What IQ correlates with: eighteen correlations ranked by size and source

0 The short answer

IQ correlates most strongly with other measures of ability, such as SAT scores (0.82) and the same test taken decades later (0.66), moderately with education and school grades (0.56 and 0.54), and only weakly with income, personality, brain volume, creativity and height (0.09 to 0.24 in size). The ranking below uses eighteen published coefficients, each from a meta-analysis or a large sample, and it marks whether the number is as observed, corrected for measurement error or range restriction, or estimated for a latent trait. A correlation describes how two things move together across a group; it does not say that IQ causes the other variable, and even 0.56 leaves most of the variation in an individual outcome unexplained.

1 What Does It Mean for IQ to Correlate With Something?

A correlation coefficient measures how closely two variables move together across a group, on a scale from minus 1 to plus 1, and it says nothing by itself about cause, about one person, or about how large the practical difference is. The coefficient is written r. A value of 0 means no linear association, a value of 1 means that every person who is higher on one variable is higher on the other in exact proportion, and a negative value means that higher scores on one tend to go with lower scores on the other. When this page says that IQ correlates 0.56 with education, it means that across the people in the studies pooled, those with more schooling tended to have higher scores, and the association was strong enough to be seen but far from exact.

Two conversions make the number easier to read, and both are our arithmetic, not a result reported by any study cited here. The first is the share of variance, which is r multiplied by itself. A correlation of 0.56 accounts for about 31 percent of the variation in the other variable, 0.24 for about 6 percent (Pietschnig and colleagues report an R squared of 0.06 for brain volume themselves), 0.16 for about 3 percent and 0.09 for under 1 percent. The second is the spread that remains around a prediction. If the relationship is linear and both variables are expressed in standard units, the typical distance between a person and the line is the square root of 1 minus r squared. For r = 0.56 that is 0.83, and for r = 0.24 it is 0.97. In other words, knowing a person's IQ shrinks the uncertainty about education by about a sixth when the correlation is 0.56, and by about 3 percent when it is 0.24.

An example makes the logic concrete. Under a linear model, a person at IQ 115, one standard deviation above the mean on the scale explained on the page about standard deviation 15, has an expected education score of 0.56 standard deviations above the mean when the correlation is 0.56. That is the average for everyone at 115, and the people at 115 still spread widely around it. The same person has an expected score of 0.09 standard deviations above the mean on a variable that correlates 0.09, which is no usable expectation at all. The ranking in this article is therefore a ranking of how much a group average moves, not of how reliably any individual outcome can be foreseen.

The definition of the thing being correlated also matters. IQ in this article means a score on a standardized test of general cognitive ability or on a battery that estimates it, described on the page about what IQ measures and organized by the CHC model into broad abilities such as fluid reasoning (Gf), comprehension-knowledge (Gc), quantitative reasoning (Gq), visual processing (Gv), working memory (Gwm) and processing speed (Gs). Some of the studies below use a single test, some a composite of several, and some a statistical factor extracted from a battery. The difference is flagged in the table where it changes how the number should be read.

2 The Ranked Table: Eighteen Correlations With IQ, Strongest to Weakest

Ranked by absolute size, the strongest correlates of IQ are other ability measures and the same test repeated over a lifetime, the next tier is schooling and work, and the long tail of personality, attitudes, brain size and body size sits between 0.09 and 0.24. The table gives each coefficient as published, with the number of studies or people the authors report, and a final column that says whether the figure is as observed, corrected, or latent. Signs are written out: a minus sign means higher IQ goes with a lower value of the other variable, and it does not mean that the association is bad or unimportant. Where two coefficients tie in absolute size, the order between them carries no meaning.

A few reading rules prevent most misuse of this table. Rows 1, 2 and 5 are correlations between ability measures, so they show consistency in what is being measured more than an outcome in life. Rows 3, 4, 7, 9 and 12 are the outcomes people usually have in mind, and the page on whether IQ predicts success covers education, work, income and health in depth, so this article keeps to what the numbers add to a ranking. Rows 11, 14 and 15 are about beliefs and attitudes, with the caveat that they are group averages with direction left to the studies. Rows 16 and 17 are the links people most often repeat and most often overstate. Each row is taken up below, in the order of the kind of evidence, with the design that stands behind it.

