Last Updated on March 29, 2023
Most people are aware that there are significant disparities between blacks and whites in the United States with regard to a wide range of important social outcomes, including crime, income, education, poverty, welfare usage, etc. For almost every measurable metric of important life outcomes, blacks perform significantly worse than whites. In this post, I will cite studies showing that many of these disparities are likely caused by the significant cognitive differences between blacks and whites. I begin by illustrating a few examples of the disparities between blacks and whites with respect to important life outcomes. Then I briefly review evidence demonstrating the predictive validity and causal influence of cognitive ability for these outcomes. Next, I present data illustrating the scope and magnitude of the black-white cognitive ability gap. Finally, I provide evidence indicating that many of the aforementioned disparities between blacks and whites are (mostly) eliminated after controlling for youth cognitive ability.
Black households have far lower average incomes than white households. US census data (2017) [archived] shows that the average black household income was more than one-third lower than the average white household income in 2017 ($40,258 vs $68,145, Table 1). This disparity has barely budged within over 50 years:
The same data shows that black households were over twice as likely to be poor as white households in the same year (21.2% vs 8.7%, Table 3). More recent data [archived] from the US census shows that, while the black poverty rate has decreased significantly since the 1960s, there are still large disparities in poverty rate between blacks and whites today:
Part of the explanation for racial disparities in household income is that black households are far less likely to be dual-income households than white households, independently of any income disparity between black and white individuals. However, data [archived] from the National Center for Education Statistics shows black workers have far lower annual earnings than white workers. The median annual earnings for black workers was $11,200 lower than the median earnings for white workers ($33,700 vs $44,900):
Controlling for educational attainment does not eliminate the disparity, as black workers have lower median earnings than white workers at each of the major levels of education:
Data from the Pew Research Center [archived] also showed significant racial disparities in wealth that have persisted for generations. In fact, the ratio of median white wealth to median black wealth is greater today than it was in the 1980s:
A report by Mazumder (2008) [archived] published by Pew Charitable Trusts used the NLSY79 to examine factors relevant to income mobility. Income mobility was examined by investigating family income in adolescence (1978-1980) and family income as adults (1997-2003). Prior to controlling for any covariates, the study finds stark racial differences in income mobility. For example, about 75% of whites raised in the bottom income quintile eventually transition out of that quintile, whereas only 56% of blacks do the same. In fact, blacks raised in the 2nd highest income quintile are equally as likely to end in the bottom quintile as are whites raised in the lowest income quintile (24.6% vs 24.9%):
At every range of parental income, black children are far less likely to exceed their parent’s income:
I wrote more on racial disparities in intergenerational mobility in this post.
More data from the National Center for Education Statistics (NCES) [archived] shows large disparities in educational attainment between blacks and whites. For example, blacks are about twice as likely as whites to fail to complete High School (Figure 27.1) and they are about half as likely to attain a bachelor’s degree (Figure 27.3).
Data from Pew Research Center [archived] shows that the black-white disparity in high school completion has narrowed significantly in recent decades, although large gaps remain. On the other hand, the college completion gap between blacks and whites has barely changed within 50 years.
Crime and misbehavior
Black people are disproportionately involved in criminality. Data from FBI crime statistics (2015) [archived] shows that despite making up only 13% of the US population, black people commit 36% of violent crime in the US. Even worse, they commit over half of the robberies and murders in the country. Among criminals under the age of 18, black youth commit over 60% of the robberies and murders in the country, and over half of the violent crime.
In 2003, the Bureau of Justice Statistics [archived] (Figure 4) released a report showing that 1 in 3 black males could be expected to go to prison if the current rates of imprisonment had remained unchanged.
These disparities show that, given 2003 rates of incarceration, black females were about as likely to be incarcerated as white males. Fortunately, incarceration rates have decreased since then, so the lifetime chances of being incarcerated have lowered since then as well. Recent data from the Bureau of Justice Statistics (2020) [archived] shows that blacks are much more likely to be incarcerated than whites, although the disparity has decreased in recent years (Figure 1). The racial disparities are greatest for the youngest males. In fact, the imprisonment rate for black males aged 18-19 years is over 12 times higher than the rate for similarly aged white males (Figure 2).
Some people might object that these statistics only show that black people are more likely to be arrested or incarcerated without showing that they black people are more likely to commit crimes. A simple test of this hypothesis involves looking at the leading causes of death by race. If black people are much more likely to be killed via homicide, then that is evidence that black people are far more likely to commit homicide (since the vast majority of homicides are intraracial rather than interracial). The CDC (2015) [archived] reports that homicide is the leading cause of death for black males aged 15-34, with nearly half of deaths for men aged 15-24 the result of homicide (see page 34). Contrast this with white males of the same age for whom homicide causes roughly between 3-5% of deaths (see page 27). In fact, the death rate due to homicides for blacks aged 20-24 (110.8 deaths per 100,000 population) is over 20 times the rate for similarly aged whites (5.4 deaths per 100,000 population).
Disparities in troublesome behavior between blacks and whites appear far before adulthood. For example, among middle-schoolers [archived], black males are suspended at 3 times the rate for white males, and black females are suspended at over 4 times the rate for white females. Furthermore, despite making up just 18% of enrolled preschoolers, black children account for 48% [archived] of preschool children suspended more than once.
I go into more detail on the racial disparities in crime and misconduct in a separate post.
Why cognitive ability matters
This section will summarize points that I’ve made about the predictive validity of cognitive ability in a separate post.
Intelligence is a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather, it reflects a broader and deeper capability for comprehending our surroundings-“catching on,” “ making sense” of things, or “figuring out” what to do. (page 13)
For a more formal definition of the meaning of cognitive ability that I have in mind, see the Cattell–Horn–Carroll (CHC) theory of human cognitive abilities, which has been deemed the “most widely accepted theory” of intelligence (Sternberg 2012). The CHC theory is a synthesis of the Cattell–Horn Gf–Gc and Carroll Three-Stratum models, which have “have emerged as the consensus psychometric-based models for understanding the structure of human intelligence” (McGrew 2009). The CHC theory affirms that there are three strata of intelligence that hierarchically relate to one another. At the top of the hierarchy (stratum III) is the general factor of intelligence (also referred to as the g factor or g), which will be explained more later in this post. In the middle of the hierarchy (stratum II) are broad abilities such as fluid and crystalized intelligence. At the bottom of the hierarchy (stratum I) are narrow abilities nested under each broad ability. The broad abilities are comprehension-knowledge (Gc), fluid reasoning (Gf), quantitative knowledge (Gq), reading & writing ability (Grw), short-term memory (Gsm), long-term storage and retrieval (Glr), visual processing (Gv), auditory processing (Ga), and processing speed (Gs). The degree to which a test measures these abilities corresponds to the degree to which the test measures my definition of cognitive ability. Fortunately, virtually all tests of mental abilities measure at least some of these abilities, and all tests measure g to some extent.
My working definition of cognitive ability is measured by IQ tests fairly accurately. It is important to understand IQ because, as Nisbett et al. (2012) [archived] notes, IQ is the measure of intelligence for which “the bulk of evidence pertinent to intelligence exists” (page 131). To start, one should understand how IQ scores are distributed. IQ scores are normed for a given population to produce a mean score of 100 and a standard deviation (SD) of 15 points. Because IQ scores are normally distributed, 32% of the population has an IQ score of more than a standard deviation away from the mean. In other words, about 68% of the population has scores between 85 and 115. About 5% of the population has an IQ score of more than two standard deviations (30 points) from the mean. In other words, about 95% of the population has scores between 70 and 130 (Neisser et al. (1996) [archived], page 78).
My working definition of cognitive ability is also measured fairly accurately by other tests that are not officially IQ tests. This is because there are many tests that are highly g-loaded even though they aren’t officially IQ tests. Highly g-loaded tests produce test results that are strong measures of g, the general factor of intelligence at the top of the hierarchy in the CHC theory of cognitive abilities. g-loaded tests are important because, as Gottfredson (2002) [archived] notes, the more g-loaded a test is, “the better it predicts performance, including school performance, job performance, and income” (page 28). She also states the following regarding g (page 27):
Theorists have long debated the definition of “intelligence,” but that verbal exercise is now moot. g has become the working definition of intelligence for most researchers, because it is a stable, replicable phenomenon that—unlike the IQ score—is independent of the “vehicles” (tests) for measuring it. Researchers are far from fully understanding the physiology and genetics of intelligence, but they can be confident that, whatever its nature, they are studying the same phenomenon when they study g. That was never the case with IQ scores, which fed the unproductive wrangling to “define intelligence.” The task is no longer to define intelligence, but to understand g.
Furthermore, g scores extracted from different test batteries correlate nearly perfectly with one another (Johnson et al. 2004, Johnson et al. 2008, Kaufman et al. 2012). That being said, we can rely on tests that are not official IQ tests as good measures of cognitive ability if they are highly g-loaded. For example, ACT tests, SAT tests, AFQT tests, and even vocabulary tests are highly g-loaded, which make them satisfactory measures of cognitive ability on my working definition.
The expert consensus is that cognitive ability (as defined earlier) is a very powerful predictor, often the most powerful predictor, of a number of important social outcomes. For example, in a recent review of intelligence research by experts in the field, Nisbett et al. (2012) [archived] summarized the predictive power of IQ as follows (page 131):
The measurement of intelligence is one of psychology’s greatest achievements and one of its most controversial. Critics complain that no single test can capture the complexity of human intelligence, all measurement is imperfect, no single measure is completely free from cultural bias, and there is the potential for misuse of scores on tests of intelligence. There is some merit to all these criticisms. But we would counter that the measurement of intelligence — which has been done primarily by IQ tests — has utilitarian value because it is a reasonably good predictor of grades at school, performance at work, and many other aspects of success in life (Gottfredson, 2004; Herrnstein & Murray, 1994). For example, students who score high on tests such as the SAT and the ACT, which correlate highly with IQ measures (Detterman & Daniel, 1989), tend to perform better in school than those who score lower (Coyle & Pillow, 2008). Similarly, people in professional careers, such as attorneys, accountants, and physicians, tend to have high IQs. Even within very narrowly defined jobs and on very narrowly defined tasks, those with higher IQs outperform those with lower IQs on average, with the effects of IQ being largest for those occupations and tasks that are most demanding of cognitive skills (F. L. Schmidt & Hunter, 1998, 2004).
