IQ and Height: Small Correlation

0 The Short Answer

Height and measured intelligence do correlate positively, and the correlation is small. The cleanest single-cohort estimate comes from Marioni and colleagues, writing in Behavior Genetics in 2014, who measured standing height and general cognitive ability in 6,815 unrelated Scottish adults from the Generation Scotland cohort. The phenotypic correlation they report is 0.16, with a standard error of 0.01. Most well-powered studies land somewhere between 0.10 and 0.20, which means the finding is real, it replicates, and it is nearly worthless for saying anything about a particular human being.

What makes the topic worth a page is not the number. It is the question of where the number comes from. A correlation of that size can be produced by at least five different mechanisms operating at once, and researchers have spent forty years trying to apportion it: childhood nutrition and disease burden acting on both traits, family socioeconomic position acting on both traits, prenatal growth acting on both traits, genes that happen to influence both traits, and a subtler cause that has nothing to do with biology at all, which is the way people choose partners. Each of those leaves a different signature in the data, and the designs that separate them are among the more elegant things behavioral genetics has built.

The one conclusion the evidence does not support is the naive causal one. Nothing in the literature says that being tall makes a person think faster, and nothing says that thinking faster makes a person tall. The arrow, where anyone has managed to isolate one, points from shared upstream causes into both outcomes, and it points weakly.

1 What the Data Actually Show

The association turns up in almost every large dataset that happens to record both variables, which is a lot of them, because conscription boards have measured height and administered cognitive tests to entire birth cohorts of young men for the better part of a century. That accident of military bureaucracy is why this literature has sample sizes that psychology usually cannot dream of. For contrast, the childhood IQ and mortality studies report something far larger out of the same kind of found data: one standard deviation of higher childhood score, roughly 15 points, is associated with a 20 to 24 percent lower risk of death over long follow-up. Those childhood scores were normed on children, so an adult who wants a comparable figure today needs adult-normed IQ testing, where the reference group spans ages 16 to 90 instead of a single school year.

Tuvemo, Jonsson and Persson, publishing in Hormone Research in 1999, analyzed 32,887 healthy Swedish conscripts born in 1976 after excluding men with growth-affecting disorders. Mean intellectual performance rose continuously across the height distribution. Men at or below two standard deviations of height averaged 4.22 on the Swedish stanine scale, which runs from 1 to 9 with a population mean near 5, against 5.17 for everyone else. That gap looks dramatic until you notice it compares an extreme tail against the entire remainder of the distribution, which is the sort of contrast that makes any weak association look strong.

The Generation Scotland estimate of 0.16 is more representative of what the association looks like when you correlate the two variables across the whole range instead of slicing off a tail. Marioni's sample had a median age of 57, so this is not a childhood artifact that fades; the correlation is visible in adults who finished growing four decades earlier. It also shows up in women, in twin samples, and across countries with very different nutritional histories, which already tells you something: whatever produces it is not a quirk of one place or one generation.

Two things about the number itself are worth fixing in mind before going further. First, 0.16 is a correlation between two continuous measurements, not a difference between two groups, so any sentence that begins "tall people are" has already lost the plot. Second, correlations of this size are the normal texture of individual differences research. If you want a sense of what larger associations look like, our page on IQ and job performance covers a relationship several times stronger, and even that one fails to predict individuals well.

2 How Small Is a Correlation of 0.16?

Correlations are notoriously hard to feel. The honest way to convey one is to translate it into predictions and then look at how bad the predictions are. Every figure in the grid below is arithmetic performed on r = 0.16, using the modern IQ scale with a mean of 100 and a standard deviation of 15, and taking one standard deviation of adult height as 6.3 cm, the value Tyrrell and colleagues used for UK Biobank in their 2016 BMJ analysis. Arithmetic like that assumes the score was produced on a properly normed scale in the first place, which is the difference between a measurement and a quiz, and it is the subject of a checklist for judging whether an IQ test is accurate.

The fourth figure deserves a second look because it is the one that settles the practical question. A predictor that shrinks your uncertainty from fifteen points to fourteen point eight has told you essentially nothing you could act on. If you want the percentile mapping behind those conversions, the IQ percentile chart shows how score ranges translate into population rank.

