The predictive validity of cognitive ability (OLD)

Last Updated on September 28, 2021

NOTE: this post is outdated. An updated version of this post has been posted here.

There is overwhelming evidence demonstrating the undeniable predictive validity of cognitive ability. Cognitive ability measured as early as age 6 has a strong association with one’s future success in a number of important outcomes, including income, educational attainment, academic performance, occupational prestige, occupational performance, criminality, etc. These associations are robust, persisting even after controlling for a number of plausible confounding variables, including parental socioeconomic status, race, job training and job experience, and other risk factors for the relevant life outcomes. The totality of evidence heavily implies that this association is causal, indicating that early cognitive ability is a powerful factor in determining a person’s chances of achieving conventional measures of success in Western societies. In this post, I will cite scholarly evidence demonstrating each of these claims. I will begin by first describing my working definition of “cognitive ability” and then explaining a few concepts that must be understood to interpret the evidence that follows.


Working definition of cognitive ability

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 precise working definition, it may be useful to frame my definition from the perspective of different theories of intelligence. My working definition of “cognitive ability” corresponds to the visual-spatial, linguistic-verbal, and logical-mathematical forms of intelligence stipulated by Howard Gardner’s Theory of multiple intelligences. It also corresponds to the “analytic” component of intelligence stipulated by Robert Sternberg’s Triarchic theory of intelligence.

I want to emphasize that my working definition of cognitive ability is not an endorsement of any particular theory of intelligence. I make no claims whatsoever about whether creativity, bodily-kinesthetic intelligence, musical intelligence, etc. are “real” forms of intelligence. My definition of cognitive ability corresponds to “analytic” intelligence, not necessarily because I think analytic intelligence is the only “true” form of intelligence, but because analytic intelligence has a number of properties that make it worth investigating such as, e.g., predictive validity and stability (as I will show in this post) which have not been demonstrated for other theorized forms of intelligence.

IQ

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).

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). Later in this post, I will provide more data useful to understand how to interpret more specific IQ ranges between 70 and 130, i.e. what life outcomes we can expect from someone with an IQ in the 80-90 range compared to someone with with an IQ in the 110-120 range.

The g factor

One seeming problem with measuring cognitive ability with IQ tests is that there are a large variety of different IQ tests, e.g. some popular forms include Raven’s Progressive Matrices, the Wechsler Adult Intelligence Scale, the Stanford–Binet Intelligence Scales, the Woodcock–Johnson Tests of Cognitive Abilities etc. Which of these tests should be considered to be the measure of cognitive ability? Furthermore, given that different tests seem to measure different components of cognitive ability (e.g., visual-spatial reasoning, verbal ability, logical reasoning, etc.), which tests measure true cognitive ability?

The first thing to note in answering these questions is that different IQ tests and subtests tend to be 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 gGottfredson (2002) [archived] gives two reasons for why tests with high g-loadings are particularly 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 Gottfredson (2002) [archived] to forcefully pronounce that (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.

That being said, it should be clear I don’t need to pick which specific IQ test is the “real” measure of cognitive ability. All IQ tests measure g to some extent and g is the primary element of my working definition of cognitive ability. The importance of g (rather than any particular mental traits measured by IQ) is important to recognize because there may be situations where we do not have access to IQ data. In those situations, we can use other mental tests that have high g-loadings as measures of cognitive ability. For example, IQ tests, SAT tests, AFQT tests, and even vocabulary tests have very high g-loadings, which make them satisfactory measures of cognitive ability (on my working definition). Now, ideally we will have direct measures of g. But data often isn’t ideal. There will be many studies cited that use some of these other kinds of tests which serve as good proxies of g.

Someone might be surprised that vocabulary has a g-loading. Why would the general factor of intelligence have such a strong association with vocabulary, which seems to be determined by the richness of one’s environment rather than their general cognitive ability? The response to this is that a person’s vocabulary is determined by both their environmental exposure and their general cognitive ability. There are a few reasons why this might be the case. One reason may be that high-g individuals are more likely to seek out cognitively stimulating environments that facilitate vocabulary development. For example, Neisser et al. (1996) state that “In a society in which plenty of words are available in everyone’s environment-especially for individuals who are motivated to seek them out–the number of words that individuals actually learn depends to a considerable extent on their genetic predispositions” (page 85). Another reason may be that if you have two persons with the same environmental exposure to language, the person with higher general cognitive ability may be more efficient at employing the underlying cognitive processes (i.e, reading context, inferring intention in language use, etc.) that facilitate vocabulary development.