Gottfredson (1997) [archived] was a very brief 3-page statement that outlines conclusions regarded as mainstream by over 50 experts in intelligence and allied fields. Some of the conclusions they reached regarding the causal influence of cognitive ability were as follows (page 14):
- IQ is strongly related, probably more so than any other single measurable human trait, to many important educational, occupational, economic, and social outcomes. Its relation to the welfare and performance of individuals is very strong in some arenas in life (education, military training), moderate but robust in others (social competence), and modest but consistent in others (law-abidingness). Whatever IQ tests measure, it is of great practical and social importance.
- A high IQ is an advantage in life because virtually all activities require some reasoning and decision-making. Conversely, a low IQ is often a disadvantage, especially in disorganized environments. Of course, a high IQ no more guarantees success than a low IQ guarantees failure in life. There are many exceptions, but the odds for success in our society greatly favor individuals with higher IQs.
- The practical advantages of having a higher IQ increase as life settings become more complex (novel, ambiguous, changing, unpredictable, or multifaceted). For example, a high IQ is generally necessary to perform well in highly complex or fluid jobs (the professions, management); it is a considerable advantage in moderately complex jobs (crafts, clerical and police work); but it provides less advantage in settings that require only routine decision making or simple problem solving (unskilled work)
- Differences in intelligence certainly are not the only factor affecting performance in education, training, and highly complex jobs (no one claims they are), but intelligence is often the most important. When individuals have already been selected for high (or low) intelligence and so do not differ as much in IQ, as in graduate school (or special education), other influences on performance loom larger in comparison.
Reeve and Charles (2008) [archived] examined the opinions of 30 experts in the science of mental abilities about their views on cognitive ability and cognitive ability testing. The study found a consensus among experts that general cognitive ability “is measured reasonably well by standardized tests”, that general cognitive ability “enhances performance in all domains of work”, that general cognitive ability “is the most important individual difference variable”, and even that general cognitive ability is “the most important trait determinant of job and training performance” (Table 1). Participants in the survey were selected from individuals on the editorial board of the journal Intelligence, from all registered members of the International Society of Intelligence Researchers, and from persons who had published three or more articles in Intelligence over the last 3 years (page 683). Experts were selected from this group by filtering down to “only individuals with a doctorate degree, and having at least five career publications on the topic of intelligence or testing” (page 683). This study was a replication of Murphy, Cronin, and Tam (2003) [doi], which found largely similar results.
Rindermann, Becker, and Coyle (2020) [doi] surveyed the opinions of over 100 experts in the field of intelligence about a variety of questions. One of the questions in the survey was “to what degree is the average socioeconomic status (SES) in Western societies determined by his or her IQ?” The survey found that “Experts believed 45% of SES variance was explained by intelligence and 55% by non-IQ factors (Table 3). 51% of experts believed that the contribution of intelligence (to SES) was below 50%, 38% above 50%, and 12% had a 50–50 opinion.” That is, experts believe that roughly half of the variance in socioeconomic status in Western societies is due to intelligence.
A meta-analysis by Strenze (2007) [archived] shows that intelligence (measured by IQ scores) is a great predictor of future socioeconomic success. Socioeconomic success was measured as educational level, occupational status, and income. The analysis found that IQ measured before age 19 was a powerful predictor of socioeconomic success after age 29 (see “best studies” on Table 1).
The analysis concludes with the following (page 415):
These results demonstrate that intelligence, when it is measured before most individuals have finished their schooling, is a powerful predictor of career success 12 or more years later when most individuals have already entered stable careers. Two of the correlations – with education and occupation – are of substantial magnitude according to the usual standards of social science.
For more concrete examples of the association between adolescent cognitive ability and socioeconomic outcomes, see Murray (1998) [archived]. In this work, Murray used data from the NLSY79 to measure the predictive power of cognitive ability on a variety of socioeconomic outcomes. He separated subjects from the NLSY79 into 5 different “cognitive classes”: those who scored in the 90th+ AFQT percentile (classified as “very bright”), those who scored in the 75th-89th AFQT percentile (“bright”), those who scored in the 25th-75th AFQT percentile (“normal”), those who scored in the 10th-24th AFQT percentile (“dull”), and those who scored below the 10th percentile (“very dull”). He then reported the average levels of socioeconomic success for each cognitive class. As expected, those from the higher cognitive classes attained far higher levels of success than those in the lower cognitive classes. Consider the following findings (taken from tables 6-1 through 6-3):
|Cognitive Class (percentile range)||Mean Years of Education (1994)||Percentage obtaining a B.A. (1994)||Mean Weeks Worked (1993)||Median Earned Income (1993)|
|Very Bright (90th+)||16.5||77%||45.4||$36,000|
|Bright (75th – 89th)||15.0||50%||45.2||$27,000|
|Normal (25th – 74th)||13.2||16%||41.8||$21,000|
|Dull (10th – 24th)||11.9||3%||36.5||$13,000|
|Very Dull (10th-)||10.9||1%||30.7||$7,500|
These findings on the IQ-income correlations were corroborated by Zagorsky (2007) [archived]. He also used the National Longitudinal Survey of Youth 1979 to examine the association between youth IQ and income and net worth measured between the ages of 33 and 41 (page 491). The benefit of this study over the previous data is that this study was able to report on outcomes at a later stages in life. The study reported medium-large correlations between IQ and income (r = 0.30) and small-medium correlations between IQ and net worth (r = .16) (Table 2). The median incomes and net worth at different IQ points were as follows:
|IQ test score||Median income (2021 dollars)||Median net worth (2021 dollars)|
|120||$48,681 ($78,587)||$127,500 ($184,875)|
|110||$40,884 ($59,282)||$71,445 ($103,595)|
|100||$36,826 ($53,398)||$57,550 ($83,448)|
|90||$30,881 ($44,777)||$37,500 ($54,375)|
|80||$18,467 ($26,777)||$10,500 ($15,225)|
|Overall||$35,918 ($52,081)||$55,250 ($80,112)|
- The raw values are the figures in 2004 dollars, taken directly from the study. The values in parenthesis are in 2021 dollars, by multiplying the raw values by 1.45.
The black-white gap in cognitive ability
In this section, I’ll briefly review data on the size of black-white gaps in cognitive ability. To review data on such gaps in more detail, see my previous post here.
The magnitude of the gap
Research on race and IQ has fairly consistently shown that the average IQ score for blacks in the US is about one standard deviation (about 15 points) lower than the average IQ score for whites (Neisser et al. 1996 [archived], page 93). Studies often put the average black IQ at around 85 and the average white IQ at around 100 (Gottfredson 1997 [archived], page 14).
The size of the black-white gap in cognitive ability was observed in a meta-analysis by Roth et al. (2001) [archived]. As far as I know, this is the most recent meta-analysis reporting the magnitude of racial gaps in cognitive ability. The meta-analysis considered studies on racial disparities in general cognitive ability (g) among adults across educational and employment settings. The black-white gap in g hovered at about 1 standard deviation across different settings. The average gap was 1.1 standard deviations across over 6 million individuals pulled from 105 samples. Table 1 shows the magnitude of the gap in standard deviations (see columns for d) across different testing settings.
IQ scores are normally distributed for both racial groups. The IQ distributions of blacks and whites appear as follows (this graph was pulled from page 279 The Bell Curve):
The ubiquity of the gap
Test score gaps are also found across all ages. In fact, studies show that racial gaps emerge before children reach formal schooling. For example, Gottfredson (1997) [archived] reports that “Racial-ethnic differences in IQ bell curves are essentially the same when youngsters leave high school as when they enter first grade” (page 15). This was a claim published in a very brief 3-page statement that outlines conclusions regarded as mainstream among over 50 experts in intelligence and allied fields.
Farkas and Beron (2004) [archived] investigated oral vocabulary scores for black and white children by age using data from the Children of the NLSY79 (CNLSY). Vocabulary is measured using the Peabody Picture Vocabulary Test (PPVT). The PPVT is described as follows (page 473):
The Peabody Picture Vocabulary Test (PPVT) was used to measure oral vocabulary. The PPVT of spoken vocabulary consists of 175 words of generally increasing difficulty. The tester reads the word to the child, and the child points to one of four pictures that best describes its meaning. Testing stops and the child’s score (‘‘ceiling’’) is established when he or she incorrectly identifies six of eight consecutive items. The child’s score is the number of words identiﬁed correctly. (We analyze this variable in its raw score form, since we are controlling the child’s age in months.)
The researchers analyze data on vocabulary of children between the years of 1986 and 2000. Regarding racial disparities, the study found large racial disparities at the earliest observation which persisted throughout the entire study period (page 477):
Beginning with the earliest observation at 36 months of age, Whites average significantly higher scores than African-Americans. This pattern is consistent over the full age span, with the White lead remaining significant through 13 years of age. To see how very substantial this vocabulary gap is, note that Whites cross the 40-word level at approximately 50 months of age, whereas African-Americans do not reach this level until approximately 63 months, which puts them 13 months, or more than one year, behind in vocabulary development.
Test score gaps persist across all levels of education as well. For example, the meta-analysis by Roth et al. (2001) [archived] shows large racial gaps among high school students, college applicants, college students, and even graduate school applicants. The magnitude of the gaps are all at or above 1 standard deviation, except for the gap for college students which is only about 0.7 standard deviations.
Gaps of similar magnitude are found regardless of whether one analyzes elementary school samples or graduate application samples.
