Burt's point that heritability estimates and polygenic scores are context- and population-dependent is well-taken and widely appreciated. However, it should not be overstated as implying that all genetic analyses are irremediably socially contingent, varying widely depending on period, culture, and context, thereby shunning any hope of identifying stable, meaningful genetic associations.
One can of course tell stories about education being something very different in a remote hunter–gather society or in the distant future, but this should not obscure the fact that the notion of educational achievement in the twenty-first century that is the current focus of genetic analysis is a well-defined and circumscribed concept that is essentially the same all over the world except for some extremely isolated cultures where schools don't exist. Even if it is true that the personality traits that were likely to attract a young woman to higher education in the 1870s and in the 2020s United States may differ to some extent, the cognitive traits (detailed further below) that would have been important for her to succeed at university in 1870 are very likely to be the same as those important in 2020, and they are also the same in the United States, in Saudi Arabia, or in Thailand, thus providing a stable basis for the genetic analysis of educational achievement. When some of these factors differ between countries or periods, this should not be cause for despair or rejection of genetic approaches, as the issue is perfectly empirically tractable: This should rather be welcomed as an opportunity to describe interesting gene–environment interactions.
Nevertheless, Burt's critique has the merit of highlighting the potential gaps between the social outcomes that are currently subjected to genetic analysis, and their cognitive basis. One should recall that social outcomes such as educational achievement or income have been genetically studied mainly because they were conveniently available in very large databases. In every genetics project, every participant answers one question about their highest obtained degree, regardless of the initial goal of the research. Thus, pooling across many projects has enabled researchers to gather the millions of participants required to compute reliable educational achievement polygenic scores (Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati and Young2022).
But to the cognitive scientist, this may seem a temporary distraction: These complex social outcomes are not phenotypes that are under direct natural selection and that should naturally be the focus of genetic analysis. The phenotypes of interest for genetic analysis are situated at the cognitive level, where stable traits can be defined and can be the target of selection. For educational achievement, these are specific cognitive abilities: Not just general intelligence (which is itself a complex emerging property; Ramus, Reference Ramus2017), but its underlying components: Verbal ability, abstract reasoning, working memory, and also more specific cognitive skills such as phonological awareness (which contributes to reading acquisition) or number sense. One should not forget the popular but ill-named “noncognitive skills” (Ramus, Reference Ramus2022) such as conscientiousness, self-control, intrinsic motivation, grit, which do explain part of the educational achievement variance and which are also genetically influenced (Demange et al., Reference Demange, Malanchini, Mallard, Biroli, Cox, Grotzinger and Nivard2021). These traits reliably underlie educational achievement regardless of time, culture, and gender of the learner, and there is every reason to think that they have a stable neural and genetic basis, which may be to a large extent similar in all populations.
Similarly, the answer to the question “have you ever had sex with someone of the same sex? Yes/No” has never been a valid phenotype for genetic analysis, but it is the one that was available for UK Biobank and 23andMe participants (Ganna et al., Reference Ganna, Verweij, Nivard, Maier, Wedow, Busch and Zietsch2019). These authors are of course well aware that the stable cognitive trait of interest is sexual orientation, that it is continuous (e.g., as on the Kinsey scale), and that its relationship with actual sexual behaviour is imperfect, subject to social norms, to opportunities, and to many life circumstances. Genome-wide research on the genetics of sexual orientation will have to wait until an appropriate scale is rated by a sufficiently large number of participants.
An additional difficulty that may be less widely appreciated is that the cognitive functions that are under genetic influence are latent, unobservable variables, that cannot simply be equated with performance in one behavioural test. This is because any test, no matter how elementary it seems, inevitably recruits several cognitive functions. For instance, even the simplest reaction time test involves not only processing speed but also vision (or audition, to perceive the signal), sustained attention, language skills (to understand instructions), and motor skills (to produce a response). Therefore, there never is a one-to-one mapping between cognitive functions and behavioural tests. Any cognitive function can only be inferred by triangulating across several behavioural tests involving it in different ways.
This implies that research into the genetics of cognitive functions is going to be much more difficult than running a genome-wide association study (GWAS) on an answer to a single question or on a single test score. It will require administering well-designed, comprehensive test batteries to very large populations.
The conclusion is that the critique of the genetics of complex social outcomes is partly well-founded, insofar as social outcomes sometimes have unsystematic relations with cognitive traits. But the correct conclusion is not to dismiss the entire field altogether. Rather, the implication of this critique is to redirect geneticists' attention to the stable cognitive phenotypes that are natural candidates for genetic analysis. Unfortunately, studying the genetics of specific cognitive functions will take greater efforts and a longer time until the necessary test results are collected in sufficiently large genotyped populations.