Finally, I would like to note that, although I take g to be the primary element of my working definition of cognitive ability, I also secondarily include non-g mental abilities in my definition as well. I do this because there appears to be evidence that non-g mental abilities have incremental predictive validity of success (above and beyond g-factors alone) for the outcomes that I’m concerned with. For example, Kell and Lang (2017) reviewed three studies suggesting that non-g factors have incremental predictive validity for job performance. Another review by Coyle (2018) also reported that a number of studies “confirm the predictive power of non-g factors” for a diverse range of outcomes, including GPAs, choice of college major, college degree attainment, and attainment of STEM jobs. Therefore, my working definition of cognitive ability includes both primary and secondary components: (1) the general factor of intelligence, or g, as the primary component, and (2) specific non-g factors (e.g., the non-g components of spatial ability, verbal ability, etc.) as secondary components insofar as these factors have predictive validity.

Stability

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. Consider the following.

  • 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).
  • 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) 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 restrictions, the estimated correlations rose to r=0.73.

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.

Predictive validity


Expert consensus

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. See the following reports/surveys.

Gottfredson (1997) [archived] reports that “IQ is strongly related, probably more so than any other single measurable human trait, to many important educational, occupational, economic, and social outcomes.” (page 14). This was published in a very brief 3-page statement that outlines conclusions regarded as mainstream by over 50 experts in intelligence and allied fields.

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.

High cognitive ability predicts success

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)

A meta-analysis by Strenze (2007) [archived] shows that intelligence (measured by IQ scores) is one of the best predictors 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.

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 seem to have much of an impact on the IQ-success correlation (Table 2). Regarding age at success, the correlation between youth IQ and education 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).

IQ is also a great predictor of occupational performance. In fact, Gottfredson (1997) [archived] has forcefully asserted that “g can be said to be the most powerful single predictor of overall job performance” (page 83) partially because “no other single predictor measured to date (specific aptitude, personality, education, experience) seems to have such consistently high predictive validities for job performance.” She further 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).

Strenze (2015) [archived] cites several meta-analyses showing the correlation between intelligence and a variety of measures of success (Table 25.1). The results showed large correlations between intelligence and academic performance in primary education (r=0.58), educational attainment (r=0.56), job performance (0.38-0.53, depending on the sample), occupational attainment (r=0.43), skill acquisition in work training (r=0.38), and several other metrics of success.

Low cognitive ability predicts social dysfunction

Low IQs predict a wide range of negative social outcomes. For example, in an article of Scientific AmericanGottfredson (1998) [archived] reports the following outcomes for non-Hispanic whites of various youth IQ scores:

  • Of those with IQs in the normal range (90-110), 6% live in poverty, 6% are High School dropouts, 8% of women are chronic welfare recipients, and 3% have been incarcerated
  • Of those with IQs between 75-90, 16% live in poverty, 35% are High School dropouts, 17% of women are chronic welfare recipients, and 7% of the men have been incarcerated
  • Of those with IQs below 75, 30% live in poverty, 55% are High School dropouts, 31% of women are chronic welfare recipients, and 7% of men have been incarcerated.

Low IQ scores predict a low likelihood of holding skilled and prestigious occupations. For example, Hauser (2002) [archived] found:

  • Half of all janitors have an IQ just above 90 (Figure 12), slightly more than 25% have IQs above 100, and very few (slightly more than 5%) have IQs above 110.
  • The average electrical engineer has an IQ over 110, a small minority (<25%) have IQs below 100, and very few (<5%) have IQs below 90.