3 Childhood Nutrition and Health as a Common Cause

Start with the mechanism that needs no genetics at all. Adult stature is one of the most sensitive records the body keeps of early childhood conditions. Chronic undernutrition, repeated infection and micronutrient deficiency all suppress linear growth, and the same insults, delivered during the same window, impair the development of the nervous system. When one upstream variable pushes two outcomes in the same direction, those outcomes correlate with each other even though neither causes the other. This is the plainest explanation available, and in populations where deprivation is common it is probably the dominant one. This is the same common cause logic that runs through the infant feeding and IQ research, where nutrition proxies and family background travel together.

Sudfeld and colleagues, in a 2015 meta-analysis in Pediatrics covering 68 studies from 29 low- and middle-income countries, quantified the link directly. Each unit increase in height-for-age z score was associated with a cognitive advantage of 0.24 standard deviations among children two years old or younger, and 0.09 standard deviations among older children. Prospectively, a unit of height-for-age before age two predicted a 0.22 standard deviation advantage in cognition measured at ages five to eleven. The strongest signal sits in the first two years of life, which is exactly the window when both skeletal growth velocity and brain development peak.

Single nutrients make the point even more sharply. Bougma and colleagues, reviewing the iodine literature in Nutrients in 2013, found effect sizes between 0.46 and 0.68 standard deviations across four study designs, which they translate to roughly 6.9 to 10.2 IQ points separating iodine-deficient from iodine-replete young children. Iodine deficiency also stunts growth. One trace element, two outcomes, correlation manufactured out of nothing but soil chemistry.

The historical record fits. The NCD Risk Factor Collaboration reanalyzed 1,472 population studies covering more than 18.6 million people in 2016, and found that South Korean women born at the end of the twentieth century were 20.2 cm taller than those born at its start, with Iranian men gaining 16.5 cm. Over a broadly comparable period, Pietschnig and Voracek's 2015 meta-analysis of 271 samples and nearly four million participants documented worldwide test score gains of roughly 0.28 IQ points per year for full-scale performance. Populations grew taller and scored higher at the same time, in the same countries, for overlapping reasons.

4 What Happens Before Birth

Push the window earlier and the same logic holds with more precision. Fetal growth restriction affects skeletal length and neural development together, and because birth length is measured routinely, this channel has been studied in enormous administrative cohorts. Cohorts of that size carry psychiatric outcomes as well, and the mental health branch of the same cognitive epidemiology mostly finds lower early scores associated with slightly higher later risk, with bipolar disorder standing out as the exception that runs the other way.

Bergvall and colleagues, in Pediatrics in 2006, followed 352,125 Swedish boys born between 1973 and 1981 through to conscription testing. Among those born preterm, low birth weight for gestational age did not by itself raise the risk of low intellectual performance. Very short birth length or very small head circumference for gestational age did, roughly doubling it. Among boys born at term, all three measures predicted increased risk. The authors' reading is that during earlier gestation, growth in length and head circumference matters more for later intellectual development than weight gain does. That is a striking result for this article, because it identifies a prenatal process that writes into both the skeleton and the brain at once.

Twin data provide a natural experiment in the same direction. Eriksen, Sundet and Tambs, in Twin Research and Human Genetics in 2012, compared 445,463 Norwegian men and found that singletons scored about 11 percent of a standard deviation above twins, an estimate that held at 11 percent when the comparison was restricted to full brothers within the same family. Adding gestational age barely moved it. Adding birth weight strongly reduced it. Twins are smaller at birth because a uterus is being shared, and that shared constraint shows up decades later in both physical and cognitive measurements.

The size of these effects deserves emphasis. Eleven percent of a standard deviation is under two IQ points. Nobody meeting a twin can detect it, and nobody should try. What matters here is not the magnitude but the direction of the causal arrow, which runs from a prenatal input into both traits rather than between them. Every mechanism in this section adds a little correlation to the population total while adding exactly zero predictive power at the level of a person. The contested part of the fluoride evidence lives in the same band, since the effect argued about at low exposure is smaller than the standard error of measurement on most index scores.