Cognitive ability gaps also persist even after adjusting for SES. From the introductory chapter of The Black-White Test Score Gap, Jencks and Phillips (1998) [archived] report that “a two-year reduction of the black-white education gap among mothers would only reduce the IQ gap by about a point for children” (page 22).They further state that “eliminating black-white income differences would reduce the IQ gap by about a point” (page 23).
The findings from Jencks and Phillips regarding parental income are corroborated by SAT scores disaggregated by race and income. Black-white differences in income do not explain a significant portion of black-white differences in SAT scores. Consider the following data [archived] on the racial gap in SAT scores in 2005:
Whites from families with incomes of less than $10,000 had a mean SAT score of 993. This is 129 points higher than the national mean for all blacks.
Whites from families with incomes below $10,000 had a mean SAT test score that was 61 points higher than blacks whose families had incomes of between $80,000 and $100,000.
Blacks from families with incomes of more than $100,000 had a mean SAT score that was 85 points below the mean score for whites from all income levels, 139 points below the mean score of whites from families at the same income level, and 10 points below the average score of white students from families whose income was less than $10,000.
Why the gap matters
- An IQ of 75 “signals the ability level below which individuals are not likely to master the elementary school curriculum or function independently in adulthood in modern societies” (page 28). They are likely to be eligible for “financial support provided to mentally and physically disabled adults” by the U.S. government. Such individuals are “difficult to train except for the simplest tasks, so they are fortunate in industrialized nations to get any paying job at all. While only 1 out of 50 Asian-Americans faces such risk, Figure 3 shows that 1 out of 6 black Americans does.”
- An IQ of 85 is another threshold considered because “the U.S. military sets its minimum enlistment standards at about this level” (page 28). The military is often viewed as a last resort by many people, but “this minimum standard rules out almost half of blacks (44%) and a third of Hispanics (34%), but far fewer whites (13%) and Asians (8%).” Individuals with IQs in this range “live at the edge of unemployability in modern nations, and the jobs they do get are typically the least prestigious and lowest paying: for example, janitor, food service worker, hospital orderly, or parts assembler in a factory.”
- An IQ of 105 can be viewed as “the minimum threshold for achieving moderately high levels of success” (page 30). People above this level are “highly competitive for middle-level jobs (clerical, crafts and repair, sales, police and firefighting), and they are good contenders for the lower tiers of managerial and professional work (supervisory, technical, accounting, nursing, teaching).” The percentages of people achieving this threshold of IQ are “53%, 40%, 27%, and 8%, respectively, for Asians, whites, Hispanics, and blacks.”
Obviously these large racial differences in cognitive ability will cause many racial differences in important social outcomes. However, this data actually understates the importance of the racial differences in cognitive ability. This is because the vast majority of black people tend to live in populations with high concentrations of black people. Therefore, black individuals are disadvantaged not only because they tend to have lower levels of cognitive ability, but also because they tend to live in communities where the mean cognitive ability is low. This means that they tend to live in areas with, e.g., high levels of crime, few occupational opportunities, little economic opportunity, high levels of social disorganization, etc. These macro-level disadvantages impose harms on their lives above and beyond the harms of their individual level of cognitive ability. Gottfredson also mentions the community-level challenges that are above and beyond the individual-level harms (page 31):
Where races tend to form separate communities, these disparities in individual-level outcomes create additional challenges at the community-level. Consider, for example, the percentages of each racial-ethnic group that score above IQ 110 vs. below IQ 90. For Asians it is 40% vs. 14%, for a ratio of almost 3:1. That is, there are 3 Asians who are at least somewhat above average in intelligence for every 1 who is at least somewhat below average. The pattern is starkly different for blacks: there is only 1 black who is at least somewhat above average in IQ for every 12 who are at least somewhat below average (59%). The percentages are 28% vs. 22% for whites, and 10% vs. 41% for Hispanics.
It should be clear that such large racial differences in cognitive ability are responsible for at least some of the social disparities that we find between blacks and whites today. In the rest of this post, I give examples of social disparities that are eliminated (or at least significantly reduced) after controlling for cognitive ability.
The relation between cognitive ability and racial inequalities
The National Longitudinal Surveys of Youth
The relation between youth test scores and life outcomes in the United States is perhaps best demonstrated using data from the National Longitudinal Surveys published by the U.S. Bureau of Labor Statistics. Most of the studies in this post will rely on analyses of these surveys, so it will be worth it to devote some time explaining this data. There were two separate longitudinal studies of two different cohorts: the 1979 National Longitudinal Survey of Youth (NLSY79) and the 1997 National Longitudinal Survey of Youth (NLSY97). Both surveys collected data on several thousands of subjects from a representative sample of the noninstitutionalized civilian population in the United States. For both surveys, data was collected at the start of the survey when participants were in their youth (in 1979 or 1997). The participants provided additional information in regularly scheduled follow-up interviews about important life outcomes, such as wages, educational attainment, marital status, etc.
- The NLSY79 includes about 10,000 participants born between 1957 to 1964. Participants were aged 14 to 22 years old when first interviewed in 1979. This cohort was interviewed in 28 separate rounds, with the most recent round being in 2018.
- The NLSY97 includes about 9,000 participants born between 1980 and 1984. Participants were aged 12 to 17 years old when first interviewed in 1997. This cohort was interviewed in 18 total rounds, with the most recent round being in 2018.
Both cohorts completed the Armed Forces Qualification Test (AFQT) during the first interview. The AFQT is comprised of the following subsections of the Armed Services Vocational Aptitude Battery (ASVAB): mathematics knowledge, arithmetic reasoning, paragraph comprehension, and word knowledge. Now, the AFQT is technically not a cognitive ability test because it was not constructed with the goal of measuring cognitive ability or intelligence. However, AFQT test scores correlate very highly with other conventional IQ test scores, such as the Wechsler Adult Intelligence Scale (WAIS) (source [archived]) (r=0.8), even more than IQ tests often correlate with themselves. Thus, as Smith et al. (2000) note, “Although the AFQT is not an intelligence test, per se, it is highly correlated with IQ and is a widely used measure of adult aptitude” (page 812). The abilities measured by the AFQT test also tests many of the broad abilities identified by the Cattell–Horn–Carroll theory of human cognitive abilities, which is the definition of cognitive ability that I’m using for the purposes of my post. Furthermore, the AFQT is “one of the most highly g-loaded tests in use (g refers to the general intelligence factor)” (Dickens and Flynn 2006 [archived], page 3). For these reasons, AFQT tests are often used as proxies for cognitive ability (e.g., Heckman et al. 2005, page 83; see Gjerde 2002, page 449; Shearer et al. 2002, page 238; Trevor 2001, page 628; Han et al. 2004, page 338). In line with existing literature, I also treat AFQT tests as proxies for cognitive ability.
Findings from The Bell Curve
The clearest presentation of data on the relation between racial disparities in life outcomes and racial disparities in cognitive ability was presented in The Bell Curve by Murray and Herrnstein (1994) [archived]. While there is significant controversy about the book’s political claims (regarding, e.g., affirmative action, welfare) and the book’s causal claims (regarding the cause of the black-white cognitive ability gap, the influence of family environment, etc.), I’m only referencing in this post the book’s statistical claims regarding the relationship between the black-white cognitive ability gap and racial inequalities. Again, this post is agnostic with respect to the cause of the black-white cognitive ability gap. I only wish to show the influence of the cognitive ability gap on racial inequalities.
This book used data from the NLSY79 to show that youth AFQT scores explains much of the black-white disparity on a wide variety of social outcomes (see chapter 14). In fact, many substantial gaps in important social outcomes completely disappeared or even reversed after controlling for these test scores. For example, the data indicates that, controlling only for age, mean black wages were about 80% of mean white wages in 1989. Adding controls for education raised this percentage to only 82%. Adding controls for education and parental SES raised this percentage to only 86% (parental SES was based on “information about the education, occupations, and income of the parents of NLSY youths”, page 131). However, adding controls for only youth IQ (as measured by AFQT scores) raised the percentage to 98%, effectively eliminating the gap (page 324):
In other words, about 90% of the wage gap is eliminated after controlling for youth test scores (the disparity reduces from 20% to 2%). In fact, there were a number of industries where mean black wages were substantially higher than mean white wages after controlling for youth ability. If we measure the black-white disparity in absolute terms, we see similar findings. Controlling only for age, mean wages for black workers was $6,378 less than the mean wages for white workers ($27,372 vs $20,994). After including controls for IQ, mean wages for black workers was only only $545 less than the mean wages for white workers ($25,546 vs $25,001). So 91% of the wage gap was eliminated after controlling for youth IQ scores (the disparity reduces from $6,378 to $545).
Data for other outcomes also show that substantial portions of the black-white disparity are eliminated after controlling for youth cognitive ability. Chapter 14 goes through a lot of this data, but all of the findings are summarized nicely in Appendix 6, where table 3 shows racial disparities in various outcomes before and after controlling for IQ (page 650):
Take poverty as one example. The data shows that, controlling only for age, the black poverty rate is about 18.4 percentage points greater than the white poverty rate (25.6% vs 7.2%). After adding controls for youth IQ, this differential decreases to 4.1 percentage points (10.6% vs 6.5%), showing that youth IQ can (statistically) explain 78% of the black-white disparity in poverty rate. Using similar calculations for some of the other outcomes in that table shows that the majority of many racial inequalities can be explained by youth IQ. I’ve included the same calculations in the following table.
|Outcome||Black-white disparity with age controls||Black-white disparity with age and IQ controls||Percent of disparity reduced after adding IQ controls|
|% High school dropout||+7.4||–2.5||134%|
|% Bachelor’s degree||-15.5||+13.5||187%|
|% High-IQ occupation||-2.3||+2.8||222%|
|% Adult poverty||+18.4||+4.1||78%|
|% Unemployment (males)||+10.4||+4.7||55%|
|% Incarceration (males)||+10.7||+2.7||75%|
|% Child poverty||+44.6||+7.9||82%|
- Black-white disparities are measured by calculating (black outcome) – (white outcome).
