Twin Studies and IQ: What the Correlation Data Actually Shows

The cleanest natural experiment in the study of intelligence is also one of the oldest: find people who share all of their genes, or half of them, raised in the same home or in different ones, and see how alike their IQ scores turn out to be. Decades of exactly that comparison produced one of the most replicated patterns in the behavioural sciences, and one of the most misread.

The logic of the comparison

Identical, or monozygotic, twins develop from a single fertilised egg and share close to all of their DNA. Fraternal, or dizygotic, twins develop from two separately fertilised eggs and share, on average, the same proportion of DNA as any two siblings — roughly half, though the actual shared fraction for any specific pair varies around that average by chance. If genes influence IQ at all, identical twins should end up more alike on it than fraternal twins, and the size of that gap, compared across pairs raised together and pairs raised apart, is what lets researchers estimate how much of the variation in IQ scores across a population tracks genetic relatedness rather than shared upbringing.

That estimate is called heritability, and the term is worth pausing on because it is one of the most commonly misunderstood figures in psychology. Heritability is not a measure of how "genetic" one person's IQ is — an individual score does not split into a genetic portion and an environmental portion any more than the area of a rectangle splits into a "length portion" and a "width portion." It is a population statistic: the proportion of variation in scores across a specific group of people, in the environments that group actually experienced, that statistically tracks genetic differences among them. Change the population or the range of environments, and the number can change with it.

What the classic studies found

The Minnesota Study of Twins Reared Apart, led by Thomas Bouchard at the University of Minnesota beginning in 1979, tracked identical and fraternal twin pairs separated in infancy and raised in different households, then reunited as adults for extensive testing. A wider review of decades of earlier twin, sibling and adoption research, published by Bouchard and John McGue in 1981, pooled the correlations across dozens of independent studies and produced the pattern the field still cites.

• Identical twins reared together — pooled correlation roughly 0.86, the highest of any pairing measured.
• Identical twins reared apart — roughly 0.75 in the Minnesota sample, and around 0.72 in the wider pooled review, despite growing up in different homes.
• Fraternal twins reared together — roughly 0.60.
• Ordinary siblings reared together — roughly 0.47, similar to fraternal twins despite the less controlled comparison.
• Adoptive siblings with no biological relationship, reared in the same home — roughly 0.15 to 0.19.
• Biologically unrelated people reared in separate homes — close to zero.

That last comparison is the one that carries the argument. Shared genes without a shared household — identical twins reared apart, roughly 0.75 — produced a stronger correlation than a shared household without matching genes — fraternal twins or ordinary siblings reared together, roughly 0.47 to 0.60. Whatever is driving the similarity, it is tracking genetic relatedness more closely than it is tracking which house someone grew up in.

A second, independent design agrees

Twin studies are not the only design pointing the same way. Adoption studies compare children raised by adoptive parents against both their adoptive and their biological parents, removing the shared-genes-and-shared-home overlap that complicates ordinary family comparisons. Research programmes including the Texas Adoption Project and the Colorado Adoption Project found the same basic pattern from a completely different angle: adopted children's IQ scores correlate more strongly with their biological parents, whom many of them never lived with, than with the adoptive parents who raised them — and that correlation with biological parents tends to strengthen as the children move into adulthood, mirroring the age trend twin studies also find. Two different research designs, run by different teams on different samples, converging on the same conclusion is a stronger form of evidence than either design produces alone.

What heritability does not mean

None of this licenses the jump people usually make from it. A heritability estimate describes variation within the specific population that was actually measured — mostly, in these studies, mid-to-late twentieth-century participants in the United States and a handful of other countries. It says nothing about why any one person scored what they scored, and it says nothing about differences between separate populations or across very different environments, because the studies were never designed to compare across those boundaries in the first place. A statistic built entirely from within-group comparisons cannot, by its own construction, explain a between-group difference. This is the same distinction that undermines every attempt to explain cross-national score comparisons with heritability figures drawn from single-country twin samples — the statistic simply was not built to travel across that kind of boundary.

Not a fixed number, even within one population

Longitudinal work that retests the same twin pairs across childhood and into adulthood has found something that surprises people expecting one constant answer: heritability estimates for IQ tend to rise with age, from more modest figures in early childhood toward the higher figures usually quoted for adults, a pattern sometimes called the Wilson effect after the longitudinal Louisville Twin Study that first tracked it. One proposed explanation is that as children grow older they increasingly select and shape their own environments in ways that align with their genetic predispositions, so genetic influence and environmental influence become harder to cleanly separate rather than easier. Whatever the mechanism, a single heritability percentage was never going to be the whole story even for the population it was measured in.

What this is useful for, and what it is not

The honest use of this research is narrow: it tells you that within the populations studied, genetic relatedness tracks IQ similarity more closely than shared household does, and it rules out the simplest environmental story — that upbringing alone explains why family members' scores resemble each other. It does not tell you what caused any individual's score, it does not transfer to comparisons between different populations or eras, and it does not diminish the role environment plays in the aggregate — a correlation of 0.75 among reared-apart identical twins still leaves real, measurable room for everything that was not genetically shared. Readers checking a specific score against a real distribution are better served by seeing where a given result sits on the population curve than by any single percentage attached to nature versus nurture.

The number that survives all the caveats is the comparison, not a percentage: identical twins end up more alike than fraternal twins, reared-apart identical twins end up more alike than reared-together non-twin siblings, and that gap has shown up in study after study for more than forty years. What it means for any one household — including whether a firstborn's edge described in our note on birth order is itself partly a shared-environment effect — is a separate, much harder question that a correlation coefficient was never built to answer alone.

Common questions