
IQ: 110
Men: 35%
Women: 42%
IQ: 120
Men: 61%
Women: 68%
College degree: NLSY97
IQ: 100
Men: 24%
Women: 35%
IQ: 110
Men: 44%
Women: 57%
IQ: 120
Men: 66%
Women: 77%
Women had an advantage for all three fitted values in both cohorts. As in the case of years of education, the female advantage increased from the NLSY79 cohort to the NLSY97 cohort.
6. Since 2003, the Census Bureau has used a set of options for self-identified ethnicity that permits combinations of two or three ethnicities. The numbers that follow are based on those who self-identified as white only, black only, or Asian only, and also reported that they were not Latino. The Latino number is based on all those who self-identified with a specific Latino population (e.g., Puerto Rican, Cuban, Chicano, Mexican American), regardless of their self-identified race. Mean years of education for 2018 are shown below:
CPS 2018
White
Years of education: 14.8
College degree: 44.1%
Black
Years of education: 14.1
College degree: 28.3%
Latino
Years of education: 12.6
College degree: 19.4%
Asian
Years of education: 15.4
College degree: 64.6%
Other
Years of education: 14.2
College degree: 28.8%
Source: CPS 2018, for persons 25–54.
Before adjusting for IQ, Asians have advantages in both years of education and the percentage with college degrees.
7. The table below shows IQ by ethnicity for the two cohorts of the NLSY and, to demonstrate how typical these results are, ethnic means for the three standardizations of the WAIS over the last 40 years. Asians are not reported for NLSY79 because that survey was limited to whites, blacks, and Latinos. The ethnic designations for NLSY79 use the Sample ID variable (variable no. R0173600). The ethnic designations for NLSY97 combine information from the Key!Race and Key!Race_Ethnicity variables (nos. R538700 and R1482600) and follow the pattern for the CPS: non-Latino whites, blacks, and Asians; and Latinos of any self-identified race.
Mean IQ by ethnicity
NLSY (1979)
White: 103.1
Black: 86.7
Latino: 90.5
NLSY (1997)
White: 103.2
Black: 89.6
Latino: 94.0
Asian: 105.9
WAIS-R (1981)
White: 101.4
Black: 86.9
WAIS-III (1995)
White: 101.4
Black: 86.8
WAIS-IV (2008)
White: 103.2
Black: 88.7
Latino: 91.6
Asian: 106.1
Sources: Reynolds, Chastain, Kaufman et al. (1987): Table 1; Dickens and Flynn (2006): Supplemental data provided by the authors.
The table below shows the results, expressed in years of education, when the number of years of education is regressed on IQ and ethnicity. The sample was limited to persons whose age at the most recent measure of years of education was 30 or above.
Years of education: NLSY79
IQ: 80
White: 11.0
Black: 12.0
Latino: 11.2
IQ: 100
White: 13.2
Black: 14.2
Latino: 13.4
IQ: 120
White: 15.4
Black: 16.4
Latino: 15.6
Years of education: NLSY97
IQ: 80
White:11.7
Black: 12.3
Latino: 11.8
Asian: 12.9
IQ: 100
White:14.0
Black: 14.6
Latino: 14.2
Asian: 15.2
IQ: 120
White:16.4
Black: 17.0
Latino: 16.5
Asian: 17.6
For the NLSY79, blacks had an advantage of 1.1 years of education over whites after adjusting for IQ, and Latinos had an advantage of 0.2 years over whites. For the NLSY97, the black advantage was 0.6 years of education and the Latino advantage was 0.1 years.
The next table shows the results of a logit regression of achievement of a college degree (yes or no) on the same independent variables. The table shows the percentages of each group obtaining a college degree. The fitted values of IQ shown in the table are 100, 110, and 120, corresponding approximately to the 50th, 75th, and 90th percentiles.
College degree: NLSY79
IQ: 100
White: 15%
Black: 26%
Latino: 13%
IQ: 110
White: 36%
Black: 53%
Latino: 33%
IQ: 120
White: 65%
Black: 78%
Latino: 61%
College degree: NLSY97
IQ: 80
White: 28%
Black: 34%
Latino: 26%
Asian: 54%
IQ: 100
White: 49%
Black: 56%
Latino: 47%
Asian: 75%
IQ: 120
White: 71%
Black: 76%
Latino: 69%
Asian: 88%
Blacks got more college degrees than whites in both of the NLSY cohorts for comparable IQs. So did Asians in the NLSY97. Latinos and whites with comparable IQs have close to the same percentage of college graduates in both NLSY cohorts, with a slight advantage for whites.
8. The indispensable source for understanding the nature of the remaining male-female income disparities is Goldin (2014). I limit this presentation to the basics provided by the CPS and the NLSY. The table below shows median earned income (combined wages or salary, income from business, income from farm) from the 2018 CPS for men and women ages 25–54 with various combinations of marital status, children in the home, and labor force status. The results use the CPS sample weights for persons.