The table is also the answer to a design question, so it is worth saying what is not in it. It leaves out associations for which a study reports a difference, a hazard ratio or a trend rather than a correlation, because those cannot be placed on the same scale without conversions that would add precision the studies do not have. A later section lists those topics and sends the reader to the pages that cover them. It also leaves out any claim whose coefficient we could not open in the source and read, which is why a few widely repeated numbers do not appear.

3 Observed, Corrected, Latent: Why the Same Link Carries Several Numbers

A published correlation with IQ can differ by 0.10 or more depending on whether it is the raw average, a sample weighted average, or a value corrected for unreliability and range restriction, so two rows in the table are only comparable when the kind of coefficient is the same. Three kinds appear in the ranking. An observed coefficient is what the sample produced. A corrected coefficient estimates what the sample would have produced if the tests were perfectly reliable or the group had not been filtered on ability. A latent coefficient is between factors extracted from several tests, such as a general ability factor, so that the quirks of any single test drop out.

Strenze's meta-analysis of longitudinal studies is the cleanest illustration, because its Table 1 prints all three columns for the same studies. For education, the plain average correlation across 59 samples was 0.46, the sample size weighted average was 0.48, and the corrected value was 0.56. The corrected value in that paper is the sample size weighted average corrected for unreliability and dichotomization, using an average test reliability of 0.83 for most of the IQ measures, and the author states that no mathematical correction for range restriction was performed. He excluded studies with considerable range restriction instead. He also notes that larger studies produced smaller correlations, which suggests that smaller and less representative samples overstate the link.

Source: Table 1 of Strenze 2007, the first row of each outcome (all studies). The "best studies" are samples tested before age 19 on average and measured for success after age 29 on average.

The same pattern of choices appears in the other rows. Roth and colleagues pooled 240 samples with 105,185 participants and report a population correlation of 0.54 between standardized intelligence tests and school grades after correcting for sampling error, measurement error and range restriction in the IQ measure. Ackerman, Beier and Boyle estimated the link between working memory and general intelligence at 0.479 using true-score estimates, which remove test unreliability. Deary and colleagues report a raw age 11 to age 80 correlation of 0.66 and a value of 0.73 once the narrowed range of the cohort is taken into account. In each of these cases the authors chose the coefficient that best answers the question of how strong the link is between the underlying traits, and where the raw value is also printed it is smaller.

The correction itself is a judgment, and it can be wrong in a way that inflates a number. Sackett and colleagues, writing in the Journal of Applied Psychology in 2022, argued that the range restriction corrections used in earlier selection research often overcorrected, and that the validity of many selection procedures for predicting job performance had been overestimated. Their revised operational validity for cognitive ability tests is 0.31, against the 0.51 offered by Schmidt and Hunter in 1998, and they report that the highest alternative value they examined, pairing alternate reliability estimates with an indirect range restriction correction, is 0.36. The paper does not conclude that cognitive ability stopped mattering. It concludes that the earlier figure was too high, and that most of the same procedures keep their rank with mean validities lower by 0.10 to 0.20.

The practical consequence is that the ranking is robust at the level of tiers and fragile at the level of positions. Whether the correlation with job performance is 0.31 or 0.51 depends on a statistical choice that specialists still debate, and under the older figure that row would sit just below school grades instead of ninth. A reader who wants to cite one number should cite the row and its label together. A reader who wants to compare should compare rows of the same kind. The page on reliability and validity explains the two ideas that every correction here rests on, and the page on how IQ scores are normed explains why a score is always relative to a reference group.

4 Why Are the Strongest Correlates of IQ Other Ability Measures?

The three largest coefficients in the table, 0.82, 0.66 and 0.479, are all links between measures of cognitive ability, which shows that these tests largely capture the same thing rather than that IQ explains an outside outcome. This is the most reliable evidence behind the idea of general cognitive ability, and it is also the part of the ranking that is least informative about life.

The 0.82 comes from Frey and Detterman's study of the SAT. In the first of their two studies they used 917 participants from the National Longitudinal Survey of Youth 1979, extracted a general ability factor from the Armed Services Vocational Aptitude Battery, and correlated it with SAT scores. The result was 0.82, or 0.86 after correction for nonlinearity. Their second study tells a more cautious story. Among 104 undergraduates, revised and recentered SAT scores correlated 0.483 with Raven's Advanced Progressive Matrices, or 0.72 once corrected for restricted range. Two things differ between the studies at once: the first sample spans the full range of ability while the second has already passed through university admission, and the first uses a factor from a battery while the second uses a single matrices test. The authors concluded that the SAT is mainly a test of general ability. The page on IQ and the SAT covers score conversion and what the correlations do and do not allow.