- “High-IQ” occupations include lawyers, physicians, dentists, engineers, college teachers, accountants, architects, chemists, computer scientists, mathematicians, natural scientists, and social scientists.
Thus, essentially all of the disparity in socioeconomic status was eliminated after controlling for cognitive ability measured at youth: controlling for youth IQ eliminated all of the disparity in educational attainment and occupation status, and over 90% of the disparity in wages. These test scores also eliminated about 80% of racial disparities in poverty. Only about half of the unemployment disparity was eliminated after controlling for youth IQ scores. Possible explanations for this residual disparity are explained in some of the following studies mentioned in this post.
The findings in this table summarizes the remaining studies cited in this section. Most of the remaining studies report similar findings using more recent data from the same dataset, although some of the studies report data from other datasets. These studies are beneficial because they replicate the findings reported in The Bell Curve, which is important since many people are skeptical of the book’s claims due to its controversy. In the rest of this section, I will cite data that replicate and extend Herrnstein and Murray’s core finding regarding cognitive ability and racial inequalities: youth cognitive ability explains the majority of black-white disparities in educational attainment, income, income mobility, occupational status, and incarceration. I start with black-white disparities in educational attainment.
Cameron and Heckman (2001) [archived] used data from the NLSY79 to examine racial differences in schooling attainment. The researchers analyzed how racial differences in schooling attainment were influenced by a number of different factors, including family income, age adjusted AFQT scores when students were 15-24 years of age, family background variables (number of siblings, parental education, broken home, urban residence, and southern residence). Four measures of schooling attainment were measured: the probability of being in grade 9 or higher at age 15, high school completion by age 24, college entry by age 24 conditional on high school graduation, and college entry by age 24 *unconditional* on high school graduation.
The results showed that controlling for AFQT scores eliminated more of the black-white disparity in schooling attainment than did any other variable. In fact, controlling for AFQT scores alone eliminated the black-white disparities for each measure of schooling attainment. For example, the researchers note that “Regardless of income and family background, at the same AFQT level, blacks and Hispanics enter college at rates that are substantially higher than the white rate. The predictions for high school completion are similarly dramatic” (page 486). Consider the following results (Table 4):
- The black-white gap in the probability of high school completion at age 24 was 6 percentage points (row 9 of panel B).. If the AFQT scores for blacks were equalized to the scores for whites, the probability of high school completion for blacks would increase by 11 percentage points (row 5 of Panel B), accounting for the entire black-white gap.
- The black-white gap in the probability of college entry (conditional on high school completion) at age 24 was 11 percentage points (row 9 of panel C). If the AFQT scores for blacks were equalized to the scores for whites, the probability of college entry for blacks would increase by 15 percentage points (row 5 of panel C), accounting for the entire black-white gap.
This data led the researchers to state the following (page 486):
Regardless of income and family background, at the same AFQT level, blacks and Hispanics enter college at rates that are substantially *higher* than the white rate. The predictions for high school completion are similarly dramatic. The role of AFQT in explaining racial and ethnic schooling differences is thus seen to be very important. It is long-run factors that promote scholastic ability that explain most of the measured gaps in schooling attainment, and not the short-run credit constraints faced by students of college-going age that receive most of the attention in popular policy discussions. The long-run factors that promote college readiness are proxied by AFQT. Even if we exclude AFQT from the analysis, parental background factors play essentially the same role as AFQT, although the effects are weaker.
This study did not report data on rates of college completion, so could not fully corroborate Herrnstein and Murray’s finding that the black-white gap in educational attainment disappears when one controls for AFQT scores. Fortunately, this finding was corroborated by Lang and Manove (2006) [archived]. Using data on highest grade completed as of 2000 for subjects from the NLSY79, they find that blacks average about “three-quarters of a year less education than do whites”, but “conditional on AFQT, blacks get more education than whites do” (page 3). In fact, they find that “Black men get about 1.2 years more education than do white men with the same AFQT. Among women the difference is about 1.3 years” (page 4). The following figures show that black men and women attain higher levels of education across all AFQT scores:
Similar findings were reported in the introductory chapter of The Black-White Test Score Gap. Here, Jencks and Phillips (1998) [archived] further corroborate Herrnstein and Murray’s finding that gaps in college graduation are eliminated after controlling for high school test scores. The benefit of this study is that it uses a different dataset, The High School and Beyond Survey. This survey tested 12th graders in 1982 and followed up with them in 1992 when they were in their late twenties. Test scores were based on the sum of vocabulary, reading, and math scores. At the followup, there was a substantial gap in college graduation rates: the study found that “only 13.3 percent of the blacks had earned a B.A., compared with 30 percent of the non-Hispanic whites” (page 7). However, this gap is completely eliminated after controlling for 12th grade test scores. In fact, the gap reverses: “Once we equalize test scores, High School and Beyond blacks’ 16.7 point disadvantage in college graduation rates turns into a 5.9 point advantage” (page 7). Figure 1-4 shows the disparity in college graduation rates after controlling for socioeconomic status and test scores:
Similar analyses were performed by Aughinbaugh (2008) [archived] using a newer dataset – the NLSY97 – to obtain mostly similar results. This study aimed to determine the predictors of the following outcomes for participants at age 20: college attendance, the type of college attended (2-year vs 4-year), and college retention. The predictor variables under consideration were race/ethnicity, sex, family background (parental education, family income, mother’s age at first birth, and whether the respondent lived with both parents at age 12), high school grades, and performance on the math-language score on the ASVAB. The study found that high school grades and ASVAB scores explained most of the black-white disparity in educational outcomes:
- College attendance (Table 4). After controlling only for sex, black respondents were 15 percentage points less likely than whites to attend college. After controlling only for family background, they were equally likely to attend college. By contrast, after controlling only for high school grades and ASVAB scores, blacks were about 5 percentage points more likely than whites to attend college. After controlling for family background and high school grades and ASVAB scores, blacks were 11 percentage points more likely than whites to attend college.
- College retention (Table 7). After controlling only for sex, black respondents were 9 percentage points less likely than whites to remain at a 4-year college for at least 12 months. After controlling only for family background, they were only 3.5 percentage points less likely (although the difference was not statistically significant). After controlling only for high school grades and ASVAB scores, blacks were about 1.8 percentage points less likely (although the difference was not statistically significant). After controlling for family background and high school grades and ASVAB scores, blacks were about 1 percentage point more likely than whites to remain at a 4-year college for at least 12 months (although the difference was not statistically significant).
Unfortunately, this study did not analyze the effects of high school grades and ASVAB scores separately. This makes it impossible to quantify the independent effects of cognitive ability because high school grades are substantially influenced by cognitive ability (Roth et al. 2015) and independent behavioral and personality traits such as conscientiousness (Poropat 2009). However, there is some reason to suggest that much of the gap in high school grades between blacks and whites may be explained by differences in cognitive ability. For example, Cucina et al. 2016 found that general mental ability (g) was a significant predictor of grades for both blacks and whites, and that blacks and whites with similar g-scores had similar grades (see Figure 3):
Thus, it’s possible that the effects of racial disparities in high school grades are actually downstream of racial disparities in cognitive ability.
The seminal work on the relationship between youth test scores and racial disparities in income was conducted by Johnson and Neal (1998) [archived] (also see Neal and Johnson 1996 for an earlier paper reporting similar findings using a similar methodology; see O’Neill 1990 for similar findings reported for earlier data). These authors examined the role of premarket skills (measured using AFQT scores) in explaining income disparities between blacks and whites. The study focused solely on respondents born between 1962 and 1964 because these respondents completed the AFQT before turning 19 years old. Consistent with prior data on racial gaps in cognitive ability, the data showed that the ” racial difference in mean scores is roughly one standard deviation for both men and women” (page 3). The study measured racial disparities in two measures of income:
- Hourly wages: Hourly wages were measured by averaging the inflation-adjusted wages from 1990 to 1993 for everyone who worked at any point during this period (page 3).
- Annual earnings: Annual earnings were measured by averaging the inflation-adjusted annual earnings from 1990 to 1992 for everyone who reported any earnings during this period (page 6).
Since respondents were born between 1962 and 1964, both measures of income were based on outcomes when respondents were in their late 20s.
Regarding wages, the vast majority of the gap was eliminated after controlling for AFQT scores (page 4):
Table 1 examines some of the determinants of our measure of wage rates. Columns (1) and (3) estimate the racial gap in wages for men and women, controlling only for workers’ age. Among men, for example, the mean of the log of wages is –.277 lower for blacks than whites. This difference implies that black men earn 24 percent less per hour than white men. For women, the −.183 log wage gap implies that black women earn 17 percent less per hour than white women of the same age. Controlling for AFQT completely changes these residual wage gaps (see columns (2) and (4)). For men, the wage gap narrows by roughly two thirds, to about 9 percent. For women, the gap is actually reversed. Black women earn five percent more per hour than white women with the same AFQT score.
In other words, the wage gap reduces from 24% to 9% for men after controlling for AFQT. For young women, black women go from a 17% disadvantage to a 5% advantage after controlling for AFQT scores (page 4).
The authors note also that the AFQT is not merely a proxy for race (page 5):
AFQT is not simply a proxy for race in these regressions. Table 2 shows the relationship between AFQT and wages for blacks, whites and Hispanics separately. Columns (2) and (5) of Table 2 show that AFQT has as large an effect on wages within the black population as it has in the whole population. Basic cognitive skills, as measured by AFQT, raise the wages of blacks at least as much as they raise the wages of whites. In short, basic skills do influence wages and a large fraction of the black-white wage gap reflects a skill gap that pre-dates labor market entry.
