Personality seems intimate and self-authored. The patience brought to a difficult conversation, the appetite for a crowded room, the tendency to worry before anything has gone wrong: these feel like products of a life rather than entries in a genome.
Yet decades of twin and family research have repeatedly found that genetic differences account for a substantial part of why people differ in broad personality traits. Depending on the study and trait, the familiar estimate is roughly 40 to 60 per cent.
A vast new genetic analysis appears, at first glance, to tell a very different story. Examining data from as many as 1.14 million people, researchers estimated that common DNA variants captured between 4.8 and 9.3 per cent of variation in the five major personality traits in their main analysis.
The missing percentage points have not disproved the twin studies, nor have they shown that genes barely matter. The two approaches measure related but importantly different things. The space between them is one version of the “missing heritability” problem, and it reveals how difficult it is to move from the observation that personality runs in families to an account of the particular DNA differences involved.
Heritability is not a percentage of a person
The word heritability invites a misleading interpretation. A personality trait cannot be divided into a genetic portion and an environmental portion inside one individual. Someone is not 50 per cent outgoing because of DNA and 50 per cent outgoing because of experience.
Heritability is a population statistic. It estimates how much of the variation among people, in a particular population living in particular environments, is statistically associated with genetic differences. Change the population, its environments, the way a trait is measured or the ages of the people studied, and the estimate can change.
A trait can also be heritable without being fixed. Eyesight is influenced by genetics, yet glasses change what a person can see. Vocabulary has genetic influences, yet education matters. In the same way, genetic influence on neuroticism or conscientiousness says nothing simple about whether a person can change.
This is also why subtracting a heritability estimate from 100 does not deliver a neat measure of “the environment.” Genes and environments can be correlated or interact with one another, measurement contains noise, and every statistical model makes choices about which effects it can see.
How twins produce the larger estimate
Twin studies do not normally identify any DNA variant. Instead, they compare patterns of resemblance. Identical twins share essentially all of their inherited DNA, whereas fraternal twins share about half of their segregating genetic variants on average. If identical twins resemble one another more strongly on a trait, models can attribute part of that extra similarity to genetic influence.
Adoption studies, pedigrees and comparisons of twins raised together or apart add other forms of information. Combined across many studies, these designs have produced the familiar conclusion that broad personality differences are moderately heritable.
A recent ScienceBlog examination of identical twins raised apart offered a useful companion lesson. Twins sharing essentially the same genes still differed when their educational histories diverged. Family resemblance can reveal genetic influence without locking a complex human outcome into place.
The strength of twin designs is their ability to detect genetic influence without knowing the variants responsible. Their vulnerability is that the estimate depends on assumptions. These include how similarly identical and fraternal twins experience relevant environments, how assortative mating is handled, and whether genetic effects are additive or involve dominance and interactions. Modern methods test and relax some assumptions, but no family design is assumption-free.
What the Nature study counted
The new Nature analysis brought together 46 cohorts of people with broadly European genetic ancestry and another ten cohorts with broadly African genetic ancestry. Depending on the personality trait, its main meta-analysis included between 611,037 and 1,136,711 participants.
Participants had completed measures of the Big Five: extraversion, agreeableness, conscientiousness, neuroticism and openness. The researchers tested roughly ten million genetic variants per trait and identified 1,260 approximately independent lead variants associated at genome-wide significance. Of those, 824 had not appeared in earlier personality scans.
Those 1,260 findings are not the source of the entire 4.8 to 9.3 per cent estimate. That number is SNP heritability, an estimate of the combined contribution tagged by common single-letter DNA differences across the genome, including variants whose individual associations are far too small to reach genome-wide significance.
In the principal analysis of European-like cohorts, common variants accounted for 4.8 per cent of measured variation in agreeableness and 9.3 per cent in extraversion, with the other traits between those points. A random-effects approach that allowed genetic effects to vary more across cohorts produced an average of 8.6 per cent, ranging from 7.4 per cent for agreeableness to 10.6 per cent for extraversion.
Personality is spread across thousands of effects
Even in a study of more than a million people, the typical detectable effect was minute. After adjustment for the tendency of initial discoveries to look larger than they really are, the median lead variant for extraversion shifted a person by about 0.009 standard deviations. The paper translates that into a movement from roughly the 50th to the 50.35th percentile.
The researchers estimated that each trait involves about 16,180 independently associated common variants. That does not mean exactly 16,180 “personality genes.” Variants can regulate genes from a distance, the same biological pathways can affect several traits, and the statistical estimate describes a diffuse genetic architecture rather than a tidy list of switches.
Large samples make it possible to detect ever smaller average associations. They do not turn those associations into destiny or make personality readable from one person’s genome with clinical precision. Polygenic scores built in the study predicted some variation in independent samples, but population-level prediction remains very different from explaining an individual life.