The 0.479 for working memory deserves the opposite reading. Several investigators had claimed over the previous decade that working memory and general intelligence are identical or nearly identical. Ackerman, Beier and Boyle meta-analyzed 86 samples and found that the average correlation between true-score estimates of the two is substantially less than 1. By our arithmetic, 0.479 squared is about 0.23, so roughly a quarter of the variation in working memory (Gwm) is shared with the general factor. That makes working memory one of the strongest correlates of IQ and still a distinct ability, which is the reason that a battery that reports a working memory index separately can show a profile that a single number hides.

The stability row is the one that surprises people in both directions. In the Lothian Birth Cohort 1921, a group of people born in 1921 who had taken the Moray House Test at age 11 in the Scottish Mental Survey of 1932, the same test taken at about age 80 correlated 0.66 in the whole sample, with 0.62 for men and 0.69 for women. An earlier follow-up of a smaller sample from the same survey had found 0.63, and the authors report disattenuated values of 0.73 for all participants after allowing for the restricted range of the cohort. A correlation of 0.66 across about 69 years is large by the standards of anything in this table. It is also, by our arithmetic, about 0.44 of the variance, so more than half of the differences in the age 80 scores were not predicted by the age 11 scores. People moved up and down relative to others, and the data show how much, not whether.

5 Education, Work and Income: The Outcome Rows

Education, school grades, occupation, job performance and income form the outcome tier of the ranking, and the order within it, 0.56, 0.54, 0.43, 0.31 and 0.20, is the order in which a link to IQ is easiest to see. The ordering is an observation, not a theorem: it reflects what the studies measured, and the page on whether IQ predicts success is the place to read the full argument for education, work, income and health. This section adds only what the ranking itself shows.

Strenze states that for every predictor in his analysis, the correlation with education is the strongest and the correlation with income the weakest. The comparison with other predictors is the more useful finding. His abstract concludes that intelligence is a powerful predictor of success but, on the whole, not an overwhelmingly better predictor than parental socioeconomic status (SES) or school grades. The corrected values show how close the race is.

Source: corrected sample size weighted correlations from Table 1 of Strenze 2007. Sample counts differ by row.

Two points follow. First, IQ is a leading predictor but not a lone one: an index of the family's education, occupation and income correlates with a person's own education almost as strongly (0.55 against 0.56). Second, school grades predict education about as well as IQ and occupation nearly as well, but predict income far worse (0.09 against 0.20). Roth and colleagues find that the strength of the link between IQ and grades varies by school subject domain, by grade level, by the type of intelligence test and by publication year, and that gender had no effect on its size. A single figure for grades therefore hides moderation, and the page on IQ and academic achievement goes through it. The page on average IQ by education level shows what group means look like when the same relationship is read as a table.

On work, the coefficient depends on the question. The 0.43 for occupational level in Strenze is a link between childhood or youth scores and the status of the job held later. The 0.31 for job performance is Sackett and colleagues' revised operational validity for cognitive ability tests, a link between a test score and a measure of how well someone performs the job, corrected for unreliability in that measure. They are different questions with different ceilings, which is why the page on IQ and job performance separates them. The 0.20 for income is the figure that most often surprises readers, and the page on IQ and income covers why a correlation of that size is compatible with large average differences between groups and with a great deal of overlap between them.

6 Do Partners Have Similar IQs? What the 0.44 Does and Does Not Measure

Partners resemble each other in measured IQ at about 0.44 in a meta-analysis of 17 samples, which makes the partner row a measure of how alike two people are rather than a measure of what IQ does to a person. Horwitz, Balbona, Paulich and Keller combined systematic reviews and meta-analyses of 22 traits and then analyzed 133 traits in UK Biobank. For the IQ score, the random-effects mean was 0.44 across 17 samples (13 published studies) and 5,672 partner pairs, with a 95 percent interval of 0.23 to 0.61. That interval is wide, so the pooled value is imprecise.

