Last Updated on March 8, 2023
This is an updated version of my old post from here.
There is overwhelming evidence showing the predictive validity of cognitive ability for important life outcomes. Cognitive ability measured as early as age 6 has a strong association with one’s future success in a number of important outcomes, including academic achievement, occupational performance, income, educational attainment, occupational prestige, criminality, self-control, and health. The associations are typically large, often making cognitive ability the best predictor for such outcomes. In this post, I will cite research showing the evidence for these claims. I will begin with some background on cognitive ability, including definitions, the distributions of IQ test scores, the stability of cognitive ability test scores, and expert consensus on the validity of cognitive ability. Finally, I will cite data demonstrating the predictive validity of cognitive ability in academic achievement, occupational performance, socioeconomic success, anti-social behavior, and health.
Much of the data presented here use correlation coefficients to show the predictive validity of cognitive ability. For someone unfamiliar with differential psychology, it may be unclear whether a given correlation coefficient in this field is large, small, or medium. So I recommend readers check out this post where I provide empirical data on the distributions of correlation coefficients within the field and I also list the correlation coefficients between many commonly understood variables. Following the standards mentioned in that post, I treat low, medium, and large correlations as correlation coefficients in the ranges of |r| < .15, .15 < |r| < .30, and |r| > .30, respectively.
When possible, I will try to demonstrate the predictive validity of cognitive ability without appealing to correlation coefficients. Instead, I will try to cite data from studies showing the average outcome at different cognitive ability ranges when such data is presented. This provides a much more intuitive representation of the strength of the association between cognitive ability and the outcomes in question.
Note: the goal of this post is simply to show that cognitive ability scores are predictive of the important outcomes that I mentioned above. My goal for this post is not to show that cognitive ability is causal. Now, I think the evidence is equally strong that cognitive ability is causal, but showing that will have to wait for a later post. The current post is already large enough!
When I say “cognitive ability”, I’m referring to the definition of “intelligence” given by Gottfredson (1997) [archived]:
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 related 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 fluid and crystalized intelligence. At the bottom of the hierarchy (stratum I) are narrow abilities nested under each broad ability. The broad abilities and some corresponding narrow abilities are as follows:
- Comprehension-Knowledge (Gc): includes general verbal information, Lexical knowledge, foreign language aptitude, communication ability, etc.
- Fluid reasoning (Gf): includes inductive reasoning, general sequential reasoning, quantitative reasoning, etc.
- Quantitative knowledge (Gq): includes mathematical knowledge and mathematical achievement
- Reading & Writing Ability (Grw): includes reading comprehension, reading speed, spelling ability, writing ability, etc.
- Short-Term Memory (Gsm): includes memory span and working memory capacity.
- Long-Term Storage and Retrieval (Glr): is the ability to store information and fluently retrieve it later in the process of thinking.
- Visual Processing (Gv): is the ability to perceive, analyze, synthesize, and think with visual patterns, including the ability to store and recall visual representations.
- Auditory Processing (Ga): is the ability to analyze, synthesize, and discriminate auditory stimuli, including the ability to process and discriminate speech sounds that may be presented under distorted conditions.
- Processing Speed (Gs): is the ability to perform automatic cognitive tasks, particularly when measured under pressure to maintain focused attention.
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). Here’s a nice graph of IQ distributions that I stole from 123test.com.
Now, for some context on how to interpret a given IQ score, consider that the DSM-5 [archived] defines intellectual disability as an IQ score of about 70 or below. “Giftedness” is not a well-defined term but, when defined using IQ scores, it is often defined as possessing an IQ of around 130 or higher (Gottfredson (1997) [archived], page 13).
I first want to review data showing the average IQs at different socioeconomic levels. This data does not show the predictive validity of IQ (since one might argue that the socioeconomic outcomes are causing/predicting one’s IQ rather than the other way around). However, the data is still useful to get an understanding of the IQ ranges that are associated with various achievements.
In Chapter 4 of Kaufman and Lichtenberger (2005), the authors report mean IQs at different educational and occupational levels. For example, they report mean composite IQs by years of education from a standardization sample (Table 4.8):
|Years of Education||Crystallized IQ||Fluid IQ||Composite IQ|
|>= 16 years||111.0||110.2||112.1|
|<= 8 years||83.9||88.6||85.7|
They also report mean IQs from a standardization sample for the Wechsler Adult Intelligence Scale-Revised (WAIS-R). Here are mean full-scale IQs by occupational group for adults at different age groups (Table 4.4):
|Occupational Group||Ages 16-19||Ages 20-54||Ages 55-74|
|Professional and Technical||107||112||114|
|Managerial, Clerical, and Sales||103||104||109|
Unfortunately, these values were based on fairly old standardization samples (e.g., the IQs by occupational group were from Reynolds et al. 1987). Recent data shows similarly sized IQ gaps between different socioeconomic levels.
Similar findings were reported in the most recent iteration of the Wechsler Adult Intelligence Scale, the WAIS-IV test which was released in 2008. In chapter 4 of WAIS-IV Clinical Use and Interpretation, the mean full-scale IQ scores by education level for adults aged 20-90 were as follows (Table 4.1):
|Educational Level||Ages 16-19||Ages 20-90|
|College graduate or above||107.0||110.8|
|High school graduate or GED||96.2||97.3|
|8th grades or less||84.1||83.0|
However, absolute scores by educational level have decreased within the past few decades because lower-IQ individuals have become increasingly more likely to attain higher levels of education. For example, Kaufman et al. (2009) [archived] used a more recent standardized sample to report the following age-adjusted IQs for crystallized intelligence and fluid intelligence by educational level (Table 3):
|Educational Level||Crystallized Intelligence||Fluid Intelligence|
|High school graduate||92.1||92.7|
For data on the relationship between cognitive ability and more specific occupations, consider data on the Wisconsin Longitudinal Study published by Hauser (2002) [archived]. The 10th, 25th, 50th, 75th, and 90th percentiles for IQ are reported for each occupation with 30 or more workers among Wisconsin men who reported a job during 1992-94 (Figure 12). Also, Huang (2001) reports data on the mean/median IQ for occupations from the same dataset (Table 2). See this site for the same figure with guide lines to help view the percentile marks. The IQ scores for different occupations were mostly as one would expect. This data indicates the rarity of persons with below-average IQ scores (i.e. <100 IQ points) among workers whose occupation require significant cognitive complexity (e.g. doctors, engineers, professors, etc.). Persons with low IQ scores are more likely to be found among unskilled or low-skilled laborers (e.g. janitors, truck drivers, manual labor, etc.).
The g factor
Before proceeding, it is important to understand the general factor of intelligence, also called g or the g factor. This is important to understand because many of the studies cited below will report data on g rather than scores from a specific IQ test. To understand the importance of g, one must recognize that there are many different IQ tests. Some popular tests include Raven’s Progressive Matrices, the Wechsler Adult Intelligence Scale, the Stanford–Binet Intelligence Scales, and the Woodcock–Johnson Tests of Cognitive Abilities. Furthermore, these tests seem to measure different kinds of cognitive abilities (e.g., visual-spatial reasoning, verbal ability, logical reasoning, etc.). Despite the variety of test content, scores from these different tests tend to be highly correlated with one another. In other words, individuals who perform above-average on one IQ test also tend to perform above-average on other IQ tests. This suggests that each of these specific tests, in addition to measuring ability in their narrow domain, also measure “some global element of intellectual ability” (Gottfredson 1998 [archived], page 24). This global element can be statistically extracted using factor analysis and is referred to as the general factor of intelligence – abbreviated as g.
Different intelligence tests can be ranked according to their g-loading, i.e. according to the degree to which they accurately measure g. Gottfredson (2002) [archived] gives two reasons for why highly g-loaded tests are important. Firstly, “more g-loaded ones are more complex, whatever their manifest content” (page 28). Secondly, and more importantly, the more g-loaded a test is, “the better it predicts performance, including school performance, job performance, and income” (page 28). These points have led her to pronounce the following (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.
Not only do all tests measure g to some extent, the g scores extracted from different test batteries correlate with one another almost perfectly. For example, Johnson et al. (2004) identified the g factors from three different test batteries — the Comprehensive Ability Battery (CAB), the Hawaii Battery, and the Weschler Adult Intelligence Scale (WAIS). The correlation coefficients from the three batteries were essentially perfect, ranging from r = .99 to r = 1. The study concluded with the following:
Our analyses indicate that g factors from three independently developed batteries of mental ability tests are virtually interchangeable. This is in spite of the fact that the batteries emphasize somewhat different aspects of mental ability, and, though there are some similarities among them, none of the tasks we administered directly overlap. (The most similar tasks are the multiple-choice synonym Vocabulary tasks from the CAB and HB, and even there the words included differ in the two forms. WAIS Vocabulary is administered orally and requires free response definitions. Other similarly titled tasks differ in specific form or method of administration.) The CAB emphasizes inductive reasoning and verbal knowledge while the HB emphasizes nonverbal reasoning and pattern visualization, both in a multiple-choice format. The WAIS is relatively balanced in these areas, but generally elicits free responses. Thus, it seems unlikely that the correlations among the g factors arose either because of common task content or common examination methods, strictly construed.
The authors state that their results provide “the most substantive evidence of which we are aware that psychological assessments of mental ability are consistently identifying a common underlying component of general intelligence.” These findings were replicated by Johnson et al. (2008) using five different test batteries and Floyd et al. (2012) using six different intelligence tests. In fact, g factors from cognitive ability tests have even been shown to be highly correlated with g factors from achievement tests. Kaufman et al. (2012) found a mean correlation coefficient of .83 for g scores pulled from various cognitive ability tests and g scores pulled from reading, math, and writing achievement tests.
So all mental tests measure g to some extent. g is important to understand not only because it often predicts outcomes better than individual IQ scores, but also because there may be situations where we do not have direct access to IQ data. In those situations, we can use other g-loaded tests as measures of cognitive ability. For example, SAT tests, AFQT tests, and even vocabulary tests correlate extremely highly with IQ scores and g, which make them satisfactory measures of cognitive ability.
An important point to note about cognitive ability is its reliability or stability across an individual’s lifetime. Neisser et al. (1996) [archived] report that “Intelligence test scores are fairly stable during development” (page 81). They note that an individual’s age 17-18 IQ correlates at r=0.86 with their age 5-7 IQ, and correlates at r=0.96 with their age 11-13 IQ. Thus, we can predict with fairly high accuracy a person’s IQ at adulthood once we know their IQ at childhood. Similar points were made by Gottfredson (1997) [archived] who states that “intelligence is highly stable beginning in childhood” (page 87). Similar points are made by Sternberg et al. (2001) [archived] who makes two observations on IQ correlations between ages from age 3 to age 12: “First, the best predictor of IQ in a given year is the IQ from the previous year. Second, the predictive power of IQ in every subsequent year increases with the child’s age” (page 15). The stability of cognitive ability has also been verified in many recent studies.