5 The Socioeconomic Channel

Families differ in income, education, housing quality, food security and medical access, and those differences track both how tall children grow and how they perform on cognitive tests. Any correlation computed across families therefore has a social component baked in before biology gets a vote. That social component reappears wherever cognitive scores get correlated with a life outcome; in the well-being research, including a very large UK Biobank analysis, the association shows up most clearly exactly where lower-scoring groups also carry worse health, lower income and less independence.

The economists Anne Case and Christina Paxson worked this channel hard. In their 2008 paper in the Journal of Political Economy, using American and British data, they showed that taller children score higher on cognitive tests on average, and that those test scores account for a large share of the well-documented earnings premium enjoyed by taller adults. Their follow-up with Mahnaz Islam in Economics Letters in 2009, using the British Household Panel Survey, found that most of the labor market height premium is explained by educational attainment and by sorting into higher-status occupations and industries.

That result cuts against the folk explanation. The usual story about tall people earning more invokes social dominance, self esteem or discrimination at the hiring table. Case and Paxson's data suggest a duller mechanism in which early childhood conditions shape both growth and cognitive development, cognitive development shapes schooling, and schooling shapes earnings, with height riding along as a visible marker of the same early history rather than as a cause of anything downstream.

Silventoinen and colleagues added a useful wrinkle in Paediatric and Perinatal Epidemiology in 2013, analyzing over 1.1 million Swedish conscripts. Twins and triplets trailed their singleton brothers slightly on test scores, height, body mass and muscle strength, and the gaps were widest in families of lower socioeconomic position. The same biological starting disadvantage produced a larger lasting effect where the postnatal environment was thinner, which is a compact demonstration that this correlation is not fixed by nature. It is partly a measure of how unequally childhoods are distributed.

6 The Behavioral Genetic Evidence

Environmental accounts run into a problem: the correlation refuses to disappear in affluent, well-fed, well-schooled populations where almost nobody is nutritionally stunted. Generation Scotland is a modern Scottish sample, not a famine cohort, and it still produced 0.16. That is what pushed researchers toward genetic explanations.

The relevant quantity is the genetic correlation, which asks whether the genetic variants that influence one trait tend to be the same variants that influence the other. Marioni and colleagues estimated it directly from measured single-nucleotide polymorphisms rather than from family structure, and reported a genetic correlation of 0.28 with a standard error of 0.09, alongside a bivariate heritability of 0.71. That last figure is the important one: about seven tenths of the observed phenotypic correlation was traceable to shared genetic influences in that cohort.

Twin studies reached similar conclusions earlier. Silventoinen, Posthuma, van Beijsterveldt, Bartels and Boomsma, writing in Genes, Brain and Behavior in 2006, followed Dutch twins from age five into middle age. Height was highly heritable throughout, with estimates from 0.93 to 0.96, and adult full-scale IQ came in at 0.83 to 0.84. Every height and IQ correlation they detected, at every age, was attributable to genetic factors common to both. Their explicit conclusion was that this association should not be read as direct evidence that childhood living conditions drive cognitive outcomes. A separate study by Silventoinen with Iacono, Krueger and McGue, in Behavior Genetics in 2012, examined 756 Minnesota twin pairs at age 11 and 626 at age 17, found that height and head circumference were the anthropometric measures most consistently linked to test scores, and again traced the links to common genetic factors.

None of this means a gene makes people tall and clever. Genetic correlation is a statistical statement about overlapping influence, and it is entirely compatible with variants that affect general developmental processes, cell growth regulation or metabolic efficiency, showing up faintly in two very distant outcomes. It also, as the next section shows, does not even require a biological link at all.

7 Assortative Mating: The Elegant Explanation

Here is the finding that reframes the whole topic, and it is the reason this article exists rather than a shorter one.