The meta-analysis is most useful for what it puts next to the IQ figure. In the same table, partners resembled each other at 0.55 for educational attainment (86 samples and about 1.9 million pairs), 0.58 for political values, 0.56 for religiosity, 0.24 for height, 0.16 for body mass index, 0.21 for openness and between 0.08 and 0.16 for the other Big Five traits. The abstract summarizes the pattern: political and religious attitudes, educational attainment and some substance use traits showed the highest correlations. Resemblance in IQ is therefore high compared with most personality traits and physical traits, and lower than resemblance in education and attitudes. It is not an unusual feature of intelligence.

Why partners resemble each other is a separate question, and the correlation cannot settle it. People may choose partners who are similar, may meet them in settings that already sort by schooling and interests, and may become more alike over time. Each path produces the same coefficient in a cross-section. The page on IQ and relationships covers the partner literature in more detail. For this ranking the point is narrower: a correlation of 0.44 between two people's scores is one of the larger entries in the table, and nothing in it says that a higher IQ makes a person more or less likely to find a partner.

7 Brain Volume and IQ: A Correlation of 0.24 With a Publication Bias Story

Total brain volume correlates with IQ at 0.24, which is about 6 percent of the variance, and the authors of a meta-analysis of 88 studies argue that the published literature overstates even that. Pietschnig, Penke, Wicherts, Zeiler and Voracek combined 88 studies covering 148 healthy and clinical mixed-sex samples with more than 8,000 individuals, using published results and unpublished results obtained by personal communication. They found a significant positive association (r = 0.24, R squared = 0.06) that generalized over age, over the type of IQ (full-scale, performance and verbal) and over sex.

The more careful part of the paper is about bias. Their publication bias analyses indicate that strong positive correlations had been reported more often than small or non-significant ones, which would inflate the average of the published literature, and that reported effects have been declining over time. They conclude that the strength of the association has been overestimated, but that it remains robust when different kinds of dissemination bias are accounted for. They also caution against reading brain size as a direct proxy for intelligence differences, noting that it is tempting to interpret the association in terms of cognitive evolution and species differences in brain size but that this is not warranted for individual differences in humans.

Two readings of the 0.24 should be kept apart. At the level of a population, a correlation of 0.24 means that people with larger brains have somewhat higher scores on average. At the level of a person, it means that a brain scan tells very little about a score: by our arithmetic, 94 percent of the variation in IQ is unrelated to brain volume, and the typical spread around the expected score is about 97 percent of the full spread. The page on intelligence and the brain goes through what imaging studies have and have not replicated, including network models and large samples, and this ranking keeps only the meta-analytic average with its bias warning.

8 Personality, Religious Belief and Political Attitudes

Personality traits, religious belief and political attitudes correlate with IQ at sizes from 0.09 to 0.24, which are real in large samples and too small to say anything about a given person. These are the rows where the group averages are most easily mistaken for personal traits, so each row is stated with its design and its limit.

For personality, Anglim, Dunlop, Wee, Horwood, Wood and Marty meta-analyzed 162,636 people (k = 272) and report that openness (ρ = 0.20) and neuroticism (ρ = minus 0.09) were the strongest Big Five correlates of intelligence. Openness correlated more with crystallized than with fluid intelligence, so the link is closer to knowledge and engagement with ideas than to raw reasoning. At the level of facets, traits related to intellectual engagement and unconventionality were more strongly related to intelligence than other openness facets, while sociability and orderliness were negatively correlated with intelligence. That means that the other three Big Five domains were weaker correlates than these two, and that the facet pattern is more informative than the domain label. The page on IQ and personality goes through it.

For religion, Zuckerman, Silberman and Hall meta-analyzed 63 studies and found a significant negative association between intelligence and religiosity. The association was stronger for college students and the general population than for participants younger than college age, and stronger for religious beliefs than for religious behavior. For college students and the general population, the means of weighted and unweighted correlations between intelligence and the strength of religious beliefs ranged from minus 0.20 to minus 0.25, with a mean of minus 0.24. The authors discuss three possible interpretations, which they present as possibilities: that intelligent people are less likely to conform, that they tend toward an analytic rather than an intuitive thinking style, and that they may have less need for some functions that religiosity serves. The meta-analysis does not test which, if any, is correct. The page on IQ and religion covers the designs.