Generally speaking, the likelihood of a sub-90 person attaining an occupation that requires complex cognitive processing (e.g. doctors, engineers, professors, analysts, etc.) is very low. They are more likely to be found in unskilled or low-skilled labor (e.g. janitors, manual labor, etc.). Gottfredson (1997) [archived] has also emphasized the scant occupational opportunities for low-IQ individuals (page 90):

…virtually all occupations accommodate individuals down to IQ 110, but virtually none routinely accommodates individuals below IQ 80 (WPT 10). Employment options drop dramatically with IQ-from virtually unlimited above IQ 120 to scant below IQ 80. Such options are virtually nonexistent today (except in sheltered settings) for individuals below IQ 70 to 75, the usual threshold for borderline mental retardation

Individuals with low IQs tend to be unable to complete all but the simplest of tasks. In fact, 10 U.S. Code § 520 [archived] outlaws low-IQ individuals from participating in the Armed Forces. Gottfredson (1997) [archived] cites evidence that these requirements are not arbitrary; the requirements are necessary 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).

Also, see Gottfredson (1998) [archived] for more on the negative consequences of low IQ.

Interpreting correlation coefficients

Many of the above studies have quantified the association between cognitive ability and various life outcomes by reporting correlation coefficients between the two kinds of variables. These correlation coefficients may be uninterpretable for someone unfamiliar with the typical correlation coefficients in social sciences. One way to interpret a correlation coefficient is to square the correlation coefficient to determinate what proportion of variance is explained by the predictor variance. While this is technically correct, this way of expressing the meaning of a correlation coefficient is misleading and often suggests that a predictor variable has less importance than it actually has. For example, imagine that we have a correlation of 0.30 between a predictor variable and outcome variable. This might seem like a small effect size since it only explains 0.3*0.3 = 9% of the total variance. However, Sackett et al. (2008) described a better way of explaining the significant of this correlation, by calculating the odds of having an above-average value for the outcome variable conditional on a certain value (or range of values) of the predictor variable (page 216):

As long ago as 1928, Hull criticized the small percentage of variance accounted for by commonly used tests. In response, a number of scholars developed alternate metrics designed to be more readily interpretable than “percentage of variance accounted for” (Lawshe, Bolda, & Auclair, 1958; Taylor & Russell, 1939). Lawshe et al. (1958) tabled the percentage of test takers in each test score quintile (e.g., top 20%, next 20%, etc.) who met a set standard of success (e.g., being an above-average performer on the job or in school). A test correlating .30 with performance can be expected to result in 67% of those in the top test quintile being above-average performers (i.e., 2 to 1 odds of success) and 33% of those in the bottom quintile being above-average performers (i.e., 1 to 2 odds of success). Converting correlations to differences in odds of success results both in a readily interpretable metric and in a positive picture of the value of a test that “only” accounts for 9% of the variance in performance

See Funder and Ozer (2013) and Rosenthal and Rubin (1982) for more statistically-informed arguments against explaining the magnitude of a correlation coefficient based on the percentage of variance explained. There are at least two other methods to express the meaning of the magnitude of the effect size of intelligence on life outcomes. Both involve comparing a correlation coefficient to other benchmarks.

One method involves comparing the correlation between a given outcome and intelligence with the correlation between the same outcome and other known predictors of the outcome. Consider the following examples:

  • A meta-analysis by Richardson et al. (2012) reported correlations between college GPA and a number of psychological correlates. College GPA correlates positively with socioeconomic status (r=0.11), SAT scores (r=0.29), ACT scores (r=0.40), and high school grades (r=0.40) (Table 6). For comparison, college GPA correlates with intelligence at r=0.20. (Note that because these correlations are limited to an elite portion of the population – i.e. university students – these correlations may understate the effects of these predictors due to range restriction).
  • A meta-analysis by Strenze (2007) reported the correlations between an adult’s SES and their parental SES. There were significant positive correlations between an individual’s educational attainment and their parents’ education attainment (r=0.40), between an individual’s occupational status and their father’s occupational status (r=0.28), and between an individual’s income and their parental income (r=0.16). For comparison, youth IQ had larger correlations with each of these outcomes than did parental SES (Table 1).
  • Koenig et al. (2008) [archived] reported the correlation between cognitive ability and performance on standardized tests reported by. IQ correlated significantly with SAT total score (r=0.82) and ACT total score (0.77) (Table 2). For comparison, these correlations are greater than the correlation between SAT Math and SAT verbal (r=0.74) and the correlation between ACT math and ACT verbal (r=0.67).