The distinction matters because the phrase “genes for personality” suggests a direct route from sequence to temperament. In reality, common variants appear to alter biological tendencies by tiny amounts, against a background of development, relationships, culture, chance and the environments people partly select for themselves.
Imperfect tests hide some heritability
Personality is harder to measure than height or blood chemistry. Cohorts used different questionnaires, different numbers of questions and sometimes different raters. A short self-report inventory does not capture a trait perfectly, and random measurement error pushes heritability estimates downward because genes cannot explain noise.
The new study found that more reliable personality instruments yielded higher SNP-heritability estimates. For tests around the median reliability in the dataset, a coefficient alpha of 0.81, the expected estimates rose to between 9.3 and 13.3 per cent. Correcting completely for measurement error raised the range to about 10.8 to 15.8 per cent.
That closes a meaningful portion of the gap, but not all of it. It also illustrates why the main 4.8 to 9.3 per cent range should not be treated as a final biological ceiling. It describes the variance captured in observed scores under a particular model, not every conceivable genetic contribution to an ideally measured personality trait.
The same problem of method-sensitive heritability appears elsewhere. A recent ScienceBlog look at Danish twins and lifespan explained how estimates can shift when researchers change which deaths count as expressions of intrinsic ageing. Personality and lifespan are very different traits, but both show that a heritability number depends on how the outcome is defined and measured.
Common SNPs cannot see every genetic difference
Genome-wide association studies are best at studying common variants that are measured directly or can be inferred from nearby markers. Rare variants may exert larger effects while contributing too little statistical information to be captured well by a standard common-variant analysis.
Whole-genome sequencing can eventually recover more of that information, but the sample sizes required for rare-variant analysis are formidable. Personality adds another obstacle: a rare variant can be genotyped exactly while the behavioural trait it is being compared with remains an imperfect questionnaire score.
Standard SNP-heritability models also focus largely on additive effects, the small contributions that accumulate across variants. If two variants interact, if genetic effects depend on an environment, or if a variant’s effect changes across development, part of that structure may not appear in the main estimate.
A long-running review of missing heritability in complex traits laid out many of these possibilities years before the present personality sample existed. Better arrays, sequencing and statistics have recovered more signal since then, but complex human behaviour remains an especially demanding test.
The assumptions are not all on one side
It is tempting to regard the twin estimate as the true total and ask only why molecular genetics falls short. That framing is too simple. Family studies can capture forms of resemblance that their models classify as genetic but that may be partly shaped by assortative mating, indirect genetic effects from parents, or environments correlated with genotype.
Conversely, molecular studies can miss genuine genetic effects because their markers, samples and models are incomplete. The 40 to 60 per cent and 4.8 to 9.3 per cent figures therefore should not be placed on a single measuring tape as if one precisely defines the amount the other has failed to find.
The new paper included within-family analyses of as many as 50,725 people. Comparing genetic associations within families helps reduce confounding from ancestry differences, parental effects and assortative mating. Those analyses produced SNP-heritability estimates ranging from 7.6 to 13.4 per cent and suggested that the main common-variant signal was not primarily an artefact of population structure.
That is reassuring, but it does not certify every assumption behind either method. The within-family sample was much smaller than the full analysis, its estimates were less precise, and it still examined variants available to the genetic data and models.
A million-person study still has boundaries
The largest analyses were conducted in cohorts grouped as European-like. The researchers also examined African-like cohorts, but their sample sizes were far smaller. Genetic associations and polygenic scores do not necessarily transfer cleanly between ancestry groups because variant frequencies, correlations among markers and environmental contexts differ.
The associations were broadly similar across instruments, age groups, raters and clusters of Western countries, but they were not identical. That heterogeneity is not a nuisance to be erased. It is evidence that “personality” is being observed through different questions, cultures, ages and social settings.
Nor can an association establish a simple biological chain from one variant to one behaviour. The study connected signals with genes, tissues and other traits to generate hypotheses, but identifying a statistical region is only the beginning of understanding what it does.
The results are strongest as a map of architecture: personality traits are influenced by very many common variants, most effects are exceedingly small, and better measurement recovers more of their collective signal.
What remains missing
Some of personality’s missing heritability may be in rare DNA variants. Some may lie in interactions that current models simplify. Some may be hidden by noisy personality tests, age-dependent effects or populations missing from the largest datasets. And some of the apparent gap may narrow as the assumptions behind older family estimates are examined with more direct genetic data.
There may never be one satisfying percentage that applies to every population and every version of a trait. Heritability is not a universal constant like the charge of an electron. It is an estimate tied to people, environments, instruments and models.
That makes the new study more revealing, not less. It has found far more of the common-variant architecture of personality than earlier efforts could see, while demonstrating how much does not fit inside a simple headline about nature versus nurture.
The missing heritability problem is not an empty space where a personality gene ought to be. It is the distance between several imperfect views of an unusually complex human outcome, and narrowing it will require better genomes, better behavioural measurements and greater care about what each estimate actually means.