Keller, Garver-Apgar, Wright, Martin, Corley, Stallings, Hewitt and Zietsch published an extended twin-family analysis in PLoS Genetics in 2013, modeling the covariation among identical twins, fraternal twins, their siblings and their parents across a combined sample of 7,905 people. That design does something ordinary twin studies cannot: it separates the effects of non-random partner choice from everything else, which matters because assortative mating inflates the resemblance of fraternal twins and siblings and thereby corrupts conventional estimates. After accounting for it, the height and IQ correlation turned out to be almost entirely genetic in origin, and the genetic part split roughly in half between two very different causes. One half was pleiotropy, meaning genuine variants influencing both traits. The other half was gametic phase disequilibrium produced by assortative mating.

That second term deserves plain language. People do not pair off at random. Partners resemble each other on height, and partners resemble each other on education and cognitive test performance. When both kinds of sorting run in the same population for generations, alleles associated with greater height and alleles associated with higher test scores start accumulating in the same individuals, purely because their carriers keep having children together. No allele is doing double duty. No biological pathway connects the traits. A statistical association is built anyway, and it is inherited like any other, because the correlation between the alleles is transmitted along with the alleles.

This is why the topic is genuinely interesting rather than merely tabloid material. Half of a genetic correlation, in the best-designed study available, is a social artifact frozen into a genome. It is also a warning about interpretation: finding that a correlation is "genetic" does not establish that it is biological, and certainly does not establish that it is fixed. Change the mating patterns of a population and that half of the correlation drifts.

8 What Happens Inside Families

The sharpest tool for stripping out confounding is to stop comparing strangers and start comparing siblings. Brothers and sisters share a household, a neighborhood, a family income and half their segregating genetic variation, and each of them received an independently randomized half of each parent's genome at conception. Comparing them therefore removes, in one move, the effects of population structure, of family environment, and of the allele correlations that assortative mating built.

Howe and colleagues assembled the largest such analysis in Nature Genetics in 2022, pooling 178,086 siblings across 19 cohorts to run genome-wide association analyses both between and within families. Their headline result matters far beyond this article: within-sibship estimates were smaller than population estimates for height, educational attainment and cognitive ability, among other traits. Genetic associations measured across a population capture direct inherited effects mixed with demography, assortative mating and the effects of relatives' genotypes acting through the environment. Comparing siblings removes that mixture, and effects shrink when it goes. The within-sibship genetic correlation between educational attainment and body mass index, to take their most vivid example, collapsed toward zero.

Does the height and cognition link survive that treatment? Partly. Case and Paxson, in a 2010 review in Demography drawing on the National Longitudinal Survey of Youth child cohort, found that even among children of the same mother, the taller sibling tended to score better on cognitive tests and to move through school faster. They also traced part of that within-family difference to differences in birth weight and birth length attributable to the mother's behavior during that particular pregnancy, which puts the residual sibling gap back in the prenatal bucket rather than in some mysterious property of tallness.

The field does not agree on the final split, and pretending otherwise would be dishonest. Sundet, Tambs, Harris, Magnus and Torjussen, studying 1,181 identical and 1,412 fraternal Norwegian male twin pairs for Twin Research and Human Genetics in 2005, attributed 59 percent of the phenotypic correlation to correlated shared environments, 35 percent to correlated genes and 6 percent to non-shared environments. That is close to the opposite of the Generation Scotland conclusion. Different cohorts, different birth years, different national histories of nutrition and schooling, different answers. The reasonable position is that the mix varies by population, and that no single number is the truth about all of them.

9 What Mendelian Randomization Adds

Mendelian randomization is the closest thing available to a randomized trial of a trait nobody can randomize. Because genetic variants are allocated at conception and cannot be caused by later life circumstances, a variant that reliably raises height can be used as an instrument to test whether height itself has downstream consequences. If genetically taller people show an outcome difference, confounding by childhood poverty or nutrition is not a plausible explanation, since those things cannot have altered the genotype.

Tyrrell and colleagues ran exactly this analysis in the BMJ in 2016 on 119,669 UK Biobank participants of British ancestry aged 37 to 73. A genetically determined one standard deviation of extra height, 6.3 cm in their sample, caused participants to finish full-time education 0.06 years later, with a confidence interval of 0.02 to 0.09, raised the odds of working in a skilled profession by a factor of 1.12, and raised annual household income by about 1,130 pounds. The effects were stronger in men. On the other side of the ledger, higher body mass index caused lower income in women.