For example, in a literature review on the stability of intelligence over time, Schneider (2014) notes that there is “broad agreement that the stability of cognitive ability varies as a function of the age of the sample but is rather high from school age on” (page 3). For example, consider Yu et al. (2018) which reported data on the Fullerton Longitudinal Study, a program launched in 1979 that followed 130 children from infancy into adulthood with a total of 12 assessments of intellectual performance from age 1 to 17. Consistent with prior studies, this study found that IQ measured at age 17 correlated significantly with age-12 IQ (r=0.82), age-8 IQ (r=0.77), age-6 IQ (r=0.67), and even age-2 IQ (r=0.43) (Table 2). The following graph of correlations show that IQ becomes fairly stable by the time children reach formal schooling:
Larsen et al. (2008) [archived] investigated the stability of general intelligence in a sample of 4000+ adult male veterans drawn from the Vietnam Experience Study (VES). Each participant was given several cognitive ability tests at their induction into the military during 1967-1971 (mean age = 19.9 years) and again in 1985-1986 (mean age = 38.3 years). The coefficient of the correlation between early scores and later scores were used to estimate the “differential stability” of cognitive ability. The resulting differential stability coefficients were 0.85 for general intelligence g, 0.79 for arithmetic ability, and 0.82 for verbal ability (page 32). The authors state their their findings “provide support for the outcome of many other longitudinal studies, suggesting that general intelligence g shows high differential stability from early adulthood to middle-age. In fact, g measured in early adulthood predicts this very ability later in life with a precision that equals the reliability of the tests” (page 33).
Deary et al. (2004) [archived] investigated the stability of mental ability using data from The Scottish Mental Surveys, which collected IQ scores for almost every Scottish person born in 1921 and 1936 and attending school on June 1, 1932 and June 4, 1947. Participants completed the Moray House Test at ages 11 and 80 in order to assess their general mental ability. Researchers found that scores at age 11 correlated shockingly well with scores at age 80. The age-11 scores and age-80 scores correlated at about r=63 or r=0.66 depending on the cohort (Figure 2). After correcting for range restriction, the estimated correlations rose to r=0.73. The following scatterplot shows the association between scores at age 11 and scores at age 80:
To appreciate the strength of these correlations, consider the correlations of one’s score on different subtests of a standardized tests. For example, Koenig et al. (2008) [archived] reports that the correlation between SAT Math and SAT verbal is r=0.74 and the correlation between ACT math and ACT verbal is r=0.67 (Table 2). By comparison, the correlation between a person’s age-17 IQ and age-6 seems to be around r=0.67. This suggests that the correlation between a person’s IQ at the end of high school and their IQ at the start of primary school is about as strong as the correlation between scores on different subtests of standardized tests. In other words, if you predicted a person’s IQ at the end of high school based on their IQ at the start of primary school, this prediction would be about as accurate as predicting someone’s ACT math score based on their ACT verbal score. Similar points can be made about predicting someone’s IQ at age 80 based on their IQ at age 11. For another reference, a recent study reported that, from childhood (ages 5 to 11 years) to middle adulthood (age 45 years), “IQ was as stable across age as height” (Richmond-Rakerd et al. 2021 [archived], page 4) with a correlation coefficient of r = .77.
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)
Surveys on expert opinion have also found consensus that cognitive ability is a substantial predictor of many important life outcomes.
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?” They 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.
Koenig et al. (2008) [archived] reported the correlation between cognitive ability and performance on standardized tests from two different studies.
- In the first study, the researchers investigated over 1,000 subjects from the National Longitudinal Survey of Youth (NLSY79) to calculate the correlation between ACT/SAT scores and the general factor of intelligence (g) extracted from ASVAB test scores. They found that g correlated significantly with SAT total score (r = .82) and ACT total score (r = .77) (Table 2).
- In the second study, researchers correlated ACT scores with Raven’s Advanced Progressive Matrices (Raven’s APM) scores among a sample of 149 college students. They found a correlation of r = .61 between Raven’s APM and Composite ACT score (page 157), which increased to r = .75 after correction for range restriction (page 158).
These large correlations led the authors to conclude that “the ACT is an acceptable measure of general intelligence” (page 158). They ended with the following discussion:
The analyses presented above demonstrate a significant relationship between measures of cognitive ability and ACT scores. Based upon correlations with conventional intelligence tests and the first factor of the ASVAB, it appears that that ACT is a measure of general intelligence. Indeed, based on the correlations among the tests in Study 1, the ACT is indistinguishable from other tests that are identified as intelligence tests. In addition, the ACT shows a high correlation with the SAT, itself considered to be a measure of intelligence (Frey & Detterman, 2004). The jackknife analysis confirms the stability of these results.
To appreciate the magnitude of the IQ-SAT/ACT correlations, note that these correlations are on par with, and sometimes greater than, the correlation between different subtests of the SAT and ACT. For example, here are the correlations between IQ scores and different subtests of the ACT and SAT (Table 2):
|SAT Math||SAT Verbal||ACT Math||ACT Verbal||ASVAB IQ||SAT Total||ACT Total|
This suggests that IQ scores correlate with SAT/ACT scores about as well as the different SAT/ACT subtests correlate with themselves! This study replicated the findings of Frey and Detterman (2004) [archived] which found similarly large correlations between g and SAT scores (r = .72 or r = .86 after correction, depending on the sample), concluding that “the SAT is an adequate measure of general intelligence” (page 377).
A meta-analysis by Roth et al. (2015) [archived] reports the average correlation between general mental ability and school grades among 240 independent samples and 105,185 total participants. After correcting for measurement error and range restriction, the correlation between general mental ability and grades was ρ = .54 (page 123). Moderator analyses (page 123) revealed greater correlations for mathematics and science (ρ = .49) than for languages (ρ = .44), social sciences (ρ = .43), fine art and music (ρ = .31), and sports (ρ = .09). Furthermore, correlation between cognitive ability and grades correlation was largest for high school students (ρ = .58), followed by middle school students (ρ = .54) and elementary school students (ρ = .45). The meta-analysis concluded with the following (page 126):
The results of our study clearly show that intelligence has substantial influence on school grades and thus can be regarded as one of the most (if not the most) influential variables in this context. Although intelligence turned out to be a significant predictor on all moderator levels, we were able to identify some scenarios in which even higher validities can be obtained. First of all, the population correlation was highest for tests relying on both verbal and nonverbal materials, indicating that a broad measure of intelligence or g respectively is the best predictor of school grades. Furthermore, the importance of intelligence increases throughout grade levels. This leads us to the conclusion that intelligence has special importance in educational contexts which deal with content that is more complex and thus can be mastered fully only with an appropriate cognitive ability level.
A similar effect was found in a more recent meta-analysis by Zaboski et al. (2018).
For more concrete examples of the association between cognitive ability and high school grades, see Cucina et al. (2016). In one of their reported studies, the authors used the 1997 cohort of the National Longitudinal Survey of Youth (NLSY97) to examine the relationship between high school grades and general cognitive ability (g) extracted from ASVAB scores. Consistent with previous literature, a large correlation was observed between g and high school GPA (r = .44, table 6). To illustrate the association more clearly, the authors also reported the distribution of GPA scores by g quartile (figure 2). Here are the percentage of students with different HSGPA averages by g quartile:
|g quartile||A average||B average||C average||D average or lower|
|Quartile 4 (highest)||19.1%||63.9%||16.6%||0.4%|
|Quartile 1 (lowest)||0.7%||39.5%||50.8%||9.1%|
Compared to students with g scores in the bottom quartile (IQ < 90), students with g scores in the top quartile (IQ > 110) were nearly 30 times as likely to earn an average letter grade of an A (19.1% vs 0.7%) and over twice as likely to earn a B or higher (83% vs 40%), whereas students in the bottom quartile were over 20 times as likely to earn a D or lower (9.1% vs 0.4%).
The above analyses included samples reported by cross-sectional studies, i.e. studies that measure the intelligence and academic performance of students at the same time. It may be more interesting to examine the correlations reported specifically by longitudinal studies, i.e. studies that measure the correlation between intelligence at one time and academic achievement measured at a later time. One such longitudinal study is Deary et al. (2006) [archived], which examined a 5-year prospective longitudinal survey of a representative sample of over 70,000 children in England. Researchers measured the relationship between the general factor of intelligence (g) measured at age 11 and GSCE test points at age 16. The results were as follows:
- The correlation between g measured at age 11 and GCSE test points at age 16 was r = .69. The largest correlation was found between g and mathematics (r = .77).
- Among students with the mean value of g, 58% achieved five or more GCSE scores at grades A* to C. Of those scoring 1 standard deviation higher on g, 91% achieved this criterion. Among those scoring 1 standard deviation lower on g, only 16% achieved this criterion (page 18).
A recent review of evidence on the association between cognitive ability and academic achievement has been published by (Malanchini et al. 2020 [archived]).
The correlation of cognitive ability with academic achievement is typically greater than the same correlation with parental SES and academic achievement. See the following studies:
- A meta-analysis by Richardson et al. (2012) [archived] reported that the correlation of intelligence with college GPA (ρ = .21) was somewhat greater than the correlation with SES (ρ = .15) (Table 6). Note, though, that this is the correlation found among college students. Therefore, the reported intelligence-GPA correlation is likely a severe underestimate due to range restriction.
- Roth et al. (2015) [archived] estimated the correlation between general mental ability and grades to be ρ = .54. By comparison, a meta-analysis by Harwella et al (2016) found that the average correlation between SES and achievement was “surprisingly modest” at only r = 0.22.
Absent range restriction, cognitive ability also predicts achievement better than any Big Five personality traits as well. With sufficient range restriction, cognitive ability typically predicts success about as well as conscientiousness, although still considerably better than any other Big Five trait:
- A meta-analysis by Poropat (2009) [archived] reports that academic performance at the primary educational level correlates with cognitive ability (ρ = .58) substantially more than with any other Big Five personality trait, including conscientiousness (ρ = .28) (Table 2). However, at higher educational levels, conscientiousness begins to predict grades about as well as cognitive ability, probably because the analysis could not correct for range restriction (page 327). For example, academic performance at the tertiary level correlated with intelligence (ρ = .23) about as well as with conscientiousness (ρ = .23), although they were both much stronger predictors than the other personality traits (ρ = −.01 to .12).
- A meta-analysis by Richardson et al. (2012) [archived] also showed that college GPA correlated with intelligence (ρ = .21) about as well as with conscientiousness (ρ = .23) (Table 6). Both intelligence and conscientiousness predicted better than any other personality trait, except for procrastination (ρ = −.25). Keep in mind that this sample was limited to students already at university, which means the correlations for intelligence (and perhaps some of the other variables) were likely underestimated because the analysis did not control for range restriction (page 375).
