Estimates of the heritability of intelligence do include gene-environment interactions

Intelligence is necessarily about gene-environment interactions, not about genes and environment acting independently

I recently wrote a piece arguing that estimates of the heritability of intelligence that derive from twin studies are likely to be more accurate than the lower estimates from GWAS studies (Genome-Wide Association Studies). This is because GWAS studies lack the statistical power to find many genes with small effects. But in defending the high estimates of "heritability" (the fraction of human differences that can be attributed to genes) it is important to realise what this actually means.

The classic "twin studies" method takes twin babies who were given up for adoption at birth and then separated and raised in different families. Comparing the similarities in later-life outcomes for identical twins (who share all their genes) to those for fraternal twins (who share half their genes, as siblings do) and those of unrelated adoptees, tells us how much of the differences in life outcomes derives from differences in our genes (the "heritability"), versus differences in our "shared" environment (all environmental factors that siblings living together would share) and also differences in our "unshared" environment, a category that includes everything else from chance factors in embryonic development to random life events.

Twin studies tell us that differences in intelligence are roughly 70 per cent genetic in origin, with 20 per cent attributable to "unshared" environment, and only 10 per cent to "shared" environment. Such results have been replicated many times and are robust. But some people regard this as counterintuitive; they intuitively think that parenting and upbringing must be more important (perhaps they see the effects of the genetic similarity of parents and children and erroneously attribute them to the family environment). Similarly, whether a child goes to a "good" school versus a "bad" school matters less for their prospects than is often supposed, since the evidence is that the difference between a good and a bad school is mostly about the intake of kids.

As an aside, and to illustrate the vast gulf between science and popular culture on this topic, a serious newspaper, The Guardian, told us that: "Growing up in a home packed with books has a large effect on literacy in later life". No, it really does not. But parents who are intelligent and love reading, both: (1) have lots of books in the house, and (2) pass on genes for liking reading to their children. The article doesn't even mention the latter possibility.

As this illustrates, attributing human differences to genetics is often considered distasteful, even though it is amply supported by the science. This leads to people pointing to the lower estimates from GWAS studies, or arguing that environment must play a larger role than indicated by the "headline" twin-studies numbers. For example, Harvard geneticist Sasha Gusev argues that much of what twin studies attributes to genes is actually the effect of complex gene–environment interactions. He's right, but let's consider whether that makes the estimates invalid.

First, it is important to understand that heritability estimates pertain only to the range of environments in the study that yielded that estimate. A study of separated twins in the UK would still have the adopted children going to fairly similar schools and having a similar education. If, as a thought experiment, half were adopted into peasant families in Medieval England, where children helped their parents in the fields and did not go to school, then the greater disparity in environments would increase the disparity in outcomes, and hence the heritability would be lower. Conversely, if one could make all the adopted environments identical, then the heritability estimates would be higher, since genetic differences would be the only differences.

Now consider sending two unrelated children, Sue and Jim, to the very same school. Sue has genes that make her naturally interested in and good at mathematics, but Jim does not. Do they (being sent to the same school) have the same environment? Likely the school would recognise Sue's ability and would encourage her interest, placing her in the top set for maths where she would be surrounded by similar kids and be stretched by advanced material. Jim would likely reside in a lower set, being given basic material, and the teachers could well react to his lack of interest by not pushing him. Obviously, Sue's genetic advantage in maths would be intensified by her school's encouragement and coaching, whereas Jim's disadvantage would also be compounded.

Hence the differences in their final exam scores would not be just down to genes, but also to how they interact with their school environment, and hence would be about complex gene–environment interactions. If all maths-able kids attend schools that encourage maths ability (which is indeed likely to be the case, because that's what schools do), then twin studies would attribute all of Sue's out-performance to genetics. That's because the environmental condition: "school that does not encourage able kids" would not be present in the study and so the study would be blind to it.

If able kids are usually able to "make their own environment" by gaining access to books, libraries, museums and adult encouragement, then the resulting boost to their abilities would be recorded as a genetic effect in the twin-studies ledger (and again, that's because the relevant "control", the able child in an environment where they cannot access those things, is unlikely to be sampled in the study).

Indeed, such an effect has long been known. The estimates for the heritability of IQ from twin studies are lower when measured in teenagers than when measured in later-life adults. That's because teenagers have less control over their environment, they're made to read books and do academic work whether they like it or not. But a less-able adult can indulge their natural inclination by never reading a book again, whereas the naturally able adult will seek out intellectual stimulation. And, again, these gene-environment interactions will be scored as a genetic effect in twin studies.

So what, then, do we make of the heretibility estimates from twin studies? Sasha Gusev writes: "The gap between low [GWAS] heritability estimates and high twin heritability estimates could thus be explained by ... [gene–environment] interactions incorrectly assigned to genetics by the latter (“missing environments”)". He then asks: "Could it be that twin studies have been estimating gene–environment interactions this whole time?".

The answer is Yes! Twin studies do arrive at high heritability estimates by attributing such gene–environment interactions to the "genetics" ledger. But is that the wrong thing to be doing? The answer (as often in science) depends on exactly what question one is asking.

Partly this comes down to how one connotes "environment". We could regard an environment as being something fixed, that makes no response to how the child behaves. But when it comes to behavioural traits that's not how the world works. Society does not treat a badly-behaved child identically to a well-behaved child.

Instead we could connote "environment" in a more responsive way, being families, schools and neighbourhoods that contain parents who play in more stimulating ways to a more responsive child; and sports teams that are available to those kids who express an interest and are talented; and adults who will encourage a child who shows an interest in music and who will teach them to play the piano. Sue and Jim would both be in that same overall "environment" but (owing to their genetic disposition) would exploit it differently.

Thus, if the question is about sending children to British schools and asking what factors explain the differences in outcomes, then the high heritability from twin studies is the appropriate answer. For two children sent to the same school, their natural dispositions and abilities are what makes the biggest difference, even though that is mediated by them interacting with their environment differently.

If the question were different, if one was asking about a hypothetical society which did not treat children any differently according to their ability, then the current heritability estimates would not be valid, because studies for those environments have not been done. In practice, the former question is likely to be the one more relevant for understanding today's world and for public policy. After all, it would be a weird school that treated a pupil who scored 8/100 on a maths test identically to one who scored 92/100.

But heritability estimates are not "fixed". They only say what happens in the range of environments studied, so they don't tell you about the effects of interventions that have not been tried. It's also worth pointing out that the same difficulties in disentangling purely environmental effects from gene-environment interactions also affect GWAS studies. They have no easy way of distinguishing the two any more than twin studies do, and information about environment that helps to do this can be fed into twin studies as easily as into GWAS studies.

Hence, echoing the conclusion of my first piece, the high heritability estimates given by twin studies do seem to be sound and valid. But when it comes to human social and behavioural traits the concept of a fixed and unresponsive environment is not appropriate. Our genetic recipe plays out in a social environment dominated by interactions with other humans.

Indeed, when it comes to behavioural traits like intelligence, the concept of a genetic component that is not about gene–environment interactions, but is purely about genes that are acting independently of environment, is not even coherent. Just for starters, a child, however innately intelligent, would not even have language except through interaction with their environment.