In fact, more recent studies of the same dataset report that AFQT gains have a greater impact on wages for black workers than white workers. For example, Lin et al (2016) report that “at age 28, a 0.1 standard deviation rise in AFQT predicts a 1.4 percent increase in white labor market incomes, compared to 2.4 percent for Hispanics and 3.0 percent for blacks” (page 27).
In addition to wages, Johnson and Neal also report the impact of AFQT scores on racial gaps in annual earnings. The analyses were presented separately for men and women. For women, the racial earnings gap was completely reversed after controlling for AFQT scores (page 6):
Table 3 shows how women’s age and AFQT scores affect their earnings. The earnings measure is the log of average annual inflation-adjusted earnings from 1990 to 1992 for everyone who reported any earnings during this period. Black women, on average, enjoy a substantial earnings advantage over white women with similar AFQT scores. Although columns (3) and (4) show that the earnings gap is smaller among highly skilled women, predicted earnings for black women remain above predicted earnings for their white counterparts over almost the entire range of black AFQT scores.
The researchers consider a number of explanations for why black women have far greater AFQT-adjusted earnings than white women, including differences in labor supply and differences in marital patterns (pages 6-9).
For men, the unadjusted racial earnings gap was far greater than the unadjusted racial wage gap (48% vs 24% gap). After controlling for AFQT, only half of the earnings gap was eliminated (page 10):
Column (1) of Table 5, parallel to Table 1’s analysis of wages, presents the differences in log earnings between black, white and Hispanic men, controlling only for age. A comparison of Tables 1 and 5 shows that the log earnings gap between black and white men is over twice as large as the wage gap. Black men earn 48% less per year than whites of the same age, even though their wages are only 24% lower. When we control for AFQT in column (2), the earnings gap between black and white men is cut in half. Consequently, while premarket skills explain a significant part of the earnings gap, they account for a smaller fraction of the earnings gap than of the wage gap.
The researchers consider a number of explanations for why black men have far lower AFQT-adjusted earnings than white men, including differences in labor supply, differences in job experience, and differences in returns to education (pages 9-16).
So AFQT scores alone can explain the entirety of racial income disparities for women and most of the income disparities for men. Interestingly, the unexplained income gap for men (i.e. the residual gap after controlling for AFQT) was isolated to men with lower levels of education and ability. For example, the study reports that “Despite the large overall earnings gap, among 27 year old men with AFQT scores more than one standard deviation above the sample mean, blacks earn only about 5% less than white men” (page 11). In fact, among male college graduates, black workers earn higher wages than white workers with similar AFQT scores (page 15):
Table 8 examines the effect of experience on wages. Columns (1), (3) and (5) show regressions of log wages on race, age, and AFQT. Each regression is restricted to workers with the same amount of education. The coefficients on the black dummy variable in these regressions describe the skill-adjusted wage gap between blacks and whites in a given education category. As one would expect from Table 6, these gaps decline with educational attainment. In fact, black male college graduates in this sample earn higher wages than white male graduates with similar AFQT scores, though the difference is not statistically significant. However, even holding AFQT constant, black high school dropouts and graduates earn substantially lower wages than their white counterparts.
Much of the unexplained wage gap in wages for the non-college-educated is explained by differences in job experience (page 16):
Columns (2), (4) and (6) show that prior work experience is strongly associated with wages for each of the education groups, but especially for the two non-college groups. For dropouts and high school graduates, each additional year of work experience adds roughly 5 percent to the wage rate. For high school graduates, roughly half the unexplained black-white wage gap can be attributed to differences in past work experience (compare the race coefficients in columns (3) and (4)). For dropouts (columns (1) and (2)), experience explains about 30 percent of the remaining black-white gap.
In summary, this study finds that all of the racial gaps in hourly wages and annual earnings for women are eliminated after controlling for AFQT scores. That is, black women have higher hourly wages and higher annual earnings than white women after adding such controls. Regarding men, about two-thirds of the hourly wage gap (gap reduces from 24% to 10%), and one-half of the annual earnings gap is explained by AFQT scores. Most of this residual wage disparity for men was isolated to those with lower ability and education. For example, whereas college-educated blacks had higher wages than similarly educated whites with similar AFQT scores, blacks without a college degree had lower wages than similarly educated whites even after controlling for AFQT scores. Much of this residual disparity is explained by differences in employment. It should be noted that a review by Farkas and Vicknair (1996) showed that the entirety of the wage gap between black and white male workers could be explained by AFQT scores and a host of other predictors (Table 1). The statistically significant predictors were AFQT scores (explains 40% of the gap), work experience (22%), region (20%), marital status (13%), and years of school (10%):
Fryer (2010) [archived] reported updated findings on racial wage gaps after controlling for AFQT scores using data from the NLSY79 and NLSY97. Wage gaps were reported for wages observed in 2006 and 2007. The NLSY79 cohort was between 42 and 44 years old and the NLSY97 cohort was between 21 and 27 years (page 3).
Data from the NLSY79 showed similar findings that were reported by Johnson and Neal 15 years earlier. That is, controlling for AFQT scores reverses the wage gap for women and eliminates the majority of the wage gap for men (page 4):
Table 1 presents racial disparities in wage and unemployment for men and women, separately. The odd-numbered columns present racial differences on our set of outcomes controlling only for age. The even-numbered columns add controls for the Armed Forces Qualifying Test (AFQT) – a measure of educational achievement that has been shown to be racially unbiased (Wigdor and Green, 1991) – and its square. Black men earn 39.4 percent less than white men; black women earn 13.1 percent less than white women. Accounting for educational achievement drastically reduces these inequalities – 39.4 percent to 10.9 percent for black men and 13.1 percent lower than whites to 12.7 percent higher for black women. An eleven percent difference between white and black men with similar educational achievement is a large and important number, but a small fraction of the original gap.
Fryer also reports data on unemployment rates (page 4):
Labor force participation follows a similar pattern. Black men are more than twice as likely to be unemployed in the raw data and thirty percent more likely after controlling for AFQT. For women, these differences are 3.8 and 2.9 times more likely, respectively
Table 1 shows that 72% and 197% of the racial wage gap for men and women, respectively, was reduced after controlling for AFQT scores in the NLSY79 data. The same controls accounted for 75% and 32% of the racial unemployment gap for men and women, respectively.
Data from the NLSY97 indicated similar patterns as the NLSY79. One large difference is that the raw wage gap for men in the new dataset (NLSY97) is much smaller than the raw gap in the older dataset (NLSY79). The adjusted wage gap for men is about the same as in the NLSY79. As a result, AFQT scores explain a smaller percentage of the NLSY97 wage gap compared to the NLSY79 wage gap (page 4):
Table 2 replicates Table 1 using the NLSY97.7 The NLSY97 includes 8,984 youths between the ages of 12 and 16 at the beginning of 1997; these individuals are 21 to 27 years old in 2006-2007, the most recent years for which wage measures are available. In this sample, black men earn 17.9 percent less than white men and black women earn 15.3 percent less than white women. When we account for educational achievement, racial differences in wages measured in the NLSY97 are strikingly similar to those measured in NLSY79 – 10.9 percent for black men and 4.4 percent for black women. The raw gaps, however, are much smaller in the NLSY97, which could be due either to the younger age of the workers and a steeper trajectory for white males (Farber and Gibbons, 1996) or to real gains made by blacks in recent years.
Data on unemployment was also reported (page 4):
Black men in the NLSY97 are almost three times as likely to be unemployed, which reduces to twice as likely when we account for educational achievement. Black women are roughly two and a half times more likely to be unemployed than white women, but controlling for AFQT reduces this gap to seventy-five percent more likely.
Table 2 shows that 39% and 71% of the racial wage gap for men and women, respectively, was reduced after controlling for AFQT scores in the NLSY97 data. The same controls accounted for 41% and 52% of the racial unemployment gap for men and women, respectively.
Fryer also reported data on racial gaps in annual income using the College and Beyond (C&B) database, which includes data on over 90,000 students who entered 34 elite colleges and universities in 1951, 1976, or 1989. Fryer reports data on annual income for the 1976 cohort after controlling for SAT scores. Annual income was recorded in 1995 or 1996 when respondents were approximately 38 years old. The results for males were similar to that of Johnson and Neal: a large initial gap that was halved after controlling for test scores. The results for females were slightly different: black females outearned white females after controlling for SAT scores, but they also outearned white females prior to such controls:
Table 4 presents racial disparities in income for men and women from the 1976 cohort of the C&B Database. The odd-numbered columns present raw racial differences. The even-numbered columns add controls for performance on the SAT and its square. Black men from this sample earn 27.3 percent less than white men, but when we account for educational achievement, the gap shrinks to 15.2 percent. Black women earn more than white women by 18.6 percent, which increases to an advantage of 28.6 percent when accounting for SAT scores.
The difference between the results from the C&B database and the data from Johnson and Neal regarding annual income gaps can likely be attributed to the fact that the C&B database includes data on students from elite universities, whereas Johnson and Neal included data on a nationally representative sample.
The finding that the ability-adjusted racial wage gaps are larger at lower levels of the ability and education distribution were also reported by studies specifically focused on investigating this issue. Consider the following examples:
- In the introductory chapter of The Black-White Test Score Gap, Jencks and Phillips (1998) [archived] reported that the racial gap in annual earnings among males was smallest among those with the highest level of ability. They reported that, from 1962 to 1993, “among men who scored between the 30th and 49th percentiles nationally, black earnings rose from 62 to 84 percent of the white average. Among men who scored above the 50th percentile, black earnings rose from 65 to 96 percent of the white average” (page 6).