For political attitudes, Onraet, Van Hiel, Dhont and Hodson meta-analyzed 67 studies with 84,017 participants and found an average effect size of minus 0.20 (95 percent interval minus 0.23 to minus 0.17) between cognitive ability and right-wing ideological attitudes, and minus 0.19 (interval minus 0.23 to minus 0.16, 23 studies, 27,011 participants) between cognitive ability and prejudice. The effect sizes did not vary significantly across different cognitive abilities and sample characteristics, but they depended strongly on the measure of attitudes used, with the strongest effects for authoritarianism and ethnocentrism. The page on IQ and political orientation places this in a wider debate, and the page on the IQ of politicians takes up what records of elected officials do and do not show.

A coefficient of minus 0.20 describes a faint tilt, not a divide. By our arithmetic it accounts for 4 percent of the variance, so that knowing a person's IQ tells almost nothing about whether the person is religious, conservative or prejudiced, and the intelligent people who hold any view are far more numerous than a headline average suggests. These rows are in the table because people ask about them, and they are placed where the numbers put them.

9 Self-Estimates, Creativity and Height: Three Popular Links That Are Weaker Than Expected

People's own estimates of their intelligence, scores on creativity tests and body height each correlate with IQ at 0.33 or less, and the most repeated of the three, height, correlates at 0.16. The three share a feature: each is a plausible link that is easy to believe from experience, and each turns out to be modest when measured.

On self-estimates, Freund and Kasten meta-analyzed 154 effect sizes from 41 published studies and estimated the overall relationship between self-estimated and measured cognitive ability at 0.33. By our arithmetic that is about 11 percent shared variance. The true effects varied both within and between studies. Self-estimates were more valid when the scale was relative, with a clearly specified comparison group, and when numerical ability was assessed rather than general cognitive ability. The assessment of less frequently considered dimensions, such as reasoning speed, significantly lowered the relationship. The practical lesson is that a person's impression of where they stand is a modest guide to a measured score, which is one reason a score and its confidence interval are reported together.

On creativity, Kim combined 447 correlation coefficients from 21 studies and 45,880 participants and found a mean correlation of 0.174 (95 percent interval 0.165 to 0.183) between creativity test scores and IQ scores. The result was small but heterogeneous, and the author characterizes the relationship as negligible. Age contributed most to the variation in the relationship, and different creativity tests contributed second. The study did not support the threshold theory, the idea that IQ matters for creativity only up to about 120. A creativity test score is also not the same as creative achievement, and the page on IQ and creativity covers the difference.

On height, the Generation Scotland Study analyzed 6,815 unrelated adults with a median age of 57 and found a phenotypic correlation between height and general intelligence (g) of 0.16 with a standard error of 0.01. Using single-nucleotide polymorphism data, the authors estimated a genetic correlation of 0.28 (standard error 0.09) and a bivariate heritability of 0.71, and concluded that the majority of the phenotypic correlation is explained by shared genetic influences. By our arithmetic, 0.16 means that about 3 percent of the variation in scores tracks height. The association exists and it is small enough that it cannot be used to guess anyone's score. The page on IQ and height covers nutrition, socioeconomic status and the twin evidence.

10 Which Correlates Are Causes? Direction, Confounding and What the Table Leaves Out

Every row in the table is a correlation from observational data, so none of them states which variable acts on which, and the studies themselves show four reasons to be careful about reading direction, size and stability. The first reason is direction. Schooling is the clearest case: a longer education could raise measured ability, higher ability could lead people to stay in school longer, and both could be pushed by family circumstances. The page on whether you can improve your IQ covers the quasi-experimental studies built to separate those paths, and the correlation of 0.56 cannot do so on its own.

The second reason is confounding. In Strenze's analysis, an index of parental SES correlates 0.55 with a person's education, almost the same as IQ at 0.56. If both IQ and family circumstances are tied to the outcome, and to each other, then the coefficient for either one carries some of the other. The third is selection and publication. Strenze reports that larger studies produced smaller correlations, and Pietschnig and colleagues report that strong correlations for brain volume were reported more often than weak ones and that reported effects have been declining over time. Both findings mean that a number taken from a single small study is likely to be larger than the number from the whole literature.

The fourth reason is that a mean coefficient hides a range. Strenze's 95 percent credibility interval for the all-studies education correlation runs from 0.36 to 0.75, and for income it runs from minus 0.01 to 0.40. That means the true link varies a great deal across samples and settings, and that for income the data are compatible with a link that is close to zero in some of them. Horwitz and colleagues' interval for the partner correlation, 0.23 to 0.61, shows the same thing from another direction. Freund and Kasten and Kim both found significant heterogeneity in their pooled results. A table of single numbers is a summary of distributions, and the page on reliability and validity explains why that matters for any measurement.