You have
Access
Burt's point that heritability estimates and polygenic scores are context- and population-dependent is well-taken and widely appreciated. However, it should not be overstated as implying that all genetic analyses are irremediably socially contingent, varying widely depending on period, culture, and context, thereby shunning any hope of identifying stable, meaningful genetic associations.
One can of course tell stories about education being something very different in a remote hunter–gather society or in the distant future, but this should not obscure the fact that the notion of educational achievement in the twenty-first century that is the current focus of genetic analysis is a well-defined and circumscribed concept that is essentially the same all over the world except for some extremely isolated cultures where schools don't exist. Even if it is true that the personality traits that were likely to attract a young woman to higher education in the 1870s and in the 2020s United States may differ to some extent, the cognitive traits (detailed further below) that would have been important for her to succeed at university in 1870 are very likely to be the same as those important in 2020, and they are also the same in the United States, in Saudi Arabia, or in Thailand, thus providing a stable basis for the genetic analysis of educational achievement. When some of these factors differ between countries or periods, this should not be cause for despair or rejection of genetic approaches, as the issue is perfectly empirically tractable: This should rather be welcomed as an opportunity to describe interesting gene–environment interactions.
Nevertheless, Burt's critique has the merit of highlighting the potential gaps between the social outcomes that are currently subjected to genetic analysis, and their cognitive basis. One should recall that social outcomes such as educational achievement or income have been genetically studied mainly because they were conveniently available in very large databases. In every genetics project, every participant answers one question about their highest obtained degree, regardless of the initial goal of the research. Thus, pooling across many projects has enabled researchers to gather the millions of participants required to compute reliable educational achievement polygenic scores (Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati and Young2022).
But to the cognitive scientist, this may seem a temporary distraction: These complex social outcomes are not phenotypes that are under direct natural selection and that should naturally be the focus of genetic analysis. The phenotypes of interest for genetic analysis are situated at the cognitive level, where stable traits can be defined and can be the target of selection. For educational achievement, these are specific cognitive abilities: Not just general intelligence (which is itself a complex emerging property; Ramus, Reference Ramus2017), but its underlying components: Verbal ability, abstract reasoning, working memory, and also more specific cognitive skills such as phonological awareness (which contributes to reading acquisition) or number sense. One should not forget the popular but ill-named “noncognitive skills” (Ramus, Reference Ramus2022) such as conscientiousness, self-control, intrinsic motivation, grit, which do explain part of the educational achievement variance and which are also genetically influenced (Demange et al., Reference Demange, Malanchini, Mallard, Biroli, Cox, Grotzinger and Nivard2021). These traits reliably underlie educational achievement regardless of time, culture, and gender of the learner, and there is every reason to think that they have a stable neural and genetic basis, which may be to a large extent similar in all populations.
Similarly, the answer to the question “have you ever had sex with someone of the same sex? Yes/No” has never been a valid phenotype for genetic analysis, but it is the one that was available for UK Biobank and 23andMe participants (Ganna et al., Reference Ganna, Verweij, Nivard, Maier, Wedow, Busch and Zietsch2019). These authors are of course well aware that the stable cognitive trait of interest is sexual orientation, that it is continuous (e.g., as on the Kinsey scale), and that its relationship with actual sexual behaviour is imperfect, subject to social norms, to opportunities, and to many life circumstances. Genome-wide research on the genetics of sexual orientation will have to wait until an appropriate scale is rated by a sufficiently large number of participants.
An additional difficulty that may be less widely appreciated is that the cognitive functions that are under genetic influence are latent, unobservable variables, that cannot simply be equated with performance in one behavioural test. This is because any test, no matter how elementary it seems, inevitably recruits several cognitive functions. For instance, even the simplest reaction time test involves not only processing speed but also vision (or audition, to perceive the signal), sustained attention, language skills (to understand instructions), and motor skills (to produce a response). Therefore, there never is a one-to-one mapping between cognitive functions and behavioural tests. Any cognitive function can only be inferred by triangulating across several behavioural tests involving it in different ways.
This implies that research into the genetics of cognitive functions is going to be much more difficult than running a genome-wide association study (GWAS) on an answer to a single question or on a single test score. It will require administering well-designed, comprehensive test batteries to very large populations.
The conclusion is that the critique of the genetics of complex social outcomes is partly well-founded, insofar as social outcomes sometimes have unsystematic relations with cognitive traits. But the correct conclusion is not to dismiss the entire field altogether. Rather, the implication of this critique is to redirect geneticists' attention to the stable cognitive phenotypes that are natural candidates for genetic analysis. Unfortunately, studying the genetics of specific cognitive functions will take greater efforts and a longer time until the necessary test results are collected in sufficiently large genotyped populations.
Financial support
This work has received support by grants ANR-17-EURE-0017 and ANR-10-IDEX-0001-02.
Competing interest
None.