Therefore, at higher levels of education, cognitive ability predicts grades about as well as conscientiousness (although the predictive power of cognitive ability is likely underestimated due to range restriction). Moreover, at the college level, there are many better predictors of GPA, such as high school GPA (ρ = .41), SAT scores (ρ = .33), and ACT scores (ρ = .40). However, at lower levels of education (see Poropat 2009, Table 2), or after controlling for range restriction (Roth et al. 2015), cognitive ability is a much better predictor than all Big Five personality traits, including conscientiousness. One might criticize this comparison because the meta-analytic correlations for cognitive ability and conscientiousness come from different samples of studies with different participants and methods. To avoid this issue, see the following more recent individual studies that directly compare cognitive ability and conscientiousness within the same sample:
- Spinath et al. (2010) examined predictors for school achievement in 1,353 Austrian 8th graders. They found that general intelligence correlated with math grades (r = .56) considerably more than did conscientiousness (r = .11 – .17 depending on gender), interest in mathematics (r = .13 – .22 depending on the gender), anxiety, or any other personality trait. The study concluded with “intelligence was the strongest predictor of school achievement in all domains for both sexes” (page 484).
- Cucina et al. (2016) examined the relationship between high school grades and various mental abilities. This study found that grades correlated significantly with conscientiousness (r = .32), but still not as great as with g (r = .37 to .40). Keep in mind though the there were no corrections for range restriction, so these correlations are likely underestimates.
Now, studies do show that composite measures of Big Five personality (rather than individual personality traits) may predict grades better than cognitive ability does. For example, Lechner (2017) reported that personality (measured using the Big Five) explained a greater percent of variance in school grades than did intelligence (18% vs 10%) in a large representative sample of German ninth-graders (interpret this result with caution since the correlation between cognitive ability and school grades found in this study was much lower than that found other studies; recall that the meta-analysis by Roth et al. 2015 found a correlation of .54 between grades and cognitive ability after correcting for range restriction, which corresponds to .54^2 = 29% of variance explained, much greater than the 10% found in this study). See this graph showing the percentage of variance explained by intelligence and personality tests:
Furthermore, there is some data suggesting that measures of self-discipline predict grades better than cognitive ability does (Duckworth and Seligman 2005, Duckworth et al. 2012, Galla et al. 2019), especially when the self-discipline measures are reported by teachers. Regardless, these studies all show that cognitive ability remains the better predictor of standardized test scores. For example, Duckworth et al. (2012) concluded with the following:
We proposed a theoretical model (summarized in Figure 1) distinguishing between competencies better assessed by report card grades and influenced by self-control and, in contrast, competencies better assessed by standardized achievement tests and influenced by intelligence. Two prospective, longitudinal studies of middle school students supported predictions from this model: self-control predicted changes in report card grades over time better than did IQ, whereas IQ predicted changes in standardized achievement test scores better than did self-control.
Also, Galla et al. (2019) had a similar findings:
Adjusting for demographic characteristics, self-regulation was a stronger predictor of HSGPA (β = .77, p < .001, 95% CI [.70, .83]) than of SAT score (β = .19, p < .001, 95% CI [.12, .27]), Wald(1) = 145.86, p < .001. Cognitive ability, on the other hand, was a stronger predictor of SAT score (β = .74, p < .001, 95% CI [.68, .81]) than of HSGPA (β = .22, p < .001, 95% CI [.15, .29]), Wald(1) = 141.32, p < .001.
So it seems as if cognitive ability may not be the best predictor of grades. Measures of self-discipline or self-regulation tend to be better predictors of grades. However, we should be cautious in interpreting these findings because of the possibility of reverse causation. Survey respondents might rate a student as having higher self-discipline because the student receives higher grades, so the self-discipline measures may in part be an effect (rather than predictor) of grades. Another reason to remain cautious is that the cognitive ability correlation may be underestimates as a result of range restriction. Despite these concerns, these studies still find that cognitive ability is the better predictor of standardized test scores, which self-discipline predicts rather poorly.
In summary, cognitive ability is a great predictor of academic achievement. The correlation with grades varies depending on the educational level and depending on whether one corrects for range restriction. The meta-analysis by Roth et al. (2015) suggests that the correlation between grades and cognitive ability is about .44 in elementary to high school, which raises to .54 after correcting for range restriction. This is substantially higher than the correlations between grades and other predictors, such as parental SES or specific personality traits, although other data suggests that self-discipline measures or personality composites predict grades better than cognitive ability does. In college, cognitive ability seems to predict grades about as well as conscientiousness does (Poropat 2009, Richardson et al. 2012), probably due to substantial range restriction. Finally, the correlations between cognitive ability and standardized test scores (SAT/ACT) are unrivaled by any alternative predictors; in fact, the correlations are so great that some consider the ACT/SAT to be tests of general cognitive ability (Koenig et al. 2008, Frey and Detterman 2004).
The association between cognitive ability and job performance is so strong that Gottfredson (1997) [archived] has forcefully asserted that “g can be said to be the most powerful single predictor of overall job performance” and that “no other single predictor measured to date (specific aptitude, personality, education, experience) seems to have such consistently high predictive validities for job performance” (page 83). I will provide some data corroborating these claims in this section.
Schmidt and Hunter (1998) [archived] is a highly cited paper that summarized 85 years of research on the predictive validity of dozens of variables for job performance and job training programs in the United States. The study considered job experience, years of education, interests, employment interviews, conscientiousness tests, work sample tests (hands-on simulations of the job to be performed by the applicant), GMA tests (general mental ability tests), peer ratings of performance, job knowledge tests, behavioral consistency procedures (applicants describe their past achievements to illustrate their ability), and job tryout procedures (applicants are hired with minimal screening and their performance is evaluated within a limited duration, e.g. several months). The correlation between job performance and some of the predictor variables were as follows (Table 1):
|Personnel measures||Validity (r)|
|Work Sample tests||.54|
|General Mental Ability tests||.51|
|Employment interviews (structured)||.51|
|Behavioral Consistency Procedures||.45|
|Job Tryout Procedures||.44|
|Years of Education||.10|
The correlation between job training and various predictor variables were as follows (Table 2):
|Personnel measures||Validity (r)|
|General Mental Ability tests||.56|
|Biographical Data Measures||.30|
|Years of Education||.20|
The predictive validity of GMA tests varied based on the job complexity. The predictive validity of GMA tests for job performance was greatest for professional-managerial jobs (r = .58), followed by high level complex technical jobs (r = .56), medium complexity jobs (r = .51), semi-skilled jobs (r = .40), and unskilled jobs (r = .23) (page 264). These results were also published in Schmidt and Hunter (2004).
An updated meta-analysis was reported in Schmidt et al. (2016) [archived], which found largely similar results. Also, see Roth et al. (2005) which presented data showing that the estimates for the predictive validity of work sample tests may be an overestimate.
Similar results were reported by meta-analyses in other countries. For example, consider the following meta-analyses:
- Bertua et al. (2005) [archived] analyzed 283 independent samples to study the association between general mental ability (GMA) tests and job performance and job training success in the United Kingdom. The mean correlation between GMA tests and job performance was r = .22, which raised to ρ = .48 after correcting for criterion reliability and range restriction (Table 1). The correlation between GMA tests and job training success was r = .29, which raised to ρ = .50 after corrections (Table 2). For job performance, they found that the operational validity of general mental ability for professional workers (r = .36, ρ = .74), engineers (r = .33, ρ = 0.70), and managers (r = .33, ρ = .69) was greater than for clerical workers (r = .14, ρ = .32) and drivers (r = .16, ρ = .37). For job training, the variation in validity by occupational group was not as great: the occupational validity of general mental ability for engineers (r = .39, ρ = .64) and professionals (r = .35, ρ = .59) and was not substantially greater than the validities for clerical workers (r = .33, ρ = .55) and drivers (r = .28, ρ = .47).
- Hülsheger et al. (2007) analyzed 99 independent samples to examine the relationship between general mental ability and job performance and job training success in Germany. Consistent with meta-analyses in other countries, the authors found that general mental ability was highly predictive of both training success (r = .31, ρ = .47 after correcting for range restriction and criterion reliability) and job performance (r = .33, ρ = .53). However, contrary to prior studies, validity for training success was greatest for jobs of low complexity (r = .35, ρ = .52), followed by jobs of medium complexity (r = .29, ρ = .45) and high complexity (r = .19, ρ = .30). The authors speculate that these results may diverge from other meta-analyses because of the structure of the German educational system. In Germany, students are stratified into different kinds of secondary schools based on their prior academic achievement. Students may be stratified into schools tailored toward basic secondary education, schools that offer more demanding curriculum, or schools that offer prep for academic careers. This stratification results in more homogeneous applicant groups, which may result in indirect range restriction and therefore reduced correlations for cognitive ability (pages 11-12).
- Salgado et al. (2003) [archived] meta-analyzed 166 independent samples to study the relationship between general mental ability and job performance and job training success in Europe. The analysis studied samples from United Kingdom (k = 36), France (k = 18), Spain (k = 13), the Netherlands (k = 10), Germany (k = 9), Belgium (k = 2), and Ireland (k = 1). The study found that general mental ability predicted job performance for low complexity jobs (r = .25, ρ = .51 after correcting for criterion reliability and range restriction), medium complexity (r = .27, ρ = .53), and high complexity jobs (r = .23, ρ = .64). The greatest correlations were found for managers (r = .25, ρ = .67), followed by sales workers (r = .34, ρ = .66), engineers (r = .23, ρ = .63), and information clerks (r = .31, ρ = .61). Moreover, general mental ability predicted job training success for low complexity jobs (r = .23, ρ = .36), medium complexity (r = .29, ρ = .53), and high complexity jobs (r = .29, ρ = .74).
For reviews of the major meta-analyses in the literature on the predictive validity of cognitive on job performance, see Salgado (2017) [archived] and Salgado and Moscoso (2019) [archived]. Salgado and Moscoso (2019) also included a new meta-analysis of all validation studies conducted by the U.S. Employment Service over the period 1950-1985.
One criticism of the previous studies is that they often rely on supervisory ratings, which may be subject to arbitrary bias. To address this issue, we can use work sample tests instead of supervisory ratings. Roth et al. (2005) meta-analyzed 43 independent samples of 17,563 total subjects to analyze the association between cognitive ability tests and work sample tests. Work sample tests are defined as tests “in which the applicant performs a selected set of actual tasks that are physically and/or psychologically similar to those performed on the job” (page 1010). The mean correlation between cognitive ability tests and work sample tests was r = .32, which increased to r = .38 after correcting for unreliability in work sample (Table 4). This correlation was somewhat reduced because military jobs were included in the meta-analysis, which suffer from significant range restriction (the military uses measures of cognitive ability in its selection process). Among non-military jobs only (K = 16, N = 5,039), the correlation between cognitive ability tests and work sample tests was r = .37, which increased to r = .44 after correcting for unreliability in work sample.