The table leaves out associations for which the studies report something other than a correlation, because converting them would add precision that the studies do not contain. Birth order, video games, sleep, exercise, humor, musical ability, chess, crime, mental health and longevity each have their own page with the design that suits them, including the pages on birth order and IQ, video games and IQ, chess and IQ and longevity and IQ. Where those pages report a correlation, it comes from a different meta-analysis with its own correction choices, and it should be read with its own label rather than slotted into this ranking. Four neighboring pages published alongside this one cover other questions: what can lower your IQ, left-handedness and IQ, the IQ of politicians and conspiracy beliefs and IQ.

11 What Do These Correlations Say About You?

A group correlation, even 0.56, cannot tell you what your own score means for your own education, income or health, because it describes how averages move across thousands of people and leaves most of the variation in any one person unexplained. Take the largest outcome coefficient in the table, 0.56 for education. By our arithmetic, it leaves 69 percent of the variance in education unexplained by IQ, and the typical person lies 0.83 standard deviations from the line, which is most of the width of the whole distribution. Among people at the same IQ, there are some with little schooling and some with a great deal, for reasons that include family, health, money, geography, interests and chance. The smaller coefficients are weaker still.

The reverse also holds. A coefficient of minus 0.09 does not mean that anxious people are less capable, and 0.16 does not mean that tall people reason better. These rows describe a faint tilt in large samples, and a faint tilt means that a person's height, personality or beliefs are poor evidence about the person's score. The same applies to anyone who tries to infer a score from a job title, a degree, a religion, a party or a hobby. A score has meaning only against a reference group, and the page on how rare a given IQ is shows what a number implies about position in the population.

Nothing in this article is a clinical or diagnostic statement about the reader. ACIS is an online, unsupervised, self-administered assessment. It reports a Full Scale IQ, six index scores in the CHC domains, percentiles and a 95 percent confidence interval, and its page on reliability and validity and its technical manual set out the documentation. ACIS is not a clinical or diagnostic instrument and is not for hiring, school accommodations or admission to high IQ societies, and it is available in English only. Its forms, read on October 13, 2026, are Quick (15 dollars, 6 subtests in 3 domains, about 45 minutes), Optimized (30 dollars, 13 subtests in 5 domains, about 110 minutes) and Full Scale (50 dollars, all 20 subtests in six domains, about 175 minutes), each a one time payment, with breaks allowed and a free trial that needs no card. Prices can change.

12 What the Evidence Supports, Stated Narrowly

The evidence supports a short list of statements: IQ correlates most strongly with other ability measures and with itself over a lifetime, moderately with education and school grades, less with work and income, and only weakly with brain volume, personality, attitudes, creativity and height, and it supports nothing about causes or about one person. Each clause rests on a named design in the table, and each is limited by the label in its row.

Six statements follow from the sources opened for this page. First, scores on tests of general ability agree with each other strongly: the SAT with a general ability factor at 0.82 in a national sample of 917, and the same mental test at age 11 and age about 80 at 0.66. Second, working memory shares less than a quarter of its variation with the general factor (0.479 as a true-score estimate), so related abilities remain distinguishable. Third, IQ correlates 0.56 with education and 0.54 with school grades after correction, and an index of parental SES correlates 0.55 with education, so IQ is a leading predictor and not a lone one. Fourth, the link weakens as the outcome depends more on circumstances: 0.43 for occupational level, 0.31 for job performance on the revised estimate, and 0.20 for income. Fifth, the biological, personality and attitude rows are small: 0.24 for brain volume, minus 0.24 for the strength of religious belief, 0.20 for openness and minus 0.20 for right-wing attitudes, minus 0.19 for prejudice, 0.174 for creativity test scores, 0.16 for height and minus 0.09 for neuroticism. Sixth, the coefficient depends on whether it is observed, corrected or latent, and on the study that produced it.

What the evidence does not support is equally specific. It does not support the claim that IQ causes any of these outcomes, because the designs are observational. It does not support using any coefficient to predict an individual outcome, because even the largest leaves most of the variation unexplained. It does not support ranking the rows as if they were measured the same way, because a corrected meta-analytic value and a single observed sample answer different questions. It does not support reading a group difference in an attitude or belief as a statement about any person. The sources vary in age, from 2004 to 2023, and the correlations may differ in other populations, so a reader should treat the figures as estimates with the uncertainty the authors report.