Another method involves comparing the correlation between a given outcome and intelligence to the distribution of correlations found in the field of differential psychology. Gignac and Szodorai (2016) collected a large sample of meta-analytically derived correlations published in the field of individual differences. Researchers gathered a total of 708 observed correlations from a sample of 87 meta-analysis. They found that the 25th, 50th, and 75th percentiles corresponded to correlations of 0.11, 0.19, and 0.29, respectively. Only about 10% of the correlations exceeded 0.40 (Table 1), and only about 2.7% of correlations exceeded 0.50 (page 75). Because of these findings, the authors recommended that the normative guidelines for small, medium, and large correlations should be 0.10, 0.20, and 0.30, respectively.

Given these normative guidelines, and given the correlations between intelligence and life outcomes reported in a meta-analysis review by Strenze (2015), the magnitude of the correlations between cognitive ability and life outcomes ranges from very small to very large depending on the outcome measured. As a general rule, the more the outcome requires cognitive complexity, the larger the correlation between cognitive ability and success on the outcome (Table 25.1). For example, cognitive ability has a very small correlation with happiness (r=0.05), a small correlation with popularity (r=0.10), a medium correlation with income (r=0.20), a very large correlation with occupational attainment (r=0.43), and even larger correlations with job performance (r=0.53) and educational attainment (r=0.56).

Controlling for confounders


In order to demonstrate that variable X has a causal influence on Y, one must demonstrate the following three conditions (page 146):

  1. There is an empirical association between X and Y
  2. X occurs before Y
  3. There is reason to believe that the association between X and Y is not spurious.

The above studies have fulfilled the first two of these requirements because they show strong correlations between an individual’s cognitive ability and that individual’s future socioeconomic success. However, the data above has not given much evidence that the association between cognitive ability and later socioeconomic success is not spurious. For any two correlated variables X and Y, there may be many plausible explanations of this correlation. One such explanation involves positing a causal relationship between X and Y. But another plausible explanation might be that some third variable Z (a confounding variable) causes both X and Y. Therefore, in order to show that a correlation between X and Y is not spurious (i.e. that X causes Y), one must rule out alternative plausible explanations by controlling for plausible confounding variables.

The studies in this section will show that the association between cognitive ability and socioeconomic success persist even after controlling for many plausible confounding variables. First, I cite findings showing that cognitive ability remains associated with occupational performance after attempts to control for job training or experience. Next, I cite studies showing that youth cognitive ability remains positively associated with future socioeconomic success. Finally, I cite studies showing that youth cognitive ability remains negatively associated with criminal offending after controlling for youth socioeconomic status and other risk factors for criminal activity.

Occupational performance

Gottfredson (1997) [archived] has reported that cognitive ability predicts occupational performance independently of training. In fact, some organizations have attempted to provide low-ability groups with additional training or instruction in order to reach parity with high-ability groups. These attempts have been mostly unsuccessful (page 86):

Additional evidence of the causal importance of g is provided by the many unsuccessful efforts to eliminate or short-circuit its functional link (correlation) with job proficiency. For example, there have been efforts to train the general cognitive skills that g naturally provides and that jobs require-such as general reading comprehension (which is important for using work manuals, interpreting instructions, and the like). Another approach has been to provide extra instruction or experience to very low-aptitude individuals so that they have more time to master job content. Both reflect what might be termed the training hypothesis, which is that, with sufficient instruction, low-aptitude individuals can be trained to perform as well as high-aptitude individuals. The armed services have devoted much research to such efforts, partly because they periodically have had to induct large numbers of very low-aptitude recruits. Even the most optimistic observers (Sticht, 1975; Sticht, Armstrong, Hickey, & Caylor, 1987) have concluded that such training fails to improve general skills and, at most, increases the number of low-aptitude men who perform at minimally acceptable levels, mostly in lower level jobs.