Cognitive ability has predictive validity for nearly every aspect of occupational success. Strenze (2015) [archived] cites several meta-analyses showing the correlation between IQ and a variety of measures of occupational success (Table 25.1). There are large correlations between IQ and job performance (r = .53 for supervisory rating and r = .38 for work samples), skill acquisition in work training (r = .38), group productivity (r = .33), and promotions at work (r = .28). Consistent with the prior meta-analyses, he also notes that cognitive ability is a better predictor of success for cognitively demanding jobs. He states “IQ tests are very useful in selecting good engineers, architects, or dentists…IQ tests are less useful for selecting good dishwashers, weavers, or garbage collectors, although, even among dishwashers, it is obvious that an intelligent worker is better than a less intelligent one” (page 407).
Cognitive ability predicts occupational performance in military research. Research on the predictive validity of cognitive ability in military research is useful because cognitive ability is often taken prior to assessment of performance (e.g., scores on test scores may be used to determine enlistment qualification). Military research is also useful because tests of job performance are often objective “hands on” tests rather than purely subjective ratings of performance by supervisors. Gottfredson states that “military research has consistently shown that highly g-loaded measures such as the Armed Forces Qualifying Test (AFQT) and its forerunners, although not always conceptualized as measures of g, are good measures of “trainability”” (page 86).
The predictive validity of cognitive ability for occupational success is so important that 10 U.S. Code § 520 [archived] actually outlaws individuals from participating in the Armed Forces if they perform sufficiently poorly on IQ tests. Gottfredson (1997) [archived] notes that this law is requirement in order to prevent the recruits from being overpopulated with expensive and untrainable members (page 90):
Lest IQ 80 seem an unreasonably high (i.e., exclusionary) threshold in hiring, it should be noted that the military is prohibited by law (except under a declaration of war) from enlisting recruits below that level (the 10th percentile). That law was enacted because of the extraordinarily high training costs and high rates of failure among such men during the mobilization of forces in World War II (Laurence & Ramsberger, 1991; Sticht et al., 1987; U.S. Department of the Army, 1965). Minimum enlistment standards since World War II have generally been higher than the 10th percentile, and closer to what they are today for the different services: the 16th AFQT percentile (Army, about IQ 85), 21st (Marine Corps and Air Force, IQ 88), and 27th (Navy, IQ 91). It should be noted that these are the enlistment standards for high school graduates. Non-graduates must score above the 27th to 65th percentiles on the AFQT, depending on the service in question (Laurence & Ramsberger, 1991, p. 11).
For an example of data showing the predictive validity of cognitive ability in military success, see Ree et al. (1994) [archived] which examined the predictive validity of general cognitive ability (g) and specific abilities (s) on job performance for 1,036 U.S. Air Force enlistees across 7 different jobs. The researchers recorded the predictive validity of g on job performance using 3 different criterion measures of success: a hands-on performance test, an interview work sample test, and a walkthrough performance test. Averaged over all jobs, the predictive validity of g ranged from r = .40 to r = .44, depending on the criterion measure used (Table 2). The study also found that g was the most potent predictor of success and that s (non-g specific abilities/knowledge) “added little to prediction” (page 522). Similar results were reported by Olea and Ree (1994), which analyzed the predictive validity for students in navigator training and pilot training.
Another study showing the predictive validity of cognitive ability for military success was conducted by McHenry et al. (1990) [archived]. Researchers of this study examined the association between various predictors and 5 components of job performance in 9 different Army jobs. There were 4,039 total incumbents studied from the following jobs: infantryman, cannon crewmember, armor crewman, single channel radio operator, light wheel vehicle mechanic, motor transport operator, administrative specialist, medical specialist, and military police. The predictors included general cognitive ability extracted from 9 ASVAB subtests, 6 tests of spatial abilities, 10 tests of perceptual-psychomotor abilities, 11 measures of temperament/personality, 22 tests of vocational interests, and 6 measures of job outcome preferences. Perceptual-psychomotor test scores were composed of components such as perceptual speed, perceptual accuracy, number speed and accuracy, reaction speed, etc. The 5 components of job performance are as follows:
- Core Technical Task Proficiency: The proficiency with which the individual performs the tasks that are specific and “central” to his or her job (MOS). The tasks represent the core content of the job that distinguishes it from other jobs.
- General Task Proficiency: In addition to the core technical content specific to an MOS, individuals in every job are responsible for being able to perform a variety of general or common tasks–e.g., use of basic weapons, first aid. This factor represents proficiency on these general tasks.
- Peer Support and Leadership, Effort, and Self Development: Reflects the degree to which the individual exerts effort over the full range of job tasks, perseveres under adverse or dangerous conditions, and demonstrates leadership and support toward peers.
- Maintaining Personal Discipline: Reflects the degree to which the individual adheres to Army regulations and traditions, exercises personal self-control, demonstrates responsibility in day-to-day behavior, and does not create disciplinary problems.
- Physical Fitness and Military Bearing: Represents the degree to which the individual maintains an appropriate military appearance and bearing and stays in good physical condition.
The correlation between the different predictors and the different measures of job performance are as follows (Table 4):
Note that cognitive ability tests best predicted outcomes requiring cognitive complexity (core technical proficiency and general soldiering proficiency), whereas temperament and personality tests best predicted the other outcomes.
There is also a significant association between cognitive ability tests and work sample tests among military jobs. Roth et al. (2005) performed a meta-analysis of 27 samples of 12,524 participants and found a correlation of r = .30 between cognitive ability and work sample tests in military jobs, which raised to r = .36 after correcting for unreliability in work sample (Table 4). The correlation was r = .48 for the 14 samples that were corrected for range restriction (N = 6,100).
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.
The meta-analysis reveals that IQ has a medium correlation with income (r = .20), a very large correlation with occupational attainment (r = .43), and even larger correlations with educational attainment (r = .56). Moderator analysis was performed to determine if the association between youth IQ and adulthood success was influenced by age at testing or age at measure of success. Regarding age at testing, Strenze found that IQ measured as early as age 3-10 significantly predicted adulthood outcomes (Table 2), although not as strongly as IQ measured at later ages. After age 10, age at testing did not significantly impact the IQ-success correlation (Table 2). Regarding age at success, the correlation between youth IQ and educational attainment was fairly constant regardless of the age at success. However, the correlation between youth IQ and occupation/income was much lower when success was measured at ages 20-24. The correlation between youth IQ and occupation/income were largest when success was measured after age 29 (Table 2), suggesting that IQ may be more important for one’s career success later in life.
For more concrete examples of the association between adolescent cognitive ability and socioeconomic outcomes, see Murray and Herrnstein (1994) [archived]. These authors used the NLSY79 to measure the predictive power of cognitive ability on a variety of socioeconomic outcomes. They 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”). They then reported the average levels of socioeconomic success for each cognitive class. This analysis was restricted to non-Hispanics whites to avoid any confounding with race. As expected, those from the higher cognitive classes attained far higher levels of success than those in the lower cognitive classes:
- High school graduation: The high school dropout rate across all subjects was 9%. The high school school dropout rate for those in the “very bright” and “bright” cognitive classes were effectively 0% (page 146). By contrast, the high school dropout rate was far higher for those in the “normal” cognitive class (6%), those in the “dull” cognitive class (35%), and those in the “very dull” cognitive class (55%).
- Poverty: The poverty rate across all subjects was 7%. The poverty rate those in the “very bright” cognitive class (2%) was lower than the poverty rate for those in the “bright” cognitive class (3%), those in the “normal” cognitive class (6%), those in the “dull” cognitive class (16%), and those in the “very dull” cognitive class (30%) (page 133).
These statistics were also reported by Gottfredson (1998) [archived] (page 28). Murray (1998) [archived] extended this analysis to include a larger set of socioeconomic outcomes (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|
Murray also performed a similar analysis after restricting the sample to what he calls the “Utopian Sample”. This includes only subjects who grew up with both biological parents married from birth at least until age seven and who had parents with income above 25th percentile. The result was a sample that “has virtually no illegitimacy, divorce, or poverty” (page 33). When the analysis is restricted to the utopian sample, the same association between cognitive ability and socioeconomic outcomes appeared (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||80%||45.6||$38,000|
|Bright (75th – 89th)||15.2||57%||45.1||$27,000|
|Normal (25th – 74th)||13.4||19%||43.0||$23,000|
|Dull (10th – 24th)||12.3||4%||39.0||$16,000|
|Very Dull (10th-)||11.4||1%||35.8||$11,500|
As you can see, the association between cognitive ability and socioeconomic outcomes remains the same even when restricting the analysis to this utopian sample.
These findings on the IQ-income correlations 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.
These numbers provide a good example of how correlation coefficients may be misleading when reporting the association between two variables. The correlation coefficient between IQ and net worth in this dataset is “only” r = .16, which is small/medium based on the benchmarks mentioned in my post here. Yet the differences in net worth at the different IQ ranges are massive. What gives? To understand what’s happening here, we need to understand what a correlation coefficient actually indicates. If two variables X and Y have correlation coefficient r, then that implies that each standard deviation increase in X is associated with an increase of r standard deviations of Y. So the correlation coefficients from this data suggests that each standard deviation increase in IQ (15 points) is associated with an increase of .16 standard deviations of net worth. But the standard deviation of net worth in this sample was $447,814! (Table 1). So an increase in IQ of 15 points was associated with an increase in net worth of about .16 * $447,814 = $71,650, which is more in line with the data presented above. In other words, the seemingly low correlation coefficient between IQ and net worth is the result of extremely high variance for net worth.
For a replication using a different sample and different country, see Fergusson et al. (2005) [archived]. This study examined a birth cohort of 1,265 children born in the Christchurch (New Zealand) urban region in mid-1977. Researchers found that IQ measured at ages 8-9 was significantly related to outcomes such as socioeconomic outcomes such as educational attainment, gross income, and unemployment. For example, consider the following outcomes that were reported by IQ range (Table 4):
After controlling for a variety of covariates (e.g., childhood conduct problems, attentional problems, and socioeconomic disadvantage), researchers found that childhood IQ was still significantly associated with these outcomes. The study concluded with the following (page 856):
Also in agreement with previous research, the findings of this study suggested that intelligence measured in middle childhood has pervasive associations with later educational achievement, university entrance, degree attainment, employment and income (Caspi, Wright, Moffitt, & Silva, 1998; Herrnstein & Murray, 1994; Jensen, 1999). In general, increasing IQ was associated with increasing educational success at school, higher rates of post-school educational/vocational attainment, degree success, lower rates of unemployment and higher income at age 25. Statistical control for a wide range of factors including early conduct problems and family, social and childhood circumstances failed to explain these associations, supporting the view that intelligence had a direct relationship to later educational, occupational and related outcomes independently of other childhood characteristics and family environment.