CPS 2018
Married with at least one child in the house
Median earned income (000s): Men: $60
Median earned income (000s): Women: $38
Female-male ratio: 63%
Married, child, worked 52 weeks
Median earned income (000s): Men: $63
Median earned income (000s): Women: $43
Female-male ratio: 68%
Married, child, worked 52 40-hr. weeks
Median earned income (000s): Men: $60
Median earned income (000s): Women: $45
Female-male ratio: 75%
Married, no child in the house
Median earned income (000s): Men: $52
Median earned income (000s): Women: $40
Female-male ratio: 77%
Married, no child, worked 52 weeks
Median earned income (000s): Men: $56
Median earned income (000s): Women: $45
Female-male ratio: 80%
Married, no child, worked 52 40-hr. weeks
Median earned income (000s): Men: $55
Median earned income (000s): Women: $45
Female-male ratio: 82%
Single with at least one child in the house
Median earned income (000s): Men: $40
Median earned income (000s): Women: $29
Female-male ratio: 73%
Single, child, worked 52 weeks
Median earned income (000s): Men: $42
Median earned income (000s): Women: $33
Female-male ratio: 79%
Single, child, worked 52 40-hr. weeks
Median earned income (000s): Men: $42
Median earned income (000s): Women: $35
Female-male ratio: 83%
Single, no child in the house
Median earned income (000s): Men: $38
Median earned income (000s): Women: $35
Female-male ratio: 92%
Single, no child, worked 52 weeks
Median earned income (000s): Men: $42
Median earned income (000s): Women: $40
Female-male ratio: 95%
Single, no child, worked 52 40-hr. weeks
Median earned income (000s): Men: $40
Median earned income (000s): Women: $40
Female-male ratio: 100%
Women had a “marriage premium”: Married women in every category earned more than single women, and yet the female-male earnings ratio for married people was lower than the ratio for single women. The explanation: The marriage premium for men was even larger than it was for women.
9. The patterns in the CPS data are evident as well after adjusting for IQ. The table below shows the fitted value of annual earned income for people who self-reported being in the labor force throughout the year and who worked at least one hour. The regression analysis was conducted separately for men and women. The independent variables were IQ, a binary variable for marriage (no-yes), and a binary variable for children living in the home (no-yes). I also conducted analyses discriminating between the presence of children under the age of five and of children five years and older, but do not report them because they did not show substantively different results. Income in the NLSY surveys refers to the calendar year (CY) prior to the interview.
TWO INCOME MEASURES AT AGES 30–35 BY IQ, SEX, MARITAL STATUS, AND PRESENCE OF CHILDREN (2018 DOLLARS)
Unmarried, no children: NLSY79, CY 1993
Income measure: Earned income
IQ: 80
Men: $26,837
Women: $27,324
Ratio: 1.02
Income measure: Earned income
IQ: 100
Men: $37,365
Women: $37,676
Ratio: 1.01
Income measure: Earned income
IQ: 120
Men: $52,023
Women: $51,949
Ratio: 1.00
Income measure: Total family income
IQ: 80
Men: $25,053
Women: $22,519
Ratio: 0.90
Income measure: Total family income
IQ: 100
Men: $36,845
Women: $34,602
Ratio: 0.94
Income measure: Total family income
IQ: 120
Men: $54,187
Women: $53,170
Ratio: 0.98
Unmarried, no children: NLSY97, CY 2014
Income measure: Earned income
IQ: 80
Men: $27,685
Women: $24,142
Ratio: 0.87
Income measure: Earned income
IQ: 100
Men: $36,909
Women: $35,458
Ratio: 0.96
Income measure: Earned income
IQ: 120
Men: $49,206
Women: $52,077
Ratio: 1.06
Income measure: Total family income
IQ: 80
Men: $32,395
Women: $29,637
Ratio: 0.91
Income measure: Total family income
IQ: 100
Men: $47,646
Women: $49,203
Ratio: 1.03
Income measure: Total family income
IQ: 120
Men: $70,079
Women: $81,684
Ratio: 1.17
Married with children: NLSY79, CY 1993
Income measure: Earned income
IQ: 80
Men: $37,408
Women: $22,968
Ratio: 0.61
Income measure: Earned income
IQ: 100
Men: $52,084
Women: $31,669
Ratio: 0.61
Income measure: Earned income
IQ: 120
Men: $72,517
Women: $43,667
Ratio: 0.60
Income measure: Total family income
IQ: 80
Men: $49,393
Women: $48,440
Ratio: 0.98
Income measure: Total family income
IQ: 100
Men: $72,641
Women: $74,434
Rat
io: 1.02
Income measure: Total family income
IQ: 120
Men: $106,832
Women: $114,375
Ratio: 1.07
Married with children: NLSY97, CY 2014
Income measure: Earned income
IQ: 80
Men: $42,799
Women: $22,706
Ratio: 0.53
Income measure: Earned income
IQ: 100
Men: $57,058
Women: $33,349
Ratio: 0.58
Income measure: Earned income
IQ: 120
Men: $76,067
Women: $48,979
Ratio: 0.64
Income measure: Total family income
IQ: 80
Men: $56,657
Women: $44,090
Ratio: 0.78
Income measure: Total family income
IQ: 100
Men: $83,332
Women: $73,197
Ratio: 0.88
Income measure: Total family income
IQ: 120
Men: $122,566
Women: $121,518
Ratio: 0.99
Unmarried women without children. The cohort of women in their early 30s in 1994 earned as much as men at comparable IQ levels. For the cohort of women in their early 30s in 2015, that held true for high-IQ women. Judging from these results, there was a meaningful male advantage only for quite low levels of IQ (e.g., the fitted value for earned income when IQ was set at 80 was 87 percent of income for unmarried males with no children in the home).
Married women with at least one child in the home. Compare the earned income for unmarried and married women. Married women in the labor force earned noticeably less than unmarried women. But the larger factor that drove down the ratio of female to male earnings was that married men earned much more than unmarried men.
Family income. I show family income in addition to earned income to make a point that will come as no surprise. Married women with children are, on average, much more prosperous than unmarried women at comparable IQ levels. For the NLS79 cohort, married women with children had more than double the fitted family income of unmarried women without children (115 percent). That marriage premium had dropped for the NLSY97 cohort but was still a substantial 49 percent.
10. The raw differences in earned income in the 2018 CPS are shown for two groups: those who were in the labor force at the time of the interview (not necessarily employed) and those who reported working for 52 weeks at 40 hours per week during the year.
Human Diversity Page 55