Gottfredson further states that differences in performance between high-ability and low-ability workers persists even as they acquire substantial experience:

Not even lengthy experience (5 years) eliminates differences in overall job performance between more and less bright men (Schmidt et al., 1988). A large study of military cooks, repairmen, supply specialists, and armor crewmen showed that performance may converge on simpler and oft-performed tasks (Vineberg & Taylor, 1972, p. 55-57). However, even that limited convergence took considerable time, reflecting large differences in trainability. It took men in the 10th to 30th percentiles of ability about 12 to 24 months to catch up with the performance levels on those tasks that were exhibited by men above the 30th percentile with no more than 3 months’ experience on the job. These findings from field settings are consistent with Ackerman’s (1987) review of the experimental literature relating skill learning and ability: individual differences in performance do not decrease with practice, and sometimes increase, when tasks are characterized by “predominantly inconsistent or varied information processing requirements .” In short, tasks that are not easily routinized continue to call forth g.

Socioeconomic success

A meta-analysis by Strenze (2007) [archived] showed that the predictive power of IQ is slightly stronger than that of parental SES (Table 1). Specifically, IQ measured before age 19 outdoes parental SES in predicting future educational attainment, occupational status, and income after age 29 (see “best studies” on Table 1). In other words, if you want to predict an adolescent’s success in adulthood along a given metric of success (e.g., income, educational attainment, or occupational status), it is more useful to know that adolescent’s IQ than to know the success of their parents along that same metric. In the conclusion of the analysis, Strenze (page 416) argues that this would be unexpected if the predictive power of IQ could be attributed primarily to its association with parental SES:

Despite the modest conclusion, these results are important because they falsify a claim often made by the critics of the “testing movement”: that the positive relationship between intelligence and success is just the effect of parental SES or academic performance influencing them both (see Bowles & Gintis, 1976; Fischer et al., 1996; McClelland, 1973). If the correlation between intelligence and success was a mere byproduct of the causal effect of parental SES or academic performance, then parental SES and academic performance should have outcompeted intelligence as predictors of success; but this was clearly not so. These results confirm that intelligence is an independent causal force among the determinants of success; in other words, the fact that intelligent people are successful is not completely explainable by the fact that intelligent people have wealthy parents and are doing better at school.

The meta-analysis does find that parental SES also correlates significantly with the future outcomes of the child. However, because youth IQ and parental SES are correlated, it is possible that some unspecified portion of the predictive power of youth IQ is due to its correlation with parental SES (or vice-versa). To get a more precise estimate of the effects of youth IQ (independent of parental SES), we need to estimate the predictive power of IQ after controlling for parental SES.

Murray and Herrnstein (1994) [archived] performed such an analysis with data from the 1979 National Longitudinal Survey of Youth (NLSY79). The NLSY79 was a longitudinal study that followed 12,686 who were aged 14 to 22 in 1979. Researchers recorded participants’ IQ scores at the beginning of the study and performed several follow-ups to track their performance along various life outcomes. Murray and Herrnstein used the NLSY79 to compare the predictive power of youth IQ and parental SES on a number of measures of success. “Parental SES” is measured based on “information about the education, occupations, and income of the parents of NLSY youths” (page 131). Murray and Herrnstein found that youth IQ outperformed parental SES in predicting adulthood poverty, educational attainment, likelihood of having illegitimate children, welfare usage, crime, and offspring IQ. The general finding reported was that individuals with low IQs and average parental SES were often worse-off than those with average IQs and low parental SES, and (inversely) individuals with high IQs and average parental SES were better-off than those with average IQs and high parental SES (see chapters 5-12). For example, whites with IQs one standard deviation below the mean (85 IQ) and average parental SES were more than twice as likely to never complete High School as whites with average IQs (100 IQ) and parental SES one standard deviation below the mean (~25% vs ~10%, see page 149). If the predictive power of IQ was solely due to its correlation with parental SES, then we would not expect IQ to predict outcomes better than parental SES.

The results of Murray and Herrnstein were partially corroborated by Rindermann and Ceci (2018) [doi]. These authors performed an analysis of the same dataset used by Murray – the NLSY79 – to compare the predictive power of childhood intelligence compared to other factors such as parental education and parental wealth. Their results showed that “children’s cognitive ability is more important than parental income for children’s later income as adults” in the United States (page 21).