A particularly large study showing the association between early cognitive ability and later socioeconomic status was conducted by Hegelunda et al. (2018) [archived]. The researchers investigated all men born since 1950 who appeared before a draft board in Denmark during periods from 1968 to 1984 and 1987 to 2015 (N = 1,098,742). IQ was assessed at age 18 using a test called the Børge Prien Prøve (BPP) which comprised four subtests assessing logical, verbal, numerical, and spatial abilities. Consistent with prior studies, researchers found that youth IQ was significantly associated with socioeconomic outcomes in early adulthood (Table 2):
A similarly large study was conducted for Swedish men in the labor force by Keuschnigg et al. (2023) [archived]. The authors analyzed the relationship between early cognitive ability and later socioeconomic success. Cognitive ability scores were available for men who were obligated to undergo military conscription at ages 18-19. Such men were required to complete tests of “verbal understanding, technical comprehension, spatial ability, and logic” upon conscription. The authors focused subjects with 11 years of labor market participation, centered around the age of 40. The result was about 60 thousand Swedish-born men with labor market data to analyze. Socioeconomic success was measured based on annual wages reported to Swedish tax authorities and occupational prestige as measured on the International Socio-Economic Index (ISEI) scale.
Consistent with prior studies, the authors find that cognitive ability is significantly associated with annual wage (correlation is 0.4). They also find that the relationship is monotonic, “rejecting the claim that past a certain threshold having even higher cognitive ability does not matter”. Interestingly, while median wages increases for all increases in cognitive ability; the reverse does not seem to be true. That is, subjects with the highest wages did not consistently have the highest cognitive ability. In particular, the average annual wage of subjects in the top percentiles of annual wage were not significantly higher than the wages of subjects in slightly lower percentiles. These findings were illustrated in Figure 3:
Somewhat similar patterns (though less dramatic) were observed regarding the relationship between cognitive ability and occupational prestige. That is, while higher levels of cognitive ability consistently predicted higher occupational prestige; the reverse did not seem true. Individuals with the highest level of occupational prestige did not have much higher levels of cognitive ability than their slightly-less-prestigious peers (Figure 4):
The association between early cognitive ability and later socioeconomic success is a consistent finding that has been replicated in numerous other countries, such as Britain (Bukodi et al. 2013, tables 3-4; Von Stumm 2009), Scotland (Von Stumm et al. 2010, Deary et al. 2005), Sweden (Bergman et al. 2014, Sorjonen et al. 2012), Ireland (O’Connell and Marks 2021), and Germany (Becker et al. 2019).
Recall the meta-analysis by Strenze (2007) [archived] which reported average correlations between adulthood SES and youth cognitive ability and many other predictors of adulthood SES (Table 1). For each of the socioeconomic outcomes, youth intelligence predicted the outcome as well or better than all alternative predictors:
- Education: educational level correlated with youth intelligence (p = .56) as well as or better than alternative predictors, including parental SES index (p = .55), academic performance (p = .53), father’s education (p = .50), mother’s education (p = .48), father’s occupation (p = .42) and parental income (p = .39).
- Occupation: occupational level (occupational prestige/status) correlated with youth intelligence (p = .45) as well as or better than alternative predictors, including parental SES index (p= .38), academic performance (p = .37), father’s occupation (p = .35), father’s education (p = .31), mother’s education (p = .27), and parental income (p = .27).
- Income: personal income correlated with youth intelligence (p = .23) as well as or better than alternative predictors, including parental income (p = .20), father’s occupation (p = .19), parental SES index (p = .18), father’s education (p = .17), mother’s education (p = .13), and academic performance (p = .09).
All correlations are pulled from Table 1 using the sample size weighted average correlations corrected for unreliability and dichotomization. Also, the correlations for intelligence are reported from the “best studies”, i.e. studies where intelligence is tested before the age of 19 and socioeconomic success is measured after the age of 29.
For comparisons between youth cognitive ability and alternative predictors from within the same sample, see Spengler et al. (2018) [archived]. Researchers in this study used data from Project Talent to compare the validity of various predictors for socioeconomic outcomes long after high school. Project Talent is a longitudinal sample of Americans followed from high school to late adulthood. The dataset contains information about each participant’s parental SES, personality traits, academic achievement, and IQ while they were in high school. Parental SES was a composite score consisting of home value, family income, parental education, father’s job status, number of books, number of appliances, number of electronics, and whether the child had a private room. The study reported information on the socioeconomic outcomes of the participants at two points during adulthood, one that was 11 years after the initial sampling and another that was 50 years after the initial sampling. The results of the 50-year follow-up are consistent with the data shown thus far, which is that cognitive ability predicts socioeconomic outcomes better than alternative predictors. Here are some of the interesting correlation coefficients between variables at youth and socioeconomic outcomes 50 years later (Table 2):
|Variable||Educational Attainment||Occupational Prestige||Income|
|Interest in school||.22||.13||.14|
Impressively, youth IQ predicted income at the 50-year follow-up (r = .35) better than income at the 11-year follow-up (r = .21), better than occupational prestige at the 11-year follow-up (r = .33), and almost as well as occupational prestige at the 50-year follow-up (r = .38). The study also reported data from various regression models showing the correlations between IQ and socioeconomic outcomes after controlling for the other variables. I don’t report on that data in this post because this post is focused on simply showing the predictive validity of IQ, not that IQ is necessarily causal.
In summary, cognitive ability measured at youth is an excellent predictor of socioeconomic outcomes measured at adulthood. In fact, the cognitive ability measured at youth is a better predictor of these outcomes than the socioeconomic outcomes of one’s parents, which is commonly taken to be a strong predictor of offspring success. Thus, if you want to predict a child’s socioeconomic success, it’s better to know how smart they are than it is to know the socioeconomic success of the child’s parents.
Crime, idleness, dependency, and illegitimacy
A meta-analysis by Ttofihi et al. (2016) [archived] investigated the extent to which intelligence may function as a protective factor against delinquency, violence, and crime. The authors investigated 15 longitudinal studies based in Europe (8), the United States (5) and New Zealand (2). The studies reported the impact of intelligence on the likelihood of offending among both high-risk and low-risk groups. “High-risk” groups includes individuals who were exposed to risk factors (other than low intelligence) for offending. These risk factors varied from study to study. Some risk factors included poor child rearing, teacher- and parent-ratings of antisocial behavior, poor concentration, marital disturbance, imprisoned father, physical abuse, etc. (table 1). The authors found that, among the high-risk group, non-offenders were about 2.32 times as likely to have a high intelligence level as offenders (page 13). Some studies also investigated the effect of intelligence on offending among low-risk groups. For this group, non-offenders were only about 1.3 times as likely to have a high intelligence level, a non-significant result (page 12). The meta-analysis concludes that “intelligence can function as a protective factor for offending”. In other words, the impact of risk factors for offending is reduced among individuals of high intelligence; or, conversely, low intelligent individuals are particularly vulnerable to be negatively impacted by the risk factors for offending. The following graph shows the percentage of offenders as a function of increasing risk within high and low IQ individuals:
Returning to The Bell Curve, Herrnstein and Murray (1994) reported a number of anti-social behaviors that are (negatively) associated with cognitive ability. Recall that the authors 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”). Also recall that these associations are reported specifically for non-Hispanic whites to avoid confounding due to race. Some of the anti-social behaviors negatively associated with cognitive ability include welfare usage, illegitimacy, and incarceration. The following table shows the percentage of subjects at each cognitive class who were incarcerated, were convicted, were unemployed, used welfare usage, or who had illegitimate children.
|Cognitive Class (percentile range)||Incarceration||Criminal Conviction||Unemployment||Welfare Usage||Illegitimacy|
|Very Bright (90th+)||1%||3%||2%||1%||7%|
|Bright (75th – 89th)||1%||7%||7%||4%||7%|
|Normal (25th – 74th)||3%||15%||7%||12%||13%|
|Dull (10th – 24th)||7%||21%||10%||21%||23%|
|Very Dull (10th-)||12%||14%||12%||55%||42%|
- “Incarceration” records the percentage of men who were incarcerated at the time of the interview (page 248). “Criminal Conviction” records the percentage of men who reported being convicted for an offense (page 247). “Unemployment” records the percentage of men who spent one month or more in 1989 (page 163). “Welfare usage” records the percentage of women who went on AFDC (Aid to Families with Dependent Children) within a year of first birth (page 194). “Illegitimacy” records the percentage of first births among women that were illegitimate (page 181).
Many of these statistics were reported by Gottfredson (1998) [archived] (page 28).
Another longitudinal study showing the association between early cognitive ability and later criminal offending was conducted by Loeber et al. (2012) [archived]. Researchers used data from the Pittsburgh Youth Study to examine the relationship between IQ measured at about age 12 with criminal history at age 28 in a sample of 422 males. IQ was measured using the Wechsler Intelligence Scale for Children–Revised (WISC-R) test. The results showed that IQ was significantly associated with arrest probability for any charge, particularly during adolescence. For example, the probability of arrest for 17-year-old males with low IQs (60-65%) was about three times the probability for those with high IQs (20-25%) (Figure 1). Low-IQ and high-IQ males were those with IQs one standard deviation below and above the mean, respectively. Note also that this difference was the difference after controlling for race, socioeconomic status, and age.
The (negative) association between cognitive ability and criminal offending has been replicated in other studies. For example, Beaver (2013) used the National Longitudinal Study of Adolescent Health to analyze the relationship between IQ scores and criminal justice outcomes. The dataset involves a nationally representative and prospective sample of American youths who were assessed at four different points in time. The first wave of data was collected in 1994-1995 when participants were adolescents. The second wave was conducted about 1.5 years later with participants aged 13-22 years. The third wave was conducted in 2001-2002 with participants aged 18-28 years. The fourth wave was conducted in 2007-2008 with participants aged 24-32 years. IQ was measured using an average of Peabody Vocabulary Test (PVT) scores administered in the first and third waves. Criminal justice outcomes were measured during interviews at wave 4 by asking participants if they were arrested and/or incarcerated. The study found significant associations between IQ and multiple measures of crime. For example, percentage of respondents with the lowest IQ scores who were arrested and incarcerated was .562 and .439, respectively (page 281). By contrast, the same percentages for respondents with the highest IQ scores was .344 and .153, respectively. Similar findings were observed when limiting the analysis to black males, white males, black females, and white females. For example, this graph shows the relationship between IQ scores and criminal justice outcomes for black males:
This shows the same relationship for white males:
The author ended with the following conclusion (page 285):
Taken together, the results revealed a robust association between IQ and multiple measures of crime even after taking into account the limitations that plague prior research. When combined with previous literature, there is good reason to believe that IQ is related to criminal involvement regardless of the sample analyzed, the measurement of crime, and the inclusion of controls for potentially confounding factors, such as executive functions. There is likely not another individual level variable that is so consistently associated with crime and other forms of antisocial behaviors than IQ. When viewed in this light, these findings, along with those of the existing literature, tend to suggest that the criminogenic effects of IQ are not a methodological artifact and that research examining the etiology of crime and antisocial behaviors is likely misspecified if a measure of IQ is not included in the statistical models.