- O’Neilla et al. (2006) [archived] used the NLSY79 to analyze racial wage gaps in 1993. Replicating the findings from Neal and Johnson, they find that “the average log wage differential falls from .26 to .06” after controlling for AFQT scores (page 348), implying that test scores explain about 77% of the gap. These authors go further than Neal and Johnson in that they report the proportion of the racial wage gap due to AFQT scores at different wage levels in the wage distribution. They found that AFQT scores explained about 40-50% of the racial wage gap among those with wages below the 40th percentile, 60-80% of the gap among those with wages between the 40th and 60th percentile, and 80-90% of the gap among those with wages above the 60th percentile (Table 3).
- Arcidiacono et al. (2009) [archived] used the NLSY79 to show that, among those with only a high school education, “blacks earn wages that are about 6 percent lower than those received by whites with the same AFQT score at the time of initial entry into the labor market. This gap increases (insignificantly) with labor market experiences so that the estimated racial wage gap at ten years of experience, conditional on AFQT, is 10 percent” (page 10). By contrast, the same dataset shows that “conditional on AFQT, college-educated blacks earn 11 percent higher wages than their white counterparts upon initial entry into the labor market. This premium declines to zero after 9 years of labor market experience” (page 12). In other words, the study finds “no racial differences in wages or returns to ability in the college labor market, but a 6-10 percent wage penalty for blacks (conditional on ability) in the high school market.”
Here’s Table 3 from O’Neilla et al. (2006):
To put the influence of test scores into perspective, we compare the size of the wage gap after controlling for AFQT scores to the size of the wage gap after controlling for other covariates. I presented such an analysis by Herrnstein and Murray earlier in the post, which shows that test scores explained a far greater portion of the wage disparity than did parental SES and educational attainment. Jackson and VanderWeele (2018) [archived] extended these findings using data from the NLSY79 for wages observed in 2006 and 2007 (the same data used by Fryer). These researchers estimated the percentage of the male wage gap that is expected to reduce under hypothetical interventions to equalize various predictors of the racial wage gap. The study compared the hypothetical gap reductions after equalizing the following three variables: AFQT scores, childhood SES, and total years of education. Childhood SES was measured as a composite including maternal education, household income, and poverty status. The results were as follows (Table 2 and Table 3):
- The gap would reduce by 26% by equalizing the distribution of childhood SES.
- The gap would reduce by 27% by equalizing the distribution of total years of education.
- The gap would reduce by 66% by equalizing the distribution of AFQT scores.
- The gap would reduce by 37% by equalizing the distribution of total years of education and childhood SES.
- The gap would reduce by 74% by equalizing the distribution of both AFQT scores and childhood SES.
Keep in mind these results are specifically for men, where racial wage gaps tend to be greater.
Perhaps the most surprising findings come from Carruthers and Wanamaker (2017) [archived]. These authors studied the relationship between the black and white earnings gap and human capital differences among Southern workers in 1940. Data was used from the Public Use Sample of the 1940 Census, the earliest year that includes individual earnings data. They then “utilize a known oddity in the World War II enlistment records to impute Army General Classification Test (AGCT) scores” as a measure of human capital. Prior to controlling for AGCT scores, the authors find an unconditional log wage gap between blacks and whites of .53 and .51 log points for weekly and annual wages, respectively. However, after controlling for AGCT scores, these wage gaps are dramatically reduced:
Results for weekly and annual wages are located in Table 7. The black-white wage gap falls substantially when we control for estimated ability. The weekly wage gap falls to 12.5–14.6 log points, depending on the imputation method. For annual wages, residual gaps range from 1.1 to 11.1 log points.
This implies that the weekly gap and annual gap in earnings reduced by about 72-76% and 78-97%, respectively, depending on the imputation method. The authors state the same findings in their conclusion:
Recent labor market studies have highlighted the importance of human capital in explaining the black-white wage gap. We ask the same question for 1940 workers: How far can human capital inequalities go in explaining the large pre-war racial wage disparity? Incorporating new data on race-specific school quality in 10 southern states, we document a predominant role of school quality and educational attainment in determining wage inequality for young men. Human capital accounts for 73% of the gap in annual wages and 64% of the gap in weekly wage rates. Once we control for estimated AGCT scores imputed from World War II enlistment records, human capital accounts for up to 97% of the gap in annual wages and 80% of the weekly wage gap.
Recall the Pew report by Mazumder (2008) [archived] that used the NLSY79 to examine factors relevant to income mobility. The report found stark racial disparities in income mobility. For example, about 75% of whites raised in the bottom income quintile eventually transition out of that quintile, whereas only 56% of blacks do the same. However, among those with a median AFQT score, there is almost no difference in the likelihood of transitioning out of the bottom quintile: 81% for whites and 78% for blacks with median AFQT scores achieve this feat (page 30). The data suggests that “test scores can explain virtually the entire black-white mobility gap” (page 30). See the following commentary:
Figure 13 plots the transition rates against percentiles of the AFQT test score distribution. The upward-sloping lines indicate that, as might be expected, individuals with higher test scores are much more likely to leave the bottom income quintile. For example, for whites, moving from the first percentile of the AFQT distribution to the median roughly doubles the likelihood from 42 percent to 81 percent. The comparable increase for blacks is even more dramatic, rising from 33 percent to 78 percent. Perhaps the most stunning finding is that once one accounts for the AFQT score, the entire racial gap in mobility is eliminated for a broad portion of the distribution. At the very bottom and in the top half of the distribution a small gap remains, but it is not statistically significant. The differences in the top half of the AFQT distribution are particularly misleading because there are very few blacks in the NLSY with AFQT scores this high.
By contrast, controlling for years of educational attainment left large residual gaps in income mobility. The report notes that “years of completed schooling explains little of the black and white economic mobility gap” (page 31). The following differences in mobility are reported for blacks and whites with the same amount of schooling:
Controlling for years of education, the black-white economic mobility gap at lower levels of education is not much smaller than it was without controlling for years of schooling, as indicated by the fact that the gap between the two lines through 12 years of schooling is nearly as wide as the overall gap between blacks and whites. For example, the transition rate for whites with 10 years of schooling is 65 percent and is substantially higher than the comparable figure for blacks, 39 percent. This 26 percentage-point gap narrows to 16 percentage points for those who complete 12 years of schooling. For those who have completed college (16 years) the gap is just four percentage points and is no longer statistically significant. Overall, this suggests that years of completed schooling do not fully reflect the skills gap as captured by test scores.
Similar data was reported in Bhattacharya and Mazumder (2011), Mazumder (2012), Mazumder (2014), and Davis and Mazumder (2018). For example, Mazumder (2014) reports that the proportion of whites who move out of the bottom quintile is 27 percentage points higher than blacks who do the same. However, for those with median AFQT scores at adolescence, the gap reduces from 27 percentage points to just 5.2 percentage points, which suggests that “cognitive skills measured at adolescence can account for much of the black–white difference in upward mobility” (page 12). Controlling for AFQT scores more consistently accounted for large portions of racial differences in economic mobility than did controlling for years of schooling or family structure (Figure 8):
In the summary of the study, the authors notes the following (page 2):
This study also tries to shed light on which factors are associated with the racial gaps in upward and downward mobility. To be clear, while the analysis is descriptive and not causal, it nonetheless provides some highly suggestive “first-order” clues for the underlying mechanisms leading to black–white differences in intergenerational mobility. It appears that cognitive skills during adolescence, as measured by scores on the Armed Forces Qualification Test (AFQT), are strongly associated with these gaps. For example, conditional on having the median AFQT score, the racial gaps in both upward and downward mobility are relatively small. Consistent with previous studies linking AFQT scores to racial differences in adult outcomes (for example, Neal and Johnson, 1996; Cameron and Heckman, 2001), I do not interpret these scores as measuring innate endowments but rather as reflecting the accumulated differences in family background and other influences that are manifested in test scores. If these results are given a causal interpretation, they suggest that actions that reduce the racial gap in test scores could also reduce the racial gap in intergenerational mobility
Nyborg and Jensen (2001) [archived] examined income and occupational status among large samples of white and black American armed forces veterans. The data set contained 4,462 males obtained from the Centers for Disease Control (CDC). The CDC obtained data on these veterans 17-18 years after induction in order to assess the long-term effects of the veterans’ military service. The average age when the participants were tested by the CDC was 37.4 years. Cognitive ability was measured by extracting the general factor (g) from a psychometric battery comprising several tests of a diverse range of abilities, information content, and cognitive skills such as visual-spatial ability, verbal reasoning, general information, concept formation, etc. Income was measured based on total household income for the calendar year preceding the study interview. Occupational status was classified using the three-digit code for occupations used by the U.S. Census; the index ranks 503 occupations based on “typical requirements for education, complexity of the job’s cognitive demands, responsibility entailed, and typical salary” (page 48). Typical high-status occupations are top-level managerial and professional workers. Typical low-status occupations are semiskilled and unskilled laborers.
The racial gap in g scores was 1.3 standard deviations (Table 2). The data showed significant correlations between g scores and socioeconomic outcomes (education, occupation, and income) within both the black and white samples (Table 3). g remained correlated with occupational status and income even after controlling for education (Table 4). The unadjusted racial gaps in income and occupational index were 0.48 and 0.28 standard deviations, respectively (Table 2). After controlling for g factor scores, blacks achieved a higher mean occupational status than whites at every g factor percentile (Figure 2), and blacks achieved earned similar or higher incomes than whites at g factor percentiles above the 40th percentile (Figure 1):
One limitation of this finding is that many of the cognitive ability tests were conducted 17 years after induction into the armed forces. The other studies cited above show that racial inequalities in outcomes are reduced after controlling for prior measures of test scores, whereas this study allows for the possibility that racial differences in socioeconomic status predict racial differences in g scores rather than the other way around. However, this hypothesis is unlikely given the high stability of cognitive ability after late adolescence (source). In fact, g scores are particularly stable as they are less susceptible to environmental gains than general IQ scores (e.g., see Ritchie et al. 2015).
