What twin studies actually measure

The reasoning compares the resemblance of monozygotic twins, who share all of their genetic material, with that of dizygotic twins, who share on average half, like ordinary siblings.

A difference in resemblance between the two types is attributed to the genetic share, under an explicit assumption: that both types of twins share an equally similar environment. It is that equal environments assumption which critics contest.

Adoption studies offer a second route, comparing a child with their biological parents and with their adoptive parents. They rest on other assumptions, notably that placements are not decided according to the families' characteristics, an assumption that is also disputed.

It does not say what share of a trait is genetic in a given person. It is a population statistic, describing the share of observed variation in a given group at a given time.

Nor does it say a trait is immutable. A highly heritable trait can be altered by a change of environment, average height over the twentieth century being the most cited example. Another classic example is phenylketonuria, a hereditary disease whose effects are largely prevented by an adapted diet: a trait of genetic origin whose expression depends on the environment.

Heritability rises when the environment becomes more homogeneous, since the remaining variation can only come from elsewhere. It falls when living conditions are very unequal.

Several American studies, including those of Eric Turkheimer, found lower measured heritability of cognitive scores in disadvantaged settings. That result has not been found in every country, and it is disputed.

A regular finding of these designs is that the share attributed to the environment shared by members of the same family is small for many traits measured in adulthood, and larger in childhood.

That finding fed a durable controversy, opened by Judith Rich Harris in 1995, about the relative weight of family and peers. Critics stress that the shared environment measure poorly captures what varies between families in very different circumstances.

Genome-wide association studies now allow the contribution of directly measured variants to be estimated. The values obtained are systematically lower than heritability estimated from twins, a gap known as missing heritability.

Polygenic scores built from that work predict a modest share of variation, and mainly in the populations on which they were built, which sharply limits their reach.

Part of the gap is thought to stem from rare variants and from effects these methods capture poorly, part from overestimation by the twin design itself; how the gap divides between these explanations remains debated.