The crime-cognitive ability association has also been found in other countries. Schwartz et al. (2015) [archived] examined a total birth cohort of Finnish males born in 1987 (N = 21,513) to investigate the relationship between IQ and criminal offending. Data on intelligence was obtained from the Finnish Defense Forces Basic Ability Test which is taken by all new recruits during the first two weeks of military service (military training is mandatory for all males in the country, with nearly 90% participating). The intelligence tests contained three subscales of cognitive ability – mathematical, verbal, and spatial reasoning – along with a “total intelligence score” which was a composite of the three subscales. Test scores were normed and coded on a nine-point scale ranging from 1 point to 9 point, with higher points indicating higher intelligence. The total score had a mean of 4.76 points and standard deviation of 2.18 (Table 1). Data on criminal offending was obtained from the Central Register for Criminal Records which covers crimes committed between the ages of 15 and 21. As expected, the results showed a negative, mostly linear correlation between criminal offending and total intelligence scores, leading the authors to conclude that “the clearest takeaway from this research is that low intelligence is a strong and consistent correlate of criminal offending” (page 115). Consider the following levels of criminal offending by men of different total intelligence scores (Table A4):
- The “high frequency (%)” is the percentage of subjects with five or more convictions. The “variety index” is the number of distinct “types” of crime, with the “types” of crime being violent, property, DUI, other traffic offenses, sexual assault, and drug offenses.
Several studies have also demonstrated a relationship between cognitive ability and self-control. For example, Petkovsek and Boutwell (2014) used data from the Fragile Families and Child Wellbeing Study (FFCWS) to examine the relationship between IQ and self-control in a sample of nearly 5,000 children born in large U.S. cities between 1998 and 2000. The self-control and intelligence of the children were measured when they reached 9 years of age. Self-control was measured by teacher report. Intelligence was measured using the Peabody Picture Vocabulary Test–III (PPVT-III), the Wechsler Intelligence Scale for Children Digit Span subtest (WISC-IV Digit Span), and two subtests from the Woodcock Johnson–III (WJ-III) battery. Each child’s overall intelligence (or g) score was estimated using a summation of the four different tests. In the regression model on child self-control, overall intelligence had a significant effect (β = −.32, lower scores mean higher levels of self-control), even after controlling for parental self-control, spanking, sex, maternal race, and age (Table 2). The study therefore concluded “higher levels of intelligence in children were consistently related to higher levels of self-control. The association between intelligence and self-control persisted regardless of the covariates included in the analysis (including parental impulsivity and parenting quality)” (page 1242).
Impressively, a longitudinal study analyzed by Boisvert et al. (2013) found that low intellectual achievement is associated with self-control even after controlling for prior measures of self-control. The authors used data from the Panel Study of Income Dynamics (PSID)-Child Development Supplement (CDS) study, a nationally representative longitudinal study of over 2,000 subjects. The subjects in the sample were assessed at childhood (mean age = 8.63 years) and at adolescence (mean age = 14.76 years). Intellectual achievement was measured at both ages using four subtests from the Woodcock-Johnson Revised Tests of Achievement for Reading and Math (WJ-R): Letter-Word Identification, Passage Comprehension, Applied Problems, and Calculations. Self-control was measured by having the child’s primary caregiver answer nine statements about the child’s behavior (e.g., child is impulsive, child cannot sit still, etc.) with each question being answered with a score from 1 (often true) to 3 (not true). The total self-control score ranged from 9 to 27 points (higher scores reflect higher self-control), with an average score of score was 22.86 points and standard deviation of about 3.67 points at adolescence (Table 1). As expected, WJ-R scores during childhood predicted self-control five years later:
Furthermore, the authors found that adolescence intellectual achievement remained associated with self-control even after controlling for race, gender, age, parental involvement, parental withdrawal, and childhood self-control. The authors concluded with the following (page 88):
Overall, the results revealed that intellectual achievement is significantly related to self-control in childhood and adolescence. Specifically, lower scores on the Woodcock-Johnson test were associated with measured decreases in self-control, when controlling for various parenting measures, age, race, and gender. These results remained significant even after controlling for prior measures of self-control obtained 5 years earlier. Thus, results from the current study provide further support for the suggestion that self-control may be interconnected with functions in the prefrontal and temporal cortex of the brain, such as intellectual achievement (Beaver et al., 2007; Ratchford & Beaver, 2009)
Many of the earlier studies provided useful illustrations of the magnitude of the correlation between cognitive ability and criminal offending, but it may be useful to compare these correlations with that of other common predictors of criminality, such as parental socioeconomic level. A number of studies have made these comparisons and have found that cognitive ability correlates with crime about as much as, and sometimes better than, other common predictors of crime. Consider the following studies.
Lynam et al. (1993) [archived] reported data on the Pittsburgh Youth Study cited earlier. In this study, researchers measured the association between IQ, social class (parental occupation and education), and self-reported delinquency at ages 12-13. The study found that delinquency correlated with IQ scores, particularly full-scale IQ and verbal IQ, more than with parental social class. For example, among white youth, delinquency correlated with full-scale IQ (r = −.22) and verbal IQ (r = −.31) more than with parental social class (r = −.11) (Table 1). Also, among black youth, delinquency correlated with full-scale IQ (r = −.25) and verbal IQ (r = −.26) more than with (r = −.11) (Table 1).
Levine (2011) [archived] examined the relationship between criminality and IQ in a representative sample of 11,437 subjects from the United States. IQ and SES were measured in 1981 at youth and incarceration frequencies were measured between 1982 and 2006. About 4.3% of participants were incarcerated during the follow-up period. The results showed significant (negative) correlations between IQ and incarceration and between SES and incarceration. For example, the study found a “moderate” effect size for IQ, with incarcerated participants having a mean IQ that was 0.77 SDs lower than the mean IQ for non-incarcerated participants (100.7 vs 89.6, page 1235). In comparison, SES had a “small effect size”, with incarcerated participants having a mean SES only 0.37 SDs lower than the mean SES IQ for non-incarcerated participants (page 1235).
The previous two studies compared the correlations of IQ and SES with criminality at the individual level. There is also data that compared these correlations at the population level. For example, Beaver and Wright (2011) [archived] used data from the National Longitudinal Study of Adolescent Health to examine the association between county-level IQ and crime rates. They estimated county-level IQ by analyzing data on 20,745 youths from 243 counties within 31 states. IQ was measured using the Peabody Picture Vocabulary Test-Revised (PPVT), which measures verbal skills and receptive vocabulary. They also included a measure of “concentrated disadvantage”, which was a composite scale that included the proportion of the population that was black, the proportion of female-headed household, the proportion of households of households with an annual income under $15,000, the rate of households receiving public assistance, and the unemployment rate. The results showed that the correlation of county-level crime rates with IQ were comparable to the correlation with these other social variables. For example, consider the following correlations (Table 1):
- The property crime rate significantly correlated with the female-headed household rate (r = .56), percent black (r = .56), IQ (r = −.40), the concentrated disadvantage composite (r = .32), the unemployment rate (r = .16), and the public assistance rate (r = .13). There was also a correlation with the percent of households earning less than $15,000 (r = .12), but the correlation was not statistically significant.
- The violent crime rate significantly correlated with the female-headed household rate (r = .76), percent black (r = .60), IQ (r = −.58), the concentrated disadvantage composite (r = .57), the public assistance rate (r = .46), the unemployment rate (r = .46), and the percent of households earning less than $15,000 (r = .23).
- The composite crime rate significantly correlated with the female-headed household rate (r = .68), percent black (r = .51), IQ (r = −.53), the concentrated disadvantage composite (r = .44), the public assistance rate (r = .29), the unemployment rate (r = .27), and the percent of households earning less than $15,000 (r = .15).
There is also some data suggesting that cognitive ability is associated with childhood self-control more than parental social class is. Richmond-Rakerd et al. (2021) [archived] used data from the Dunedin Multidisciplinary Health and Development Study to investigate the relationship between childhood self-control and aging outcomes in midlife. While the study was not directly interested in discovering the association between cognitive ability and self-control, the dataset contained IQ and social class data which was found to be correlated with self-control. The study investigated 1,037 individuals born in Dunedin, New Zealand between 1972 and 1973. IQ was measured using the Wechsler Intelligence Scale for Children-Revised when children were ages 7, 9, and 11 years. Self-control was measured every two years between the ages of 3 and 11 years based on “observational ratings of children’s lack of control; parent and teacher reports of impulsive aggression; and parent, teacher, and self-reports of hyperactivity, lack of persistence, inattention, and impulsivity” (page 8). Childhood social class was also measured based on parental income and educational levels. Researchers found that childhood self-control was significantly correlated with childhood IQ (r = .45), even more than with childhood social class (r = .27) (Table 1).
Health and mortality
Cognitive epidemiology integrates differential psychology and epidemiology to study the relationship between cognitive ability and health outcomes. Researchers in this field have recently discovered that cognitive ability is a powerful predictor of mental and physical health and illness, as well as mortality. For a recent review of the association between cognitive ability and health outcomes, see Deary et al. (2021) [archived].
First, consider mortality. A meta-analysis by Calvin et al. (2011) [archived] investigated the association between youth intelligence and life expectancy across 16 studies. The studies comprised 22,453 deaths among 1,107,022 participants from five Western countries: the United Kingdom (n = 7), the United States (n = 5), Sweden (n = 2), Australia (n = 1) and Denmark (n = 1). The mean age at time of testing ranged between 7 and 20 years depending on the study. The follow-up duration ranged between 17 and 69 years. The researchers found that a 1 standard deviation increase in cognitive test scores was associated with a 24% lower risk of death (Table 2). When focusing on individuals followed up 40-69 years after testing, a 1 standard deviation increase in test scores was associated with a 20% lower risk of death. Impressively, the association between cognitive ability and mortality did not seem to be driven by childhood SES; only 4% of the association was attenuated after adjusting for childhood SES. In other words, 96% of the association between cognitive ability and mortality persisted after adjusting for childhood SES. The meta-analysis concluded with the following:
The present meta-analysis of 16 published prospective cohort studies, comprising over 1.1 million participants and 22,453 deaths, demonstrates and quantifies the consistently-reported association between higher premorbid intelligence and lower mortality risk. A 1-SD advantage in intelligence in childhood and youth was associated with a 24% lower risk of mortality. The effect was similar in men and women, and was not explained by socio-economic differences in early life, as indicated by parental occupation or income. The association was attenuated by approximately a third after adjusting for adult SES and by approximately a half after adjusting for educational experience. Intelligence remained a predictor of mortality after these attenuating effects, and removal of one study that carried by far the largest weighting in the models did little to change the magnitude of these effects.