Bjerk (2007) [archived] used data from the NLSY79 to investigate the black-white wage gap across different occupational sectors. Consistent with prior studies, the study shows that youth AFQT scores accounted for the majority of the wage gap. After controlling only for age, black workers earned about 28% less than white workers. After adding controls for region and AFQT scores, “the black-white wage gap falls to only an 8 percent differential” (page 402). In other words, the racial gap in AFQT scores explained “over two-thirds of the unconditional racial wage gap” (page 402). By contrast, controlling for years of education only reduced the black-white wage gap from 28% to 22% (Table 1):
Also consistent with prior studies, the study showed that the adjusted racial wage gap was greatest at lower skilled occupational sectors. Consider the following findings:
- In white collar and service jobs, black wages were about 17% and 24%, respectively, lower than white wages (Tables 3 and 5). After controlling for age, AFQT scores, years of education, region, and parent’s education, there was no statistically significant wage gap. White collar jobs include professional and managerial jobs. Service jobs included janitors, cooks, waiters, etc. (Table 2).
- In blue collar and sales jobs, black wages were about 24% and 32%, respectively, lower than white wages (see Tables 4 and 5). After controlling for age, AFQT scores, years of education, region, and parent’s education, the gap reduced to about 12 percent (explaining 50% of the gap) and 18 percent, (explaining 44% of the gap), respectively. Blue collar jobs included clerical workers, craftsmen, operatives, and laborers. (Table 2).
These results suggested that “all of the residual wage inequality that remains after controlling for measured academic skills” arises from “racial wage inequality between workers of similar measured academic skill working in blue-collar and sales jobs” (page 409). On the other hand, accounting for academic skills “can account for all of the racial wage inequality among workers working in white-collar and service jobs” (page 409).
Another interesting finding is that black workers are more likely to work in a white collar job after adding controls for AFQT scores. For example, the study found that “the unconditional probability that a black worker works in the white-collar sector is almost 50 percent less than the corresponding probability for whites” (page 413). However, “after controlling for where an individual lies in the AFQT distribution, black workers appear to be equally or more likely than white workers to work in the more academically skill-intensive white-collar sector” (page 416). Importantly, the study also finds that controlling for parental education, family structure, and parental occupational attainment is not sufficient to account for racial differences in white collar job attainment. One needs to include controls for AFQT scores in order to account for the gap (page 416):
The results confirm that without conditioning on the academic skill of each worker, black workers are significantly less likely to work in the more highly skill-intensive white-collar sector than white workers. The second specification shows that while the gap shrinks somewhat, black workers are still less likely than white workers to work in the white-collar sector if we further control for parental education, whether each parent worked in a professional occupation, whether the respondent lived with both parents at age 14, and the region in which the respondent resides. However, the third specification shows that if we additionally control for premarket academic skills via AFQT scores, black workers are actually significantly more likely to work in the more highly skill-intensive white-collar job sector than their white counterparts. To give an indication of the magnitude of the coefficient estimates, if all other characteristics are held fixed at the population means, the results in Specification 3 imply that a black worker with an AFQT score one standard deviation above the population mean is about 30 percent more likely to work in the white-collar sector than a white worker with the same AFQT score.
The first point to note here is that cognitive ability predicts incarceration within both the black and the white population. For example, Beaver (2013) examined a nationally representative sample of American youths from the National Longitudinal Study of Adolescent Health to study the relationship between IQ and criminal justice outcomes. The results of the study indicated a “consistent association between IQ scores and the odds of arrest and incarceration for all males and for Black males” (page 282). Given this, it’s worthwhile to examine if controlling for test scores can account for racial disparities in incarceration.
Let’s return to Fryer (2010) [archived]. In addition to data on racial gaps in wages, he also reported data on the relationship between AFQT scores and racial gaps in incarceration. He found that controlling for AFQT scores in the NLSY79 cohort explained a substantial portion of racial gaps in incarceration (page 5):
Adjusting for age, black males are about 3.5 times and Hispanics are about 2.5 times more likely to have ever been incarcerated when surveyed. Controlling for AFQT, this is reduced to about 80% more likely for blacks and 50% more likely for Hispanics. Again, the racial differences in incarceration after controlling for achievement is a large and important number that deserves considerable attention in current discussions of racial inequality in the United States. Yet, the importance of educational achievement in the teenage years in explaining racial differences is no less striking.
In other words, controlling for AFQT scores in the NLSY79 accounts for about 69% of the racial disparity in incarceration for males and all of the racial disparity for females. The same pattern was observed in the NLSY97. In this dataset, black males and females are about 2.3 times and 1.2 times, respectively, as likely to be incarcerated as their similarly-aged white counterparts. After adjusting for AFQT scores, black males and females are only about 1.4 and 0.7 times as likely to be incarcerated as their white counterparts. In other words, controlling for AFQT scores in the NLSY97 accounts for about 69% of the racial disparity in incarceration for males and all of the racial disparity for females.
Jackson and VanderWeele (2018) [archived] extended the findings by Fryer by comparing incarceration gap among men after adjusting for AFQT scores to the gap after adjusting for childhood SES and education (Table 2 and Table 3):
- The gap would reduce by 45% by equalizing the distribution of childhood SES.
- The gap would reduce by 13% by equalizing the distribution of total years of education.
- The gap would reduce by 65% by equalizing the distribution of AFQT scores.
- The gap would reduce by 46% by equalizing the distribution of total years of education and childhood SES.
- The gap would reduce by 81% by equalizing the distribution of both AFQT scores and childhood SES.
McNulty et al. (2012) [archived] analyzed data from the NLSY97 to examine the role of verbal ability in explaining black-white differences in adolescent violence. Researchers analyzed a number of variables to predict adolescent violence such as race, basic demographic factors, verbal ability, family income, urbanicity, “neighborhood disadvantage” and a host of other covariates. “Neighborhood disadvantage” was a composite comprising the percentage of the population in poverty, the percentage unemployed, and the percentage of households headed by a female.
The authors constructed a number of regression models to determine the influence of different factors by iterating adding more variables into the model. Model 2 included controls for race, sex, age, urbanicity, whether the subject lived in a single-parent family, and whether the subject reported use of drugs. With these controls, black adolescent were significantly more likely to be involved in violence than white adolescents (page 13):
Model 2 controls for race and gender at Level 2 and age and the additional control variables at Level 1, which establishes the greater involvement of Black adolescents in violence compared with Whites (.509; p < .001). The control variables are also significant in expected directions. Violence event rates are significantly higher among males, among those who reside in urban areas and in single-parent families, and especially among those who are involved with drugs.
Model 3 incorporates control for neighborhood disadvantage, which accounts for about half of the black effect (black coefficient decreases from .509 to .249), thus leaving about half of the association unexplained. Model 4 adds verbal ability to the model, which explains 90% of the remaining black effect (black coefficient decreases from .249 to .026), to the point that the black effect is no longer statistically significant (page 13):
Model 3 incorporates the neighborhood disadvantage index at Level 3, which has a strong, positive effect on violence event rates (.242; p < .001). Controlling for neighborhood disadvantage reduces the Black coefficient substantially by 51% (.509 to .249), although the effect remains significant at the .05 level. This result is consistent with literature that suggests that the disproportionate involvement in violence among Black adolescents is partly confounded with neighborhood disadvantage.
Model 4 adds verbal ability to the equation, which has the hypothesized negative effect on violence (−.380; p < .001). Most important, incorporating verbal ability fully explains the Black effect, reducing the coefficient by an additional 90% (.249 to .026). Thus, contradicting prior research (Bellair & McNulty, 2005; Sampson et al., 2005), individual differences in verbal ability are shown to contribute substantially to explanation of the Black–White disparity in violence. We also argued above that the effect of neighborhood disadvantage on violence may partly operate through verbal ability. Findings in Model 4 provide some support, showing that the effect of neighborhood disadvantage, although remaining significant, is reduced by 32% when verbal ability is added to the equation (.242 to .165).
These models and other models suggest that verbal ability plays a critical role in explaining racial disparities in adolescent violence. The authors interpret this evidence as suggesting that some of the effect of neighborhood disadvantage on adolescent violence is mediated by verbal ability, and that much of the effect of verbal ability on violence is mediated by scholastic attainment. The authors conclude with the following discussion (page 15):
This article integrates an individual difference approach that emphasizes variation in verbal ability with a sociological approach that highlights neighborhood disadvantage, both of which are relevant to explanation of the race difference in violence. Black children are far more likely than their White counterparts to grow up in neighborhoods featuring high rates of structural disadvantage, which has repercussions for the acquisition of verbal skills that are crucial for achievement in school and the labor market. Our results show that low verbal ability and diminished school attainment are criminogenic risk factors that are in part outcomes of exposure to neighborhood disadvantage. Verbal ability partly mediates the effect of disadvantage at the neighborhood level and in turn provides a succinct explanation for the racial disparity in violence.
Although sociological variables also explain the race disparity, verbal ability in conjunction with neighborhood disadvantage reduces the Black– White gap in violence to zero and is thus part of the explanation. This contradicts recent research that has found that verbal ability scores, while related to violence, contribute little to the explanation of race and violence (Bellair & McNulty, 2005; Sampson et al., 2005). Yet that the effect of disadvantage on violence is mediated by verbal ability and school achievement indicates that exposure to neighborhood disadvantage is a critical part of the process underlying the disproportionate involvement in violence among Black adolescents.
The above studies show that most of the disparities in educational attainment, occupational status, income, incarceration, etc. are explained away after controlling for cognitive ability. However, some disparities are not impacted much after incorporating such controls. For example, consider the following findings by Murray and Herrnstein (1994) [archived]:
- Only 20% of the disparity in illegitimacy rates is erased when fixing to persons of average IQ. For 29-year-old mothers, the probability of having a child out of wedlock was 62% for blacks and 12% for whites. For persons of average IQ, the number dropped to 51% for blacks and 10% for whites (page 331).
