Peter Visscher on the Genomics of Complex Human Traits
In NPJ Science of Learning [corrected 5/15/22], the quantitative geneticist Peter Visscher published a short piece about the application of genomics in general and polygenic scores in particular to human social science, education and neuroscience. (https://www.nature.com/articles/s41539-022-00124-z) The article is a good example of an overenthusiastic view of the relevance of genomics to complex human traits, especially behavioral ones.
Visscher starts with Paige Harden’s characterization of inheritance as a “genetic lottery.” In fact, as Visscher points out, there are two lotteries: the first involving who your parents are, and the second being the random process by which you inherit DNA from your parents, conditional on the outcome of the first lottery. He doesn’t dwell on the fact that the first lottery is, from the point of view of genetics, thoroughly rigged: the DNA your parents carry is related to their ethnicity, their skin color, their socioeconomic status, everything about them that is related to the trait of interest. Even referring to parental selection as a “lottery” manifests the old-fashioned hereditarian error, dating from Galton; "of misattributing observed differences among classes of people to differences in their genes;; It is– and I apologize for using the word, because I am sure Visscher doesn’t intend it that way– eugenic.
The second lottery, the one that differentiates you from your siblings, actually deserves the metaphor. The coolest scientific development of the SNP era has been the newfound ability to differentiate random genetic processes operating within families from the socially stratified ones that operate between them. (I use the term stratification loosely here, to refer to population stratification per se, as well as other environmental, familial, and population-level confounds of genetic associations.) But a funny thing happened on the way to genetic explanations of human differences. As the recent report of the educational attainment of 3 million individuals demonstrated– Visscher was an author– a polygenic score estimated between families loses three-quarters of its predictive power when it is applied within sibling pairs.
After pointing out the between- and within-family distinction at the beginning of the piece, Visscher never mentions it again. He advertises that the stratified estimate accounts for 15% of the variance of EA, without re-emphasizing that this is between families, including every single way that families are stratified in our unequal world; The within-pair estimate is less than 5%. Every study Visscher cites uses only the between pair PGS. This is the standard move among contemporary center-hereditarians: emphasize the between family estimate when you are trying to impress with effect size, then switch (as eg. Paige Harden or Robert Plomin do; Visscher doesn’t bother) to the within-family estimate when you are differentiating yourself from the eugenicists or making a philosophical case for genetic causation. But most of the time, you can’t have it both ways. If the within-pair estimate is the one we believe to be scientifically accurate and socially responsible, that has to be the one we actually use.
You can see the consequences of evading the difference between stratified and unstratified estimates of genetic effects in both the scientific and applied realms. In population genetics as opposed to social science, a scientific conclusion about genetic causation that was based on a genetic effect that was known to be up to three-quarters stratification would not be taken seriously. (See, for example, the assertion and eventual retraction of claims of selection for height along a north-south cline in Europe, based on far less stratification than appears to be at play for education.) Visscher’s vague assertions that genes “matter” and that human differences are “due to” genetic differences may well be true as a matter of statistics, but they must be quantified specifically. We have learned that about 15% of stratified differences among families are “due to” genetic differences, but only a quarter of that percentage is plausibly “due to” unconfounded differences among siblings. What is the proper scientific conclusion? I would say we have learned that once social stratification in its various forms is controlled, genetic differences are a statistically significant but causally minor source of educational differences. Such small statistical effects may well find useful applications in social science, but if you want to make it the basis for your hereditarianism you are going to have an uphill climb. Arthur Jensen, it is worth remembering, based his theorizing about human intelligence on a heritability coefficient of .8.
The implications for real-world applications of genomics are even more severe. Visscher is a bit coy about whether he thinks the educational applications of polygenic scores that he imagines would actually be a good idea, but consider a small child (let’s give her poor parents and brown skin) seeking assignment to a first-grade curriculum. Which polygenic score do you want to use? The one that is confounded with discrimination based on her skin color and her family’s socioeconomic history, or the one that makes potentially causal distinctions among her and her siblings? If the latter, then your justification needs to be based on an instrument that accounts for less than 5% of differences in educational outcome. If what we care about is scientific assessment of her intellectual ability, why not use a well-validated cognitive measure instead? Based on current knowledge, the only reason to prefer genomics is a misplaced belief that somehow genes get at her essential intellectual potential, at which point we are back to Galton.
Yes, I know: When EA17 is based on 30 million individuals instead of 3, the percentages of variance explained may be higher. But they also may not, or we might know a lot more about how stratification inflates them. In any case, our theorizing now should be based on the available science now. Modern genomics has changed very little in our understanding of the causes of human socioeconomic and educational inequality.