Charles Murray has a new book out. Yay.
Hadn’t heard of Human Diversity: The Biology of Gender, Race, and Class? Well, that’s because his publisher didn’t make review copies available. So the first places to get their reviews out were those feeling no need to take a critical view of the book, like openly white supremacist sites and The Federalist. Unsurprisingly, they think it’s great.
The reviews that do engage critically will take longer. “Critically” here means “Does this accurately represent our best knowledge of the subject?”, rather than “Ugh, don’t like.” Not that those are mutually exclusive. Asking those questions take time.
Some scientists and science communicators have already gotten a head start, however. They can do that without seeing the book because they’ve been dealing with the “evidence” and the arguments on this topic for ages. And of course, because heritability is an easily misunderstood topic, there are some good explainers out there.
So what should you read if you want to learn about heritability from experts rather than political scientists whose prior work on the topic hasn’t held up? Try these articles on the basics and methodological challenges of studying heritability. (Please note that some of these sources uncritically discuss the history of scientific work aimed at a “cure” for autism.)
The National Institute of Health has a good, basic definition of “heritability” as used among scientists:
Heritability does not indicate what proportion of a trait is determined by genes and what proportion is determined by environment. So, a heritability of 0.7 does not mean that a trait is 70% caused by genetic factors; it means than 70% of the variability in the trait in a population is due to genetic differences among people.
Nature provides an overview of how heritability is estimated and some of the problems involved:
Interestingly, heritabilities are not constant. For example, estimates of heritability for first lactation milk yield in dairy cattle nearly doubled from approximately 25% in the 1970s to roughly 40% in recent times. Heritability can change over time because the variance in genetic values can change, the variation due to environmental factors can change, or the correlation between genes and environment can change. Genetic variance can change if allele frequencies change (e.g., due to selection or inbreeding), if new variants come into the population (e.g., by migration or mutation), or if existing variants only contribute to genetic variance following a change in genetic background or the environment. The same trait measured over an individual’s lifetime may have different genetic and environmental effects influencing it, such that the variances become a function of age. For example, variance in weight at birth is influenced by maternal uterine environment, and variance in weight at weaning depends on maternal milk production, but variance of mature adult weight is unlikely to be influenced by maternal factors, which themselves have both a genetic and environmental component. Heritabilities may be manipulated by changing the variance contributed by the environment. Empirical evidence for morphometric traits suggests lower heritabilities in poorer environments, but not for traits more closely related to fitness (Charmantier & Garant, 2005). Understanding how heritability changes with environmental stressors is important for understanding evolutionary forces in natural populations (Charmantier & Garant, 2005).
Jay Joseph of Mad in America dives more deeply into the assumptions underlying heritability research and problems with them:
Flamingos provide an example from nature of the fallacy of partitioning genetic and environmental factors into separate additive influences. Flamingos become pink by ingesting a diet of shrimp and other foods rich in alpha and beta carotenoid pigments. Those whose diet does not include carotenoid pigments do not become pink. Flamingos are therefore born with a genetic potential to have pink feathers, but require environmental influences to achieve this potential.
Rutter cited flamingos as an example of gene-environment interaction.22 He noted that both genes and environment play a crucial role in the ability of flamingos to turn pink, and that “you could feed seagulls for ever on the same diet and they would never turn pink.” He concluded that “it would make no sense to say that flamingos’ color was 50 percent due to genes and 50 percent due to diet. It is 100 percent due to genes (which have to be present) and 100 percent due to the environmental diet (which has to be present).”
The Stanford Encyclopedia of Philosophy has a somewhat technical overview of the “missing heritability problem”:
The development of alternative methods of estimating heritability (e.g., GWAS and SNP heritability) has given rise to what is commonly referred to as the ‘missing heritability problem’, which has seen ample attention by behavioral geneticists and some attention by philosophers of science and biology. At face value, the missing heritability arises out of a numerical gap between traditional heritability estimates and SNP-based heritability estimates of the same trait. For example, traditional heritability estimates for IQ obtained using twin and family studies range between .5 and .7; while SNP-based heritability estimates of IQ are currently no greater than .25 (Plomin and von stumm 2018). Missing heritability is greatest among complex, behavioral traits.
Wikipedia also has one in less-technical text. Note that the standard or traditional genetics methods referred to in both these articles involve studies of closely related people, most rigorously twins.
This led to a dilemma. Standard genetics methods have long estimated large heritabilities such as 80% for traits such as height or intelligence, yet none of the genes had been found despite sample sizes that, while small, should have been able to detect variants of reasonable effect size such as 1 inch or 5 IQ points. If genes have such strong cumulative effects – where were they?