Eid (2018) [archived] performed a similar analysis as Murray and Herrnstein using a newer data set: the 1997 National Longitudinal Survey of Youth (NLSY97). The NLSY97 includes information on 8,984 individuals about most of the same variables as the NLSY79. The primary differences is, as one might expect from the name, that NLSY97 studies those who were in their youth in 1997 rather than 1979. Eid focused the analysis on investigating the relative predictive power of IQ versus parental SES on adulthood poverty. The results of the study corroborate the findings from Murray and Herrnstein on the relative predictive power of IQ and parental SES, although the predictive power of both are smaller (page 3):

Without making any meaningful changes to HM’s methodology, we reaffirm the hypothesis that IQ is more important than family SES in avoiding poverty, though both of these covariates’ effects are smaller than those found by HM. Running a logistic regression with IQ, SES, and Age in 1997 as independent variables and poverty status in 2007 as the dependent variable, wefind the IQ effect to be approximately three times the size of the SES effect.

Another analysis comparing the effects of intelligence and socioeconomic background (SEB) on wages was performed by Ganzach (2011) [archived]. He investigated a sample of high school graduates from the 1979 National Longitudinal Survey of Youth (NLSY79). He used two measures of SEB: (a) a narrow index measured as a composite of parental education, family income, and parental occupational status, and (b) an extended index which included a number of other variables, including number of siblings, whether the participant lived in a two-parents home at age 14, a school composite based on the percent of economically disadvantaged students and non-white students, and a number of other variables (see page 124 for the full list). The results showed that “SEB affected wages solely by its effect on entry pay whereas intelligence affected wages primarily by its effect on mobility. The effect of intelligence on entry pay seems to be weaker than the effect of SEB” (page 127). In other words, both intelligence and SEB impacted entry pay, but only intelligence affected the pace of pay increases throughout one’s career.

Judge, Klinger, and Simon (2010) discovered similar findings while investigating the relationship between general mental ability (GMA) and career success over a 28-year period among participants in the National Longitudinal Survey of Youth (NLSY79). General mental ability was measured using the Armed Forces Qualifying Test (AFQT) in 1980. Researchers controlled for age, gender, race, and a SES composite at the onset of the study. Participants were placed into two groups: high-GMA participants (those scoring one standard deviation above the mean) and low-GMA participants (those scoring one standard deviation below the mean). Researchers found that outcome gaps between high-GMA and low-GMA widened dramatically over time. For example, the income gap between high-GMA and low-GMA participants grew about 25-fold from $1,575 in 1979 ($5,191 vs $3,616) to $38,819 in 2006 ($62,301 vs $23,482) (Figure 2a). The occupational prestige gap grew 6-fold from 7.67 points in 1979 (39.54 vs 31.87) to 49.79 points in 2006 (82.47 vs 32.68) (Figure 2b). Similar trends were found regarding human capital accumulation over time: the gap in education, training, and job complexity between high-GMA and low-GMA participants widened significantly over time (Figure 3). Finally, improvements in education, training, and job complexity were more likely to translate into larger improvements in income and occupational prestige for high-GMA participants (Figures 4 and 5).

Fergusson et al. (2005) [doi] examined a birth cohort of 1,265 children born in the Christchurch (New Zealand) urban region in mid-1977. Prior to controlling for other variables, researchers found that IQ measured at ages 8-9 was significantly related to outcomes such as crime, education, occupation, etc. at ages 15-25. For example, compared to individuals with childhood IQs below 85, individuals with childhood IQs above 115 were much more likely to gained school qualifications (98% vs 41%, Table 4), were much more likely to gain a university degree by age 25 (59% vs 2.1%, Table 4), and had higher mean incomes (37,433 vs 23,686). 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, indicating that “intelligence had a direct relationship to later educational, occupational and related outcomes independently of other childhood characteristics and family environment” (page 856).