As hinted in the conclusion, after adjusting for adult SES and education, the association between cognitive ability and mortality was attenuated by about 34% and 54% respectively. This may suggest that the effects of cognitive ability on mortality may be mediated through these variables. In other words, part of the reason why cognitive ability effects mortality may be due to its effect on adult SES and education, which then have an effect on mortality.
The association between early cognitive ability and mortality has been replicated in many later studies across many different countries. For example, consider the following recent studies:
- Christensen et al. (2016) [archived] examined more than 700,000 men to test whether the “well-established” association between early intelligence and morality differed by cause of death for a large cohort of men in Denmark. The mean age at time of assessment was about 19 years (Table 2) and the mean follow-up was about 37 years after assessment (page 3). Intelligence was assessed using a test called the Børge Prien Prøve (BPP) which comprised four subtests assessing logical, verbal, numerical, and spatial abilities. The mean age-adjusted hazard ratio (HR) for mortality per 1 standard deviation decrease in BPP score was 1.28 (in other words, there is a 28% higher risk of death per 1 SD decrease in BPP score). After controlling for educational level at time of assessment, the HR decreased to 1.21. There was an inverse association between intelligence and mortality for all causes of death except for skin cancer. The causes of death most associated with lower intelligence were homicide and respiratory disease, with respective hazard ratios of 1.65 and 1.61, which reduced to 1.31 and 1.45, respectively, after adjusting for educational level (Table 3).
- Bratsberg and Rogeberg (2017) examined the majority of Norwegian males born in the 1962-1990 birth cohorts (n = 720,261) to measure the relationship between IQ at age 18-19 and mortality by age 40. IQ was transformed to an “ability score”, which is a score on a 9-point scale (1-9) with a median score of 5 points and standard deviation of 2 points. There were large differences in mortality risk found by ability. In fact, individuals with ability scores of 3 (1 SD below the mean) and 1 (2 SDs below the mean) had mortality risks 1.53 and 2.15 times that of those with median ability. By contrast, individuals with ability scores of 7 and 9 had mortality risks that .66 and .43 times that of those with median ability. These relative risks were all after controls for birth year and parental SES (see Table 2, column 4).
- Čukić et al. (2017) [archived] examined nearly the entire Scottish population born in 1936 (n = 70,805) to track the relationship between IQ measured at age 11 and mortality 68 years later. This is about 94% of the 1936-born population of Scotland. The study found that the hazard risk for mortality risk for IQ was 0.8. In other words, each 1 standard deviation increase in IQ was associated with a 20% reduction in mortality risk. The comparisons of mortality risks at the extremes of IQ were particularly impressive. Men and women in the highest decile of IQ at age 11 were 50% and 41%, respectively, as likely to die as men and women in the lowest decile by age 79 (Table 3). In other words, individuals in the lowest decile of IQ at age 11 were 2 times as likely (for men) and about 2.4 times as likely (for women) to die by age 79.
A number of studies have also shown significant correlations between early cognitive ability and later risk of unintentional injuries. The largest such study was perhaps done by Whitley et al. (2009) [archived]. This study analyzed a cohort of 1,109,475 Swedish men to examine the relationship between youth IQ and later risk of unintentional injury. The cohort comprised all non-adopted men born in Sweden from 1950 to 1976 with both biological parents identified in the Multi-Generation Register. IQ was measured during mandatory military service conscription examination that occurred between 1969 and 1994, when the participants were between 16 years and 25 years old (average age was 18 years old). The IQ tests comprised four subtests which represented verbal, logical, spatial, and technical abilities. IQ scores were standardized to fit a Gaussian-distributed score between 1 and 9, with higher values indicating greater cognitive ability. Rate of unintentional injuries were measured using hospital admissions records between 1969 and 2006 for drowning, poisoning, fire, road traffic accidents, medical complications, and falling. The average follow-up duration of 24 years. About 18% of men had at least one admission for an unintentional injury during the follow-up period. The study found that lower youth IQ was significantly associated with elevated risk of each type of unintentional injury. For example, the results were as follows:
- Each standard deviation decrease in IQ was associated with a 15% higher risk of any unintentional injury, a 53% higher risk of poisoning, a 36% higher risk of fire, a 25% higher risk of road traffic accidents, a 20% higher risk of medical complications, and a 17% higher risk of falling.
- Relative to subjects with IQ scores between 7 and 9, participants with scores between 1 and 2 were 1.59 times as likely to have a hospital admission for falling, 1.99 times as likely for a road accident, 2.55 times as likely for poisoning, and 1.71 times as likely for medical complications (Table 4).
Batty et al. (2009) [archived] studied the same dataset to find that lower youth IQ was also significantly associated with elevated risk of mortality due to unintentional injuries. For example, relative to individuals with the highest IQ scores (7-9), individuals with the lowest IQ scores (1-2) were 5.82 times as likely to die from poisonings, 4.39 times as likely to die from fire, 3.17 times as likely to die from falls, 3.16 times as likely to die from drowning, and 2.17 as likely to die from road injury.
Similar findings were also found in other countries. For example, Osler et al. (2007) [archived] examined the relationship between early cognitive ability and later risk of unintentional injury in a sample of 11,532 males born in Copenhagan, Denmark in 1953. The participants were followed from 1978 until 2001. IQ was assessed at age 12 using information the Härnquist test (a tests that consists of spatial, arithmetic, and verbal subtests) and at age 18 using a cognitive test administered during military conscription, which is mandatory for all men on attaining the age of 18 years. IQ measured at both ages was significantly associated with death or hospital admissions due to unintentional injury. For example, the hazard ratio for death or injury per standard deviation increase in age-12 IQ was 0.82 for any form of trauma, 0.87 for a road traffic injuries, 0.77 for falls, and 0.64 for poisoning (excluding medicines) (Table 4). The same hazard ratio per standard deviation increase in age-18 IQ was 0.78 for any form of trauma, 0.77 for road traffic injury, 0.74 for falls, and 0.52 for poisoning (excluding medicines) (Table 5).
Similar findings were reported by Lawlor et al. (2007) [archived], which found that IQ measured at age 7 was associated with hospital admissions for unintentional hospital admissions in adulthood.
There is a also a well-established relationship between cognitive ability and physical health and illness. For example, a review article by Deary et al. (2019) [archived] summarized the literature on this association as follows:
In addition to mortality, intelligence test scores are associated with lower risk of many morbidities, such as cardiovascular disease, cerebrovascular disease, hypertension, cancers such as lung cancer, stroke, and many others, as obtained by self-report and objective assessment [12, 13, 14]. Higher intelligence in youth is associated at age 24 with fewer hospital admissions, lower general medical practitioner costs, lower hospital costs, and less use of medical services, and intelligence appeared to account for the associations between education and such health outcomes [15,16]. Higher intelligence is related to a higher likelihood of engaging in healthier behaviours, such as not smoking, quitting smoking, not binge drinking, having a more normal body mass index and avoiding obesity, taking more exercise, and eating a healthier diet [16, 17, 18].
The most frequent form of physical illness that has been studied in relation to intelligence has been cardiovascular disease, probably because, as Deary et al. (2010) [archived] note, this is “the most common cause of death and disability is cardiovascular disease”. A number of studies have found significant associations between childhood cognitive ability and later risk of cardiovascular disease. For example, Hart et al. (2004) [archived] examined 938 participants from the Midspan prospective cohort studies in Scotland to measure the association between youth IQ and later cardiovascular disease. Childhood IQ was measured when participants were 11 years old using the Moray House Test in 1932. Cardiorespiratory studies were conducted on the participants in the 1970s. Each one standard deviation decrease in childhood IQ was associated with a 11% increased risk of cardiovascular disease, a 16% increased risk of coronary heart disease, and a 10% increased risk of stroke (defined as either a hospital admission or death) (Table 2). One surprising finding was that childhood IQ was associated with these outcomes only before age 65, not after age 65. That is, the risk of these outcomes after age 65 was not correlated with childhood IQ. As a result, if we focus on these outcomes before age 65, the associations with childhood IQ are even large than reported earlier: for hospital admissions or death before age 65, each one standard deviation decrease in childhood IQ was associated with a 22% increased risk of cardiovascular disease, a 29% increased risk of coronary heart disease, and a 47% increased risk of stroke.
Childhood cognitive ability is associated with other health problems as well. This can been demonstrated in the United States using data from the National Longitudinal Survey of Youth (NLSY-1979), an ongoing longitudinal study with data on early cognitive ability and adult health outcomes for over 5,000 participants. Cognitive ability was measured using the AFQT when participants were 15 to 23 years old. A number of studies have reported that childhood ability is associated with several health outcomes and health behaviors until at least age 50. Consider the following:
- Wraw et al (2015) [archived] used data from the NLSY-1979 to examine the correlation between childhood cognitive ability and for 16 different health outcomes at age 50. Researchers found that 13 of the health outcomes were significantly and negatively correlated with childhood IQ. For example, 1 SD increase in IQ was associated with: 70% greater odds of having good, very good, or excellent health (Table 2), 20% lower odds of high blood pressure or hypertension, 29% lower odds of chronic lung disease, 21% lower odds of general heart problems, 35% lower risk of stroke, and 41% lower odds of heart attack. Adjusting for childhood SES did not substantially attenuate the correlations between childhood IQ and adult health outcomes, leading the authors to conclude that “it appears that social background does not account for premorbid intelligence–health associations”.
- Wraw et al. (2018) [archived] also used the NLSY-1979 to examine the link between childhood cognitive ability and health behaviors at middle age (mid age = 51.7 years). Consistent with prior studies, higher youth IQ was associated with a number of positive health behaviors. For example, 1 SD increase in youth IQ was associated with 25% lower odds of consuming a sugary drink in the previous week, 33% lower odds of drinking alcohol heavily (6+ drinks in one occasion in the past 30 days), 40% lower odds of being a smoker, 47% higher odds of flossing. Interestingly, the 1 SD increase in IQ was also associated with 58% higher odds of having drunk in the past 30 days (so more intelligent participants were more likely to drink occasionally drink, despite being less likely to drink heavily), 10% higher odds of having skipped any meals in the past week, and 37% higher odds of having snacked between meals in the past week. Also, similar to the previous study, the authors found that “the associations in the present study appeared to be largely independent of early life socio-economic position”, suggesting that early SES does not confound the relationship.