Two honest qualifications. First, the outcomes were socioeconomic, not cognitive test scores, so this is evidence about a causal path from stature into life outcomes rather than into measured reasoning. Second, look at the size. Six hundredths of a year is about three weeks of schooling per 6.3 cm. If somebody wanted to argue that height causes cognitive ability through educational exposure, this is the best instrument anyone has published, and it buys three weeks.

There is also a methodological caution that Howe's within-sibship work makes unavoidable. Mendelian randomization run on unrelated individuals inherits the same contamination as any population genetic association, including assortative mating and indirect parental effects, so published estimates of this kind are plausibly upper bounds. The defensible summary is that if a causal effect of height on cognitive outcomes exists at all, it is small and it operates through social and educational channels rather than through anything happening inside the skull.

10 Two Traps: Growth Timing and Brain Size

Two recurring errors inflate this correlation in people's minds, and both are worth naming.

The first is measurement timing. Case and Paxson observed that children with higher test scores tend to enter their adolescent growth spurt earlier, which means that height measured during the teenage years is partly a measure of maturational tempo rather than of eventual adult stature. A fourteen-year-old who is tall for the class may simply be twelve months ahead on the developmental clock, and classmates will catch up. Any study that correlates adolescent height with adolescent test scores is therefore capturing a timing effect layered on top of whatever the adult association is, and will overstate it. This is a general problem in developmental data and a specific reason to trust adult samples like Generation Scotland more than school-based ones.

The second trap is the intuitive chain that runs from bigger body to bigger brain to higher score. Each link is weaker than it feels. Height and head circumference do travel together, as the Minnesota twin analysis showed. Brain volume and test performance also correlate, and Pietschnig, Penke, Wicherts, Zeiler and Voracek quantified that link in a 2015 meta-analysis in Neuroscience and Biobehavioral Reviews covering 88 studies and 148 samples of more than 8,000 people: r = 0.24, accounting for about 6 percent of the variance. Notably, they also showed that published effect sizes had been inflated by reporting bias and had been declining over time as methods improved, and they argued explicitly against treating brain size as a stand-in for intelligence differences.

Multiply weak links and you get something weaker still. A chain of correlations around 0.2 does not deliver a meaningful path from stature to cognition, and neither does any anatomical story built on it. If you want to understand what cognitive tests actually sample, our explainer on what IQ measures lays out the domains, and fluid and crystallized reasoning covers the distinction that anatomical shortcuts always ignore.

11 Why This Tells You Nothing About Any Individual

This section exists because the preceding ten can be misread, and the misreading is the only genuinely harmful thing on this page. Height is only the most visible member of a large family of such markers, and the traits people read as signs of high intelligence are, every one of them, weak to moderate correlations that cannot be stacked into a measurement.

Return to the arithmetic. With r = 0.16, the standard deviation of your prediction error, after you have been told a person's height, is 14.8 IQ points instead of 15.0. Every practical consequence follows from that sentence. Two people who differ by 15 cm, which is an obvious visible difference, differ in expectation by under six points, and the spread around that expectation is more than twice as wide as the expectation itself. The taller of two randomly selected strangers holds the higher score about 55 percent of the time, meaning height loses this bet almost as often as it wins. Nothing in that neighborhood justifies an inference, a screening rule, a hiring hunch or an assumption about a child.

The point generalizes beyond height. Small population correlations describe the shape of a cloud of millions of points; they say nothing about where any single point sits. Even the strongest predictors in individual differences research would be considered unusable as individual diagnostics, and this one is not among the strongest. Scores also carry their own measurement uncertainty, which is why any serious assessment reports a confidence interval rather than a bare number. A three-point expected difference sitting inside a wider band is not a signal. It is noise that happens to have a name.

There is a second reason for caution that has nothing to do with statistics. Physical traits have a long history of being pressed into service as evidence of worth, always badly and often maliciously. The empirical case for doing that with height is as thin as the numbers above. Someone's stature carries no information about their reasoning that is worth acting on, and treating it as though it does is both a factual error and a social one. For where actual scores sit relative to the population, average IQ and what counts as a good IQ give the calibration.