Another source of evidence for the SES-independent predictive validity of IQ can be found by analyzing the outcomes of siblings raised in the same family. If IQ has predictive validity independently of family SES, then we would expect higher-IQ siblings to achieve more success than their lower-IQ siblings. Sternberg et al. (2001) [archived] reports that this is exactly what we find. When comparing brothers with different IQs from the same family, one finds that the higher-IQ brother achieves higher socioeconomic success as a result of pursuing higher education (page 9):

Jencks (1979) observed that if two brothers who grew up in the same family were compared on their SES as adults, the brother who had the higher IQ in adolescence would tend to have the higher adult social status and income. This path, however, is mediated by amount of education. The higher-IQ brother would be more likely to get more education and, correspondingly, to have a better chance of succeeding socioeconomically.

Criminal offending

Frisell, Pawitan, and Långström (2012) [archived] examined the association between cognitive ability at age 18 and violent offending among all men in Sweden born between 1961–1975. In total, the researchers analyzed 238,390 full-brothers, 17,594 half-brothers raised together, and 25,148 half-brothers raised apart. Data on adolescent cognitive ability was gathered using the Conscript Register (until 2007, conscription at age 18–20 was mandatory for all Swedish men). Information on socioeconomic characteristics (income, single mother, and urbanicity) was retrieved from the 1970 and 1975 national censuses. Finally, data on criminal convictions was gathered using the The Crime Register, which “covers all convictions in lower court from 1973 and onwards”. Violent offenses included “homicide, assault, robbery, threats and violence against an officer, gross violation of a person’s/woman’s integrity, unlawful coercion, unlawful threat, kidnapping, illegal confinement, arson, and intimidation.” Unsurprisingly, researchers the found a strong association between low cognitive and violent offending. They note that “men convicted of violent crime had more than a standard deviation lower cognitive ability than those without such convictions.” The authors did find that socioeconomic factors partially mediates the association between cognitive ability and violent offending. However, they note that “most of the association remains even after adjusting for such factors.” The authors conclude that “most of the association is not due to confounding by childhood environment”.

For other studies demonstrating the predictive power of cognitive ability on criminal offending after controlling for parental SES, see Moffitt et al. (1981) [archived], Lynam et al. (1993) [doi], and Levine (2011) [archived].

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, low 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.

Some comments on controls

At the beginning of this section, I stated that, in order to demonstrate a causal relationship from X to Y, one should control for a possible confounding variable, Z, that may be causing both X and Y. One might be tempted to control for confounding by controlling for any third variable that is associated with both X and Y. The problem with this approach is that it may introduce bias. Consider two examples:

  • If Z is a mediator variable, then controlling for Z may result in an underestimate of the causal effect of X on Y. In other words, if the causal path from X to Y involves Z as an intermediate variable, then controlling for Z is not appropriate because it will block the very causal effect that we intend to estimate.
  • If Z is a collider variable, then controlling for Z may result in an overestimate of the causal effect of X on Y. In other words, if Z is an effect of both X and Y, then this can produce an association between X and Y even if X has no effect on Y. For a brief overview of collider bias with examples, see here [archived].

For more examples of good and bad controls during causal inference, see Cinelli et al. (2020) [archived]. For a primer in causal inference in statistics, see this short book [archived] by Judea Pearl.

I mention this because it is possible that some of the earlier studies may have masked the effects of cognitive ability due to controls for possible mediator variables. For example, some of the studies reviewed by Ttofihi et al. (2016) estimated the association between intelligence and offending after controlling for various “risk factors” for offending (by stratifying subjects into “high risk” and “low risk” groups). But some of these factors (e.g., antisocial behaviors, low concentration, etc.) may actually be mediator variables in the causal path from intelligence to criminal offending, i.e. a child’s intelligence may influence their likelihood of engaging in early antisocial behavior, which may influence their likelihood of later criminal offending. The fact that intelligence had a large effect on offending even after controlling for these risk factors shows the impressive effect of intelligence on offending; but it should be noted that the magnitude of this effect may actually be underestimated if we make the plausible assumption that the effect of intelligence on criminal offending is mediated through some of the controlled risk factors.