Similar results can be found using data from Great Britain. This can be shown using the 1970 British Cohort Study, an ongoing longitudinal study involving those born in Great Britain in 1970. This dataset contains information on various health outcomes at age 30 and cognitive ability at ages 5 and 10 years for over 8,000 participants. Several studies have found that cognitive ability measured early in life (sometimes as early as 5 years of age) is linked to a variety of health outcomes. Consider the following studies:
- Batty et al. (2007) [archived] examined the association between mental ability at childhood and risk factors for premature mortality at adulthood. Risk factors for premature mortality were measured using self-reports about their height, weight, and their history of smoking, high blood pressure, and diabetes. Childhood ability measured as early as age 5 years was associated with risk factors at age 30. For example, each 1 SD increase in mental ability at age 5 was associated with 14% lower odds of being a current smoker, 23% higher odds of having given up smoking (among those who smoked in the past), 11% lower odds of being overweight, and 16% lower odds of being obese (Table 3). There was no relationship between childhood IQ and diabetes. Also, consistent with the previous studies, adjusting for early SES only attenuated about one-quarter or less of these associations.
- Batty et al. (2007) studied the relation between childhood mental ability and adult habits regarding food consumption and physical activity. Participants were asked how often they consumed different kinds of foods, including fresh fruit, cooked vegetables, red meat, poultry, fish, eggs, fried food, etc. They were also asked if and how often they participated in physical activity. Researchers found that participants with higher cognitive ability at age 10 were more likely to engage in healthier dietary practices. For example, those with higher ability at age 10 were more likely to have greater frequency of consumption of fresh fruit (odds ratio per 1 SD increase in ability = 1.25), cooked vegetables (1.26), wholemeal bread (1.22), fish (1.25), and food fried in vegetable oil (1.23) (Table 2). Those with higher childhood ability were also more likely to take regular exercise (1.21) and get out of breath or sweaty more frequently when exercising (1.35) (Table 3). Adjustment for childhood social class did not substantially attenuate these relationships.
This association between childhood cognitive ability and adult health outcomes and behaviors has been replicated in many other countries, many with rather accessible health care systems, such as Luxembourg (Wrulich et al. 2013), Denmark (Batty et al. 2005), and Scotland (Batty et al. 2006).
There is a also a well-established relationship between cognitive ability and mental health. For example, a review article by Hill et al. (2019) [archived] summarized the literature on this association as follows:
Individual differences in intelligence are predictive of mental illness, where a higher level of intelligence in childhood is predictive of a lower level of self-reported psychological distress decades later . This link between intelligence and mental illness also extends to severe psychiatric conditions where individuals who have a level of intelligence one standard deviation below the mean have, on average, a 60% greater chance of being hospitalized for schizophrenia, a 50% increase of being diagnosed with a mood disorder, and a 75% greater risk for having an alcohol-related disorder, over a two-decade follow up period . A higher risk for several psychiatric illnesses has also been associated with a lower level of intelligence, including major depressive disorder (MDD) [6,7], autistic spectrum disorder (ASD), attention/deficit hyperactivity disorder (ADHD) , as well as bipolar disorder [9,10], although a higher level of intelligence, particularly as measured by tests of crystallized ability, may also be a risk factor for bipolar disorder .
The largest study from that excerpt was conducted by Gale et al. (2010) [archived]. These authors analyzed 1,049,663 Swedish men that took intelligence tests during mandatory military conscription when they were around 18.3 years of age. Participants were followed up for hospital admissions for mental disorders after a mean period of 22.6 years. IQ scores were standardized to fit a Gaussian-distributed score between 1 and 9, with higher values indicating greater cognitive ability. The excerpt has already reported the increased risk of hospital admissions associated with each standard deviation decrease in IQ (e.g., 60% greater risk of hospitalization for schizophrenia per 1 SD decrease in IQ). The gap in hospitalization rates for individuals at the extremes of the IQ distribution were expectedly more drastic. For example, relative to individuals with an IQ score category of 9, individuals with an IQ score category of 1 were more than 7 times as likely to be hospitalized for schizophrenia (5.16 vs 0.68 per 1,000 person-years), more than 5 times as likely to be hospitalized for mood disorders (5.98 vs 1.05), more than 20 times as likely to be hospitalized for alcohol-related disorders (16 vs 0.78), and more than 25 times as likely to be hospitalized for other substance-use disorders (7.52 vs 0.27) (see appendix).
The study on psychological distress was conducted by Gale et al. (2009) [archived]. The researchers used data on two British birth national birth cohorts to examine the relationship between general cognitive ability measured at youth (age 10 and 11) and psychological distress at adulthood (age 30 or 33). The cohorts used were the 1958 National Child Development Survey (N = 6,369) and the 1970 British Cohort Study (N = 6,074). The study found that a standard deviation increase in IQ was associated with a 39% and 23% reduction in psychological distress in the 1958 and 1970 cohorts, respectively. Psychological distress was measured using the Rutter’s Malaise Inventory, a 24-item self-completion scale that asks questions such as “Do you often feel depressed?”, “Are you easily upset or irritated?”, etc. with a “yes” response counting as one point to one’s total score. A score of 7 or more was used to indicate presence of psychological distress, since this has been shown to identify cases of clinically diagnosed depression (page 594).
Contrary to the rest of the data in this post, there is evidence that cognitive ability is associated with a specific set of negative mental health outcomes. For example, Lopez et al. (2010) [archived] meta-analyzed 30 peer-reviewed studies and found that people with anorexia nervosa had a mean IQ that was 6 – 11 points higher than the mean IQ of the general population. There is also some data suggesting that bipolar disorder is more common with individuals with extremely high IQ. For example, consider Gale et al. (2012) [archived] which found that, while there is an inverse correlation between intelligence and bipolar disorder, there was higher risk for bipolar disorder among those with the highest intelligence in a prospective cohort study of 1,049,607 Swedish men.
The above studies have demonstrated that cognitive ability is significantly associated with health and mortality. At this point, it may be useful to compare the associations between health outcomes and cognitive ability with the associations between health outcomes and other known risk factors. Batty et al. (2010) [archived] performed such a comparison in a population-based study that analyzed a cohort of 1,145 men from the West of Scotland Twenty-07 study. The authors compared the association between mortality and IQ with the association between mortality and well-established risk factors including smoking, body mass index, high blood pressure, physical inactivity, and socioeconomic disadvantage. The associations for each of these predictors were compared by using the relative index of inequality (RII), which indicates the relative odds of mortality for the most disadvantaged persons with respect to a given predictor relative to the most advantaged persons. The relative index of inequality for total mortality for the top 5 risk factors was 4.60 for cigarette smoking, 3.48 for IQ, 2.90 for income, 2.27 for physical activity, 2.07 for education, and 1.84 for occupational social class. The study summarized the findings as follows:
[I]n sex-adjusted analyses, cigarette smoking (ranked first), IQ (second) and income (third) were most strongly related to mortality risk, whereby the most disadvantaged experienced over three times the risk of death relative to the advantaged. These were followed in order of the magnitude by a group of risk factors — physical activity, education and occupational social class — which were associated with around a doubling of mortality risk. Following mutual adjustment for all the risk factors, these gradients were attenuated, although the order of magnitude of the effect estimates was broadly retained such that IQ continued to be ranked the second strongest predictor.
The authors concluded as follows:
In conclusion, in this, the first study to examine the relative strengths of IQ and established risk factors for total and CVD mortality, there was evidence that low IQ was one of the most powerful predictors
The above studies show that cognitive ability is an excellent predictor of many important life outcomes, including academic achievement, occupational performance, socioeconomic outcomes, anti-social behavior, and health. Now, these studies only show that cognitive ability is predictive, so it’s still an open question as to whether cognitive ability is causal. For example, one might say that the correlation between cognitive ability and these outcomes is the result of confounding with a common cause such as family background or personality. In a later post, I show evidence that cognitive ability also predicts important life outcomes after controlling for common confounders, indicating that cognitive ability is actually causal.
Even without showing evidence that cognitive ability is causal, cognitive ability is still important because of it’s predictive powers. For example, imagine that cognitive ability has no causal impact on any of the above outcomes, but is instead only correlated with the outcomes because of shared association with family background, personality, self-control, and a host of other confounders that we may or may not be able to reliably measure. Even if this is the case, cognitive ability would still be an excellent measurable index of a person’s expected future success. It would still be useful to measure the cognitive ability of children in order to reasonably know whether they are on the right track to success. That being said, I believe the evidence shows that cognitive ability is in fact causal, which I demonstrate in my next post.
- 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.
- 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. They show they cognitive ability tests are valid for predicting success and performance and that this validity is not an artifact of socioeconomic status.
- Kuncel et al. (2010). “Individual differences as predictors of work, educational, and broad life outcomes” [archived]. A brief report on the predictive validity of various measures of individual differences for important life outcomes. Focuses on sources of individual differences other than just cognitive ability (e.g., personality, admissions tests, etc.).
- 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 APA article.
- Malanchini et al. (2020). “Cognitive ability and education: how behavioural genetic research has advanced our knowledge and understanding of their association” [archived]. This article reviews the research on the association between cognitive ability and academic achievement and educational attainment. Also explores possible genetic and environment explanations of the association.
- Deary et al. (2021). “Intelligence, health and death” [archived]. This article gives a good review of research in the field of cognitive epidemiology, which studies the association between cognitive abilities and health outcomes. The review covers possible causes of the association and covers recent genetic research that may provide insight on the association.
- Linda Gottfredson (1997). “Mainstream Science on Intelligence: An Editorial With 52 Signatories, History, and Bibliography” [archived]. This is a very brief 3-page statement that outlines conclusions regarded as mainstream among researchers on intelligence. The statement was signed by 52 experts in intelligence and allied fields to promote more reasoned discussion of research in the field.
- Charlie Reeve and Jennifer Charles (2008). “Survey of opinions on the primacy of g and social consequences of ability testing: A comparison of expert and non-expert views” [archived] examined the opinions of 30 experts in the science of mental abilities about their views on cognitive ability and cognitive ability testing. This study was a replication of Murphy, Cronin, and Tam (2003) [doi].
- Heiner Rindermann (2020). “Survey of expert opinion on intelligence: Intelligence research, experts’ background, controversial issues, and the media” [doi] surveyed the opinions of over 100 experts in the field of intelligence about a variety of topics concerning intelligence testing, including the validity of IQ tests, the relationship between intelligence and success in Western societies, the impact of genes on group differences in intelligence.
- Strenze (2007). “Intelligence and socioeconomic success: A meta-analytic review of longitudinal research” [archived]. Meta-analyzed dozens of studies involving hundreds of thousands of participants reporting the correlation between youth intelligence and adulthood educational attainment, occupational status, and income.
- Strenze (2015). “Intelligence and Success” [archived]. Reviews several meta-analyses showing the correlation between intelligence and a variety of conventional measures of success, such as educational attainment, income, job performance, etc. and less conventional measures of success such as physical attractiveness, happiness, popularity, and creativity.
- Ttof et al. (2016). “Intelligence as a protective factor against offending: A meta-analytic review of prospective longitudinal studies” [archived]. Meta-analyzed 15 studies investigating the extent to which intelligence functions as a protective factor against various risk factors for delinquency, violence, and crime.
- 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.