12 How to Read the Next Small Correlation You Meet

Height and intelligence is a useful training case, because it contains every trap that shows up in headlines about diet, birth order, handedness, screen time and anything else claimed to predict cognition. Four questions handle nearly all of them.

• What is r, and what does it predict? Square it for variance explained, and convert it into an expected difference on the outcome scale. If the answer is a few points on a scale with a standard deviation of fifteen, the story is over regardless of how many zeros the sample size has.
• Was the comparison between families or within them? Across-family comparisons carry family income, neighborhood, schooling and parental behavior along with them. Sibling and twin designs strip those out, and effects usually shrink when they do, as the 178,086-sibling analysis demonstrated across many traits at once.
• At what age was each variable measured? Developmental tempo masquerades as trait level throughout adolescence. Adult measurements of both variables are the only ones that answer the question people think they are asking.
• Does "genetic" mean biological here? Often it does not. Assortative mating manufactures genetic correlations between traits that share no pathway, and it accounted for roughly half the genetic overlap in the best-designed study of this particular pair.

Run those four on a claim and most of them dissolve into a common cause, a confound or a rounding error. The residue that survives is usually smaller and more interesting than the headline promised, which is exactly what happened here: an association of 0.16 that turns out to be a compressed record of prenatal growth, childhood nutrition, family circumstance and centuries of non-random partner choice, none of which can be read off a person standing in front of you. Where the claim involves an exposure rather than a fixed trait, one further question earns its place, which is whether the difference predates the exposure: in the Danish conscript cohort behind the alcohol and IQ evidence, the men who later acquired an alcohol related diagnosis were already 5.5 points behind at conscription, roughly 60 percent of the gap that was visible in midlife.

If the honest version of this question left you curious about your own profile rather than about population averages, that is measurable. ACIS is an online self-assessment, not a clinical or diagnostic instrument, and it reports results as a profile across separate cognitive domains with confidence intervals rather than as a single number. That is the format any measurement worth discussing should take.

13 Frequently Asked Questions

14 Sources Behind This Page

The relationship discussed here comes from published research, and the honest reading includes its limits. These are the primary sources behind the numbers on this page.

• Plomin, R. & Deary, I.J. (2015). Genetics and intelligence differences: five special findings. Molecular Psychiatry. Open access. The standard modern review of what twin and DNA evidence does and does not show.
• Nisbett, R.E. et al. (2012). Intelligence: new findings and theoretical developments. American Psychologist, 67(2). The broad APA review of what moves measured intelligence and what does not.
• Schmidt, F.L. & Hunter, J.E. (1998). The validity and utility of selection methods in personnel psychology. Psychological Bulletin, 124(2). The meta-analysis behind general ability as the strongest single predictor of job performance.
• Strenze, T. (2007). Intelligence and socioeconomic success: a meta-analytic review of longitudinal research. Intelligence, 35(5).
• Deary, I.J., Strand, S., Smith, P. & Fernandes, C. (2007). Intelligence and educational achievement. Intelligence, 35(1). A five year cohort of over 70,000 English schoolchildren.
• Kuncel, N.R., Hezlett, S.A. & Ones, D.S. (2004). Academic performance, career potential, creativity, and job performance: can one construct predict them all? Journal of Personality and Social Psychology, 86(1).
• Whalley, L.J. & Deary, I.J. (2001). Longitudinal cohort study of childhood IQ and survival up to age 76. BMJ, 322(7280). Open access.
• Batty, G.D. et al. (2007). IQ and mortality: a population based cohort study. International Journal of Epidemiology, 36(3), 538-553.
• Zagorsky, J.L. (2007). Do you have to be smart to be rich? The impact of IQ on wealth, income and financial distress. Intelligence, 35(5).
• Judge, T.A., Colbert, A.E. & Ilies, R. (2004). Intelligence and leadership: a quantitative review and test of theoretical propositions. Journal of Applied Psychology, 89(3).