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014), the joint publication of the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education, as listed on the Standards' own site which we opened on October 13, 2026, frames the question that a ranking like this one cannot answer. Whether a score supports a particular decision about a particular person is a matter of evidence for that use, together with the score's reliability and error and the group it is compared against. A correlation between IQ and an outcome in a population is one piece of that evidence, and it does not replace the others. Anyone who reads a score, their own or another person's, should read it with its confidence interval, its conditions and its purpose, and should not use it in place of a professional assessment where a decision with real consequences depends on it.

13 Sources Behind This Page

Every journal article below was checked against its DOI record in Crossref on October 13, 2026, and each coefficient used on this page was read in the article's text or abstract on that date.

• Ackerman, P. L., Beier, M. E., and Boyle, M. O. Working memory and intelligence: The same or different constructs? Psychological Bulletin, 2005, volume 131, issue 1, pages 30 to 60.
• Anglim, J., Dunlop, P., Wee, S., Horwood, S., Wood, J., and Marty, A. Personality and intelligence: A meta-analysis. Psychological Bulletin, 2022, volume 148, issues 5 and 6, pages 301 to 336.
• Deary, I. J., Whiteman, M. C., Starr, J. M., Whalley, L. J., and Fox, H. C. The impact of childhood intelligence on later life: Following up the Scottish Mental Surveys of 1932 and 1947. Journal of Personality and Social Psychology, 2004, volume 86, issue 1, pages 130 to 147.
• Freund, P. A., and Kasten, N. How smart do you think you are? A meta-analysis on the validity of self-estimates of cognitive ability. Psychological Bulletin, 2012, volume 138, issue 2, pages 296 to 321.
• Frey, M. C., and Detterman, D. K. Scholastic Assessment or g? The relationship between the Scholastic Assessment Test and general cognitive ability. Psychological Science, 2004, volume 15, issue 6, pages 373 to 378.
• Horwitz, T. B., Balbona, J. V., Paulich, K. N., and Keller, M. C. Evidence of correlations between human partners based on systematic reviews and meta-analyses of 22 traits and UK Biobank analysis of 133 traits. Nature Human Behaviour, 2023, volume 7, issue 9, pages 1568 to 1583.
• Kim, K. H. Can only intelligent people be creative? A meta-analysis. Journal of Secondary Gifted Education, 2005, volume 16, issues 2 and 3, pages 57 to 66.
• Marioni, R. E., Batty, G. D., Hayward, C., and colleagues (Generation Scotland). Common genetic variants explain the majority of the correlation between height and intelligence: The Generation Scotland Study. Behavior Genetics, 2014, volume 44, issue 2, pages 91 to 96.
• Onraet, E., Van Hiel, A., Dhont, K., Hodson, G., Schittekatte, M., and De Pauw, S. The association of cognitive ability with right-wing ideological attitudes and prejudice: A meta-analytic review. European Journal of Personality, 2015, volume 29, issue 6, pages 599 to 621.
• Pietschnig, J., Penke, L., Wicherts, J. M., Zeiler, M., and Voracek, M. Meta-analysis of associations between human brain volume and intelligence differences: How strong are they and what do they mean? Neuroscience and Biobehavioral Reviews, 2015, volume 57, pages 411 to 432.
• Roth, B., Becker, N., Romeyke, S., Schäfer, S., Domnick, F., and Spinath, F. Intelligence and school grades: A meta-analysis. Intelligence, 2015, volume 53, pages 118 to 137.
• Sackett, P. R., Zhang, C., Berry, C. M., and Lievens, F. Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range. Journal of Applied Psychology, 2022, volume 107, issue 11, pages 2040 to 2068.
• Strenze, T. Intelligence and socioeconomic success: A meta-analytic review of longitudinal research. Intelligence, 2007, volume 35, issue 5, pages 401 to 426.
• Zuckerman, M., Silberman, J., and Hall, J. A. The relation between intelligence and religiosity: A meta-analysis and some proposed explanations. Personality and Social Psychology Review, 2013, volume 17, issue 4, pages 325 to 354.
• American Educational Research Association, American Psychological Association, and National Council on Measurement in Education. Standards for Educational and Psychological Testing. 2014 edition, page opened October 13, 2026.

14 Frequently Asked Questions