A 2025 reanalysis of identical twins raised apart found a striking gradient in their IQ scores. Pairs classified as having very different schooling differed by an average of 15.1 points, while pairs with similar schooling differed by 5.8 points.

The result came from a historical dataset, not an experiment that assigned children to different schools. Jared C. Horvath and Katie Fabricant assembled individual IQ scores and educational biographies for 87 monozygotic twin pairs reported across a century of research, then asked whether the twins’ resemblance changed with the similarity of their education.

This is one study, not settled consensus. The finding shows that near-identical inherited DNA does not guarantee near-identical test scores. It cannot establish that education alone caused the full 15-point gap.

Why separated twins became a natural experiment

Monozygotic twins develop from the same fertilized egg and share almost all of their inherited DNA. They are not literally identical in every biological detail, but they offer researchers a rare way to hold genetic similarity unusually constant.

If twins separated early in life remain alike as adults, that resemblance is evidence for genetic influence. If they differ, the contrast directs attention toward environments, experiences, measurement error and biological changes that occur after the embryo splits. Neither side of the comparison works as a pure meter of “nature” or “nurture.”

The most influential modern example was the Minnesota Study of Twins Reared Apart. Thomas Bouchard Jr. and colleagues reported in Science in 1990 that about 70 percent of IQ variation in their sample was associated with genetic variation. The authors themselves noted that this did not diminish the importance of education or parenting.

Heritability is a population statistic, not a permanence rating for an individual.

The researchers rebuilt a century of scattered data

For their paper in Acta Psychologica, Horvath and Fabricant searched group studies and individual case reports for separated twin pairs that had both person-level IQ results and usable schooling histories. Only three of seven relevant group studies disclosed scores for individual pairs. Sixteen of 33 case studies met the inclusion criteria, and two more pairs came from personal communication with an earlier researcher.

The resulting dataset covered 87 pairs for the absolute IQ-gap analysis and 84 for the within-pair correlation. The authors also excluded the published twin cases of psychologist Cyril Burt after much of his work was found to be fraudulent.

Schooling difference was reconstructed across three dimensions. The researchers considered whether twins attended different types of schools, whether they studied in different countries or US states, and whether their time in education differed by more than a year. Each pair received a score and was placed in a similar, somewhat dissimilar or very dissimilar category.

This was an ingenious use of imperfect records, but the imperfections matter. The biographies were written for different purposes across many decades. School names were generally unavailable, curricula and teaching were inferred from location and school type, and an unknown item received a midpoint score. The categories are the researchers’ reconstruction, not measurements planned when the twins were children.

What the 15.1-point average does and does not mean

Across all 87 twin pairs, the mean absolute IQ gap was 8.6 points. The gradient appeared when the researchers separated the educational histories:

  • 52 pairs with similar schooling differed by 5.8 points on average.
  • 25 pairs with somewhat dissimilar schooling differed by 12.1 points.
  • 10 pairs with very dissimilar schooling differed by 15.1 points.

The headline number therefore rests on 10 pairs. Their gaps also varied considerably, with a standard deviation of 7.1 points. A larger sample could move the estimate in either direction.

“Absolute difference” ignores direction. It measures the distance between two scores, regardless of which twin scored higher. The paper does not show that one specific kind of school raised IQ by 15 points, nor that a child placed in a different system would predictably gain or lose that amount.

The correlation analysis tells a complementary story. The intraclass correlation, which describes resemblance within pairs, was 0.87 among twins with similar education and 0.56 in the very-dissimilar group. The lower figure suggests substantially less resemblance, but it also shows that the twins did not become statistically unrelated.

Education probably matters, but causation remains unresolved here

The broader literature makes an educational effect plausible. Stuart Ritchie and Elliot Tucker-Drob’s 2018 meta-analysis in Psychological Science combined 142 estimates from 42 datasets involving more than 600,000 participants. Using designs based on policy changes, school-entry cutoffs and earlier intelligence measurements, it estimated an improvement of roughly one to five IQ points for each additional year of education.

That study used quasi-experimental methods designed to get closer to causation. The 2025 twin paper did not. It sorted retrospective observations according to schooling difference, so it remains vulnerable to reverse causation and confounding.

A twin’s earlier cognitive performance could influence how long they stayed at school or which educational track they entered. Different countries, school systems and durations may also coincide with differences in language, family income, nutrition, health, discrimination, books at home and exposure to stress. The new analysis could not isolate those factors from schooling itself.

IQ should also be read as performance on standardized cognitive tests, not as a complete inventory of judgment, knowledge or creativity. ScienceBlog’s earlier explanation of why scores on different mental tests tend to move together describes why IQ captures a real, repeatable pattern. That does not make the score independent of learning, test familiarity or context.

Genetic influence is not genetic destiny

The tempting response is to make the paper a victory for one side of the nature-versus-nurture argument. I think its contribution is more precise. It asks whether researchers have treated separated twins’ environments as more different than they really were, and whether pooling all pairs conceals an educational pattern.

The result does not erase the substantial resemblance repeatedly observed among identical twins raised apart. Across the full sample in the new paper, the IQ intraclass correlation was 0.80. Genes plainly remain part of the explanation.

At the same time, heritable does not mean unchangeable. An open review in npj Science of Learning explains that heritability describes variation within a particular population under particular environmental conditions. It does not say what fraction of one person’s intelligence was “caused” by DNA, and it does not place a ceiling on environmental effects.

A stronger follow-up would measure twins before their educational paths diverged, document schooling and other conditions prospectively, and test both members with the same instruments at the same ages. Because identical twins genuinely raised apart are so rare, the next advance may instead come from releasing person-level records from older studies and testing whether the gradient survives different, preregistered ways of classifying education.