- Only half of the disparity in welfare usage is erased when fixing to persons of average IQ. For 29-year-old women, the probability of having been on welfare was 49% for blacks and 13% for whites. For persons of average IQ, the number dropped to 30% for blacks and 12% for whites (page 332).
- The disparity in marital rates is almost unaffected when fixing to persons of average IQ. For persons over the age of 30, the probability of being married by age 30 is 54% for blacks and 78% for whites. For persons of average IQ, the number rises to 58% for blacks and 79% for whites (page 329).
It is likely that these disparities are the result of cultural differences regarding marriage. These cultural differences probably also explain some of the residual gap in poverty rates after controlling for cognitive ability (since married households are substantially less likely to be poor since they tend to be dual-income households). Unlike many of the other outcomes presented here, racial differences in marital rates do not seem to vanish after adjusting for cognitive ability.
Conclusion and implications
Causation or confounding?
The data above finds that there are large associations between the cognitive ability gap and racial disparities in important life outcomes, such that racial disparities in those outcomes are significantly reduced (oftentimes eliminated) when racial disparities in cognitive ability are statistically eliminated. In other words, the black-white cognitive ability gap can (statistically) account for much of the racial disparities in life outcomes. The simplest explanation of these statistical findings is that the cognitive ability gap causes much of the racial disparities in important life outcomes.
The alternative explanation is that the cognitive ability gap is not actually causally responsible for these disparities. Perhaps the cognitive ability gap is caused by an omitted third variable (a confounding variable) which is actually causally responsible for the disparities in life outcomes. Thus, it’s possible that the cognitive ability gap has no impact on racial disparities in life outcomes; the associations reported above might be merely spurious relationships produced because both the cognitive ability gap and racial disparities are effects of a common confounding variable. I find this alternative hypothesis unlikely for a number of reasons.
- There is strong evidence that cognitive ability has a causal influence on these outcomes in the general population. In order for the alternative hypothesis to be true, there would need to be an explanation for why this causal influence vanishes when considering racial disparities.
- This proposed confounding variable cannot be commonly proposed explanations of racial disparities such as parental income, parental education, or years of education. This is because the above studies show that these variables do not explain racial disparities as much as cognitive ability does. Furthermore, cognitive ability explains a significant portion of racial disparities even after one controls for those variables.
- For similar reasons, this proposed confounding variable cannot be heavily correlated with covariates such as parental income, parental education, of years of education (if the confounding variable were highly correlated with those covariates, then statistically controlling for those covariates would implicitly also control for the confounding variable, which runs into the same problems as (2)). Thus, the proposed confounding variable would need to (a) correlate highly with cognitive ability but not be caused by cognitive ability, (b) cause racial disparities in important life outcomes, but (c) not be statistically accounted for by parental income, parental education, or years of education. It’s not clear that any theoretically plausible variable could fulfill this role.
For these reasons, until (a) there is empirical evidence of a confounding variable that fulfills the role I described above and (b) there is an explanation for why cognitive ability is causal these outcomes in the general population but is not causal for these outcomes within the context of racial disparities, I believe we should (provisionally) accept the most parsimonious explanation of the findings reported above: cognitive ability is causally responsible for much of the racial disparities outlined in this post.
If I’m right that the cognitive ability gap is causally responsible for many racial disparities in important life outcomes, then I believe this has at least the following important implications:
- Whatever model you prefer to explain racial inequalities, your model should place heavy importance on disparities in cognitive ability. Now, your model might specify many possible variables to explain racial disparities in cognitive ability (e.g., structural racism, economic disparities, cultural differences, parenting practices, genetic differences, schooling differences, etc.). The variables that explain the cognitive ability gap depends on further empirical inquiry (which is beyond the scope of this post) and perhaps one’s theoretical leanings. Regardless, cognitive ability must play a role as a core component in the causal pathway that explains racial inequalities in income, educational attainment, incarceration, etc. No adequate model that hopes to explain racial inequalities can afford to omit a causal pathway that runs from cognitive ability to these racial inequalities.
- Your model to explain racial cognitive disparities and racial inequalities must rely on racial differences that can explain the properties of the cognitive ability gap outlined in this post. This post showed that the cognitive ability gap has persisted for several generations, emerges within the first few years of life before children reach formal schooling, persists at all levels of education, and persists at all levels of parental socioeconomic status. Therefore, your explanation of cognitive differences (and thus racial inequalities) must likewise rely on racial differences that have persisted for several generations, that emerge within the first few years of life, that persist at all levels of education, and that persist at all levels of parental socioeconomic status.
- If your goal is to eliminate racial inequalities in the outcomes mentioned in this post (e.g., income, income mobility, incarceration, etc.), then your plan to realize this goal will probably need to address racial inequalities in cognitive ability in order to be effective. The studies above show that eliminating disparities in parental income, parental education, years of education, etc. is not enough to eliminate disparities in important life outcomes; you probably also need to eliminate disparities in cognitive ability. I say “probably” because it’s possible that your plan could eliminate racial inequalities by improving some other component of black human capital (e.g., so called “non-cognitive” traits such as contentiousness, self-regulation, etc.) in a way that offsets their lower cognitive ability. While this is possible, it seems unlikely that there could be any plan that improves non-cognitive skills to the point that it completely compensates for the large deficiency in cognitive ability. Furthermore, even if such a plan did exist, other groups (e.g., whites, Asians) would likely take advantage of this plan as well; if so, these groups would have comparable levels of non-cognitive skills and higher levels of cognitive ability, causing racial inequalities to persist.
- This is related to the previous point: if your goal is to eliminate racial inequalities in the outcomes mentioned in this post, there should be a strong effort to scientifically study the causes of racial disparities in cognitive ability. In order to solve a problem, we ought to know the causes of that problem. For example, if the racial disparity in cognitive ability were driven entirely by racial differences in exposure to leaded paint, then the obvious implication of this is that we ought to support intensive governmental programs to remove leaded paint from impacted homes. Of course, racial disparities in cognitive ability are probably not driven entirely by a single factor as simple as leaded paint; there’s almost certainly a complex multivariate explanation of racial disparities in cognitive ability. Regardless, the main point still holds: if we better understand the causes of racial disparities in cognitive ability, then we will better understand the causes of racial disparities in important life outcomes (since the former is responsible for the latter), suggesting that we will be better equipped to solve racial disparities in those outcomes.
Other experts have expressed the same concerns outlined here. For example, In a recent review of intelligence research by experts in the field, Nisbett et al. (2012) have made this same point (page 131):
IQ is also important because some group differences are large and predictive of performance in many domains. Much evidence indicates that it would be difficult to overcome racial disadvantage if IQ differences could not be ameliorated. IQ tests help us to track the changes in intelligence of different groups and of entire nations and to measure the impact of interventions intended to improve intelligence.
In The Black-White Test Score Gap, Jencks and Phillips (1998) advanced the same argument (page 3):
In a country as racially polarized as the United States, no single change taken in isolation could possibly eliminate the entire legacy of slavery and Jim Crow or usher in an era of full racial equality. But if racial equality is America’s goal, reducing the black-white test score gap would probably do more to promote this goal than any other strategy that commands broad political support. Reducing the test score gap is probably both necessary and sufficient for substantially reducing racial inequality in educational attainment and earnings. Changes in education and earnings would in turn help reduce racial differences in crime, health, and family structure, although we do not know how large these effects would be.
As stated earlier, to solve a problem, we should understand the cause of the problem. Therefore, assuming that we have a moral imperative to address racial inequalities in important life outcomes, we should work to understand the cause of the black-white cognitive ability gap. I attempt to start this investigation in a later post.
For works to develop a general understanding of cognitive ability, see the sources linked at the bottom of this post. See the works below to understand the implications of the racial cognitive ability gap.
- Neisser et al. (1996). “Intelligence: Knowns and unknowns” [archived]. Inspired by the heated debate following the release of The Bell Curve, the Board of Scientific Affairs of the American Psychological Association established a task force of 11 experts on intelligence to prepare an authoritative report surveying the current state of the field. The report was continually revised and discussed until the report received unanimous support from each member of the task force.
- Linda Gottfredson (2004). “Social Consequences of Group Differences in Cognitive Ability” [archived]. This paper provides a good summary of research regarding the importance and nature of IQ and g. More importantly, it summarizes research on the magnitude of group differences in cognitive ability and the social implications of these differences.
- Sackett et al. (2008). “High-Stakes Testing in Higher Education and Employment” [archived]. The authors defend cognitive ability tests from criticisms against use for employment and higher education admissions decisions.
- Nisbett et al. (2012). “Intelligence: New Findings and Theoretical Developments” [archived]. An authoritative review of the field of intelligence to update the “Intelligence: Knowns and unknowns” (1996) article. This review surveyed many new findings since the publication of the original article.
- Herrnstein and Murray (1994). The Bell Curve. A highly impactful book that ignited much of the debate regarding race and intelligence. Murray and Herrnstein demonstrate the importance of cognitive ability in causing individual and group differences in success across a wide range of social outcomes, including income, employment, education, incarceration, etc. You can focus on Chapter 14 to see discussion of the relation between cognitive ability and racial inequalities.
- Jencks and Phillips (1998). The Black-White Test Score Gap. A comprehensive collection of works outlining the history, causes, and impacts of Black-White test score gaps. Unlike the previous works, this has a stronger focus on academic achievement rather than IQ gaps. The first chapter can be found here [archive].
- Earl Hunt (2011). Human Intelligence. A comprehensive survey of our scientific knowledge about human intelligence. The book discusses intelligence tests and their analysis, contemporary theories of intelligence, biological and social causes of intelligence, the role of intelligence in determining success in life, and the nature and causes of variations in intelligence across social groups.
- Nicholas Mackintosh (2011). IQ and Human Intelligence. An authoritative overview of the main issues in intelligence research, including the development of IQ tests, the heritability of intelligence, theories of intelligence, environmental effects on IQ, factor analysis, intelligence in the social context, and nature and causes of intelligence differences between different social groups.