This point is particularly relevant for a study cited earlier: Fergusson et al. (2005) [doi]. In this study, researchers found that, prior to controlling for covariates, IQ measured at ages 8-9 was significantly related to adulthood personal adjustment problems (“personal adjustment” included measures of criminal activity, substance use, mental health, and sexual behavior). For example, compared to participants with IQs above 115, participants with IQs below 85 were significantly more likely to become arrested/convicted for a crime during ages 21-25 (15.8% vs 6.6%, table 1) and become or get someone pregnant by ages 15-18 (13.8% vs 2.3%, table 1). After controlling for a number of covariates (e.g., childhood conduct problems, attentional problems, and socioeconomic disadvantage), the association between childhood IQ and these outcomes was no longer statistically significant. However, IQ may may causally influence the covariates that the researchers controlled for, so the reduction in the association between childhood IQ and adulthood success may not imply a comparable reduction in the causal impact of IQ. The researchers express this point in their discussion (page 856):

[T]hese results raise important questions about the processes which link early conduct problems and IQ. Three explanations seem possible…If the association is explained by common genetic, social, family and related factors then the association between IQ and later adjustment is non-causal and reflects the consequences of common factors. If IQ, in some way, influences predisposition to conduct problems then the effects of IQ on later adjustment are causal and are mediated via early adjustment. Finally, if the association between IQ and conduct problems arises because conduct problems lead in some way to a lower measured IQ, the association between IQ and later adjustment is non-causal and reflects the common influence of early adjustment on both later adjustment and measured IQ.

As the studies above make clear, an individual’s cognitive ability has a large influence on their life outcomes even after controlling for parental socioeconomic success. However, many studies also found that the association between cognitive ability and life outcomes is somewhat reduced after controlling for parental SES, indicating that some of the association between cognitive ability and life outcomes may actually be the result of the influence of parental SES. I would like to note that this finding also indicates the importance of cognitive ability. Parental SES is also substantially influenced by parental cognitive ability. Therefore, even the finding that parental SES has some influence on a child’s life outcomes may be a reflection of the importance of cognitive ability. Rather than showing the effect of offspring cognitive ability on offspring outcomes, this finding would show the effect of parental cognitive ability on offspring outcomes.

Conclusion


There is overwhelming evidence that cognitive ability involves a stable set of traits that play a strong role in determining a person’s future income, educational attainment, occupational success, criminality, etc. These findings have significant political implications. These findings suggest that any plan to improve success in these outcomes may need to focus on improving the cognitive ability of the subjects of concern. Furthermore, because of the lifetime stability of cognitive ability, this suggests that interventions may need to target improving the cognitive ability of children at a very young age (perhaps even before birth). Cognitive ability is a significant factor in success for nearly all societal outcomes that we care about. If we ignore this crucial factor, we are unable to develop a working understanding of the causes of success (and failure) of individuals and groups in Western societies, and we may be unable to develop informed solutions to address inequalities in certain outcomes when these inequalities are the results of differences in cognitive ability.

In closing, I would like to note that cognitive ability is not everything. There are plenty of other factors that influence an individual’s success. For example, some of the studies cited above show the importance of parental socioeconomic status. Additionally, there is a growing body of evidence that “non-cognitive” abilities have a tremendous impact on success. Many studies show that childhood self-regulation predicts adulthood success even after controlling for IQ and parental SES (Moffitt et al. 2011, pages 2694-2695; Fergusson et al. 2013, Table 3). Also, Duckworth et al. (2012) conducted two longitudinal studies of middle school students and found that self-control had a larger impact on report card grades than intelligence (although intelligence had a larger impact on standardized achievement test scores). Conscientiousness has also been shown to predict success independently of cognitive ability. For example, conscientiousness has been shown to predict occupational performance (Hurtz and Donovan 2000, table 2; Dudley et al. 2006, Table 4). Impressively, conscientiousness has sometimes been shown to predict academic achievement almost as well as intelligence (Poropat 2009, Table 2; Richardson  et al. 2012, Table 4). Furthermore, a number of meta-analyses have shown that an internal locus of control is significantly correlated with academic performance (Richardson et al. 2012) and job performance (Ng et al. 2006Judge and Bono 2001).

Relevant works


Surveys

Meta-analyses

Reviews

Books
  • 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.

Leave a Reply

Your email address will not be published.