Alexander Young discusses challenges in current methods of estimating the effects of rare genes that can be lost in large statistical analyses like these:
While my methodological concerns about GREML-WGS might be answered through further analyses for a trait like height, my own work has shown that the GREML approach leads to substantial overestimation of heritability for traits like educational attainment . This is due to the influence of indirect genetic effects (‘genetic nurture’) from relatives , which are the effects of genetic variants in relatives (mostly siblings and parents) on an individual through their environment. Family data is required to adjust for indirect genetic effects from relatives. Therefore, solving the problem of missing heritability for traits like educational attainment will require large samples of families with WGS data.
Brian Palmer argues in Slate that this heritability may not be missing at all, because twin studies are “pretty much useless”:
Twin studies rest on two fundamental assumptions: 1) Monozygotic twins are genetically identical, and 2) the world treats monozygotic and dizygotic twins equivalently (the so-called “equal environments assumption”). The first is demonstrably and absolutely untrue, while the second has never been proven.
Jay Joseph digs into the equal environments assumption directly:
While conceding the point that MZs are treated more similarly and share more common experiences when growing up, psychiatric twin researchers and textbook authors much less often mention the fact that MZ pairs also experience much higher levels of identity confusion related to their twinship, attachment to each other, and emotional closeness than experienced by DZ pairs, which will (presumably) lead to greater MZ behavioral resemblance. For example, Kringlen performed a “global evaluation of twin-closeness” based on 117 Norwegian pairs, and found that 65% of the MZ pairs had an “extremely strong level of closeness,” which was true for only 17% of the same-sex DZ pairs. Fully 90% of Kringlen’s MZ pairs had experienced “identity confusion in childhood,” which was experienced by only 10% of the DZ pairs.9 These findings by Kringlen have never been cited or discussed in any psychiatric publication that I am aware of. Other researchers investigating the question of whether twins were “mixed up for each other as children,” experienced “extreme or strong closeness or interdependence in childhood,” were “hard for strangers to tell…apart,” were “never separated from [their] twin,” or were “closely attached” found that MZ twins answered yes to these questions much more often than did DZ twins.
Joseph is also highly critical of studies of “twins raised apart”, both in the assumption that these twins had significantly different environments and for the lack of transparency in MISTRA, the most recent U.S. study:
And yet, in practice, Segal, Bouchard and their colleagues decided against making this preliminary “important first step” determination, and based their conclusions on behavioral genetic “model-fitting” statistical analyses, which depend on the clearly false assumption that “all resemblance between reared apart relatives is because of genetic factors.”17 In Born Together—Reared Apart, Segal wrote that the MISTRA personality studies found “that personality similarity between relatives seems to come mostly from their shared genes,”18 a conclusion arrived at circularly on the basis of the MISTRA assumption that, as Segal wrote elsewhere in her book, “shared genes underlie similarity between relatives.”
Kevin Bird just released a paper on methodological weaknesses of prior work on comparing heritability of educational attainment across populations and reruns the analysis using proper tests. There’s also an accompanying Twitter thread that provides his results in less specialist language.
I use cutting edge and easily available genomic analyses to assess two major claims made by proponents of the genetic hypothesis: 1. that substantial genetic differences between Black and white populations exist, and which cause the observed gap in cognitive ability and academic achievement, and 2. that these genetic differences associated with cognitive ability are the result of divergent natural selection. Results presented here indicate that known biases from population structure, assortative mating, indirect genetic effects, gene-environment interplay, and derived allele frequency differences between African and non-African populations bias polygenic score analysis. I show, using techniques comparing the squared difference of within-family effect size polygenic scores to the difference of matched randomly generated polygenic scores, and comparing genetic differentiation at trait-associated SNPs to a matched set of control SNPs, that these biases likely produced false signals of polygenic selection in recent analyses, and that there are no signals of divergent polygenic selection between African and European populations.
I further show the predicted genetic contribution to the Black-white gap in IQ score across a range of heritability estimates was substantially smaller than observed phenotypic gaps, suggesting 80% or more of the IQ variance between Africans and Europeans is environmental in nature. These results are inconsistent with the genetic hypothesis of substantial genetic contribution to the Black-white IQ gap.
ETA: This letter from Michelle N. Meyer, Patrick Turley, and Daniel J. Benjamin is a nice, quick summary of what all this means for Murray’s argument:
Consider IQ. IQ is not a fixed attribute of individuals and can be affected — for better and worse — by the environment in myriad ways. For example, in a society where people of color are denied access to childhood enrichment programs or adequate nutrition, a polygenic score for IQ might reflect genetic variants associated with skin pigmentation. Relatedly, in a sexist society, variants on the X and Y chromosomes, which determine biological sex, might be related to a variety of socio-economic phenotypes. Such polygenic scores would indeed moderately predict the IQ of people on average, but — and this is key — much of that predictive power would simply reflect social choices, not innate or immutable biology. (Note that polygenic scores for IQ are actually poor predictors of any one individual’s phenotype, as Mr. Murray acknowledges.)
If nothing else, understanding what heritability is and isn’t will give you some insulation from the hype around these claims. It will help you understand what nonsense it is when “scientific” racists like Murray and the people who promote him suggest that disagreement with him is science denialism.