2.2.1 Cultural diffusion

In human societies, mechanisms such as conformist-biased learning (learning from the majority or plurality at a rate above population frequency; Boyd & Richerson, Reference Boyd and Richerson1985; Cavalli-Sforza & Feldman, Reference Cavalli-Sforza and Feldman1981; Muthukrishna, Morgan, & Henrich, Reference Muthukrishna, Morgan and Henrich2016), payoff biases (e.g., learning from successful others; Boyd & Richerson, Reference Boyd and Richerson1985; Cavalli-Sforza & Feldman, Reference Cavalli-Sforza and Feldman1981), and norm enforcement (identifying norms and punishing norm violators; Chudek & Henrich, Reference Chudek and Henrich2011) result in particular behaviors, beliefs, and norms disseminating widely across a society. To the degree that these cultural traits mask the effects of genes (such as in the skin cancer and vitamin D examples), heritability is reduced. In contrast, to the degree to which these cultural traits unmask or interact positively with genes without masking, heritability is increased.

One example that can illustrate both masking and unmasking through diffusion is genes that support the perception of lexical tone in tonal languages such as Cantonese and Yoruba (Dediu & Ladd, Reference Dediu and Ladd2007; Wong et al., Reference Wong, Kang, Wong, So, Choy and Geng2020). To the degree that tonal languages such as Cantonese or Yoruba diffuse in the population (and holding constant other genetic contributions), heritability of language ability would increase in a manner proportional to variation in these genes; to the degree that non-tonal languages such as Norwegian or Russian diffuse in the same population, heritability of language ability would decrease. As another example: It is known that fertility is predicted by genes (Zietsch, Kuja-Halkola, Walum, & Verweij, Reference Zietsch, Kuja-Halkola, Walum and Verweij2014), but Briley, Paige Harden, and Tucker-Drob (Reference Briley, Paige Harden and Tucker-Drob2015) report a large rise in the heritability of fertility in the United States over the mid-twentieth century, likely because of the increase in the variety of acceptable reproductive choices that was brought about by the diffusion of new social values. In this case, an increase in cultural variance unmasked the effect of genes associated with reproductive behaviors and preferences. Conversely, cultural or policy changes that reduce variation in reproductive practices – such as rigid childbearing norms or a one-child policy – would be expected to mask the genetic effect.

The diffusion of cultural traits is not random and is, in fact, well studied within cultural evolution and elsewhere (Henrich, Reference Henrich2001; Muthukrishna & Henrich, Reference Muthukrishna and Henrich2016; Rogers, Reference Rogers2003). One mechanism for cultural diffusion that is important in industrialized societies is formal education. We use this example, because the effect of this particular cultural institution on heritability has been studied. Samuelsson et al. (Reference Samuelsson, Byrne, Olson, Hulslander, Wadsworth, Corley and Willcutt2008) measured the heritability of reading and spelling test scores. Australian twins demonstrated a narrow-sense heritability of 0.84 in kindergarten and a similar score of 0.80 in Grade 1. In contrast, Swedish and Norwegian twins demonstrated a heritability of only 0.33 in kindergarten, rising to 0.79 in Grade 1. Heritability was at the same level in both the Australian and Scandinavian children in Grade 1, but not in kindergarten. Why? Cultural diffusion of literacy. Australian children begin receiving compulsory literacy instruction in kindergarten, while in Scandinavia the kindergarten curriculum emphasizes social, emotional, and esthetic development – literacy instruction only begins in Grade 1. Here, we see the effect of national curriculum policy differences affecting cultural diffusion of literacy and thus heritability, in a case where the cultural trait interacts with genes. Australian kindergarteners are exposed to standardized environmental input and much of the remaining variation in reading ability is explained by genetic differences, whereas for the Swedish and Norwegian kindergarteners, variation in the amount of reading instruction received at home is much larger than any genetic differences. In line with this interpretation, Samuelsson et al. (Reference Samuelsson, Byrne, Olson, Hulslander, Wadsworth, Corley and Willcutt2008) show that the boost in heritability among the Scandinavian children was also accompanied by an almost equivalent decrease in phenotypic variance attributed to the common (home) environment, which would include home instruction.

If we were to assess the genetic basis of literacy skill in schoolchildren without accounting for the impact of their particular educational curricula on cultural diffusion and environmental variation, we would be subjecting ourselves to a selection bias, with no idea of the magnitude of this bias. This would distort our understanding of the generalizability of our finding to samples that have undergone different educational curricula, and even more so to those with different levels of educational attainment. Note that even the literacy instruction provided in the home environment is already shaped by cultural evolution, both in terms of the content being transmitted (reading and spelling), and the structures that are transmitting (family organization in Western countries; Henrich, Reference Henrich2020; Schulz et al., Reference Schulz, Bahrami-Rad, Beauchamp and Henrich2019). In societies that produce literate children, culture impacts the heritability of literacy from the moment that variation in this trait emerges in development, virtually sealing off the possibility of assessing “baseline” heritability without cultural interference, even at the very start of life (e.g., children born in literate societies are typically surrounded by writing and literate adults). Heritability is a composite measure that captures both genetic and cultural effects, and without knowledge of the cultural context, it is difficult or impossible to judge what is being measured. When we say that the heritability of reading among Scandinavian children jumps up to 0.79 when they enter Grade 1, this measurement reveals just as much if not more about the disseminative power of modern schooling than it does about the genetic basis of literacy.

Looking toward a broader social context, several insightful studies have shown that the heritability of educational attainment increases with equality of opportunity. We see evidence for this within countries over time (Heath et al., Reference Heath, Berg, Eaves, Solaas, Corey, Sundet and Magnus1985), within and between countries over time (Engzell & Tropf, Reference Engzell and Tropf2019; de Zeeuw, de Geus, & Boomsma, Reference de Zeeuw, de Geus and Boomsma2015), and within countries following a major policy change (Colodro-Conde, Rijsdijk, Tornero-Gómez, Sánchez-Romera, & Ordoñana, Reference Colodro-Conde, Rijsdijk, Tornero-Gómez, Sánchez-Romera and Ordoñana2015; Rimfeld et al., Reference Rimfeld, Krapohl, Trzaskowski, Coleman, Selzam, Dale and Esko2018; Ujma et al., Reference Ujma, Eszlári, Millinghoffer, Bruncsics, Török, Petschner, Antal, Deakin, Breen, Bagdy and Juhász2022). Although there is also some evidence going in the other direction (Silventoinen et al., Reference Silventoinen, Jelenkovic, Sund, Latvala, Honda, Inui and Tomizawa2020), the overall pattern appears to be that diffusion of educational opportunity results in genes explaining a relatively greater proportion of variance in educational attainment. These studies serve as robust demonstrations of how heritability can be boosted by changes in social context broadly, and diffusion of particular forms of environmental exposure more specifically. Although these patterns are broadly consistent with our thesis, both the consistent and inconsistent results may obscure cultural structuring, such as the hidden cluster problem and cultural Simpson's paradox, as we discuss in section 3.

2.2.2 Cultural innovation

The human capacity for cumulative cultural evolution (Dean et al., Reference Dean, Vale, Laland, Flynn and Kendal2014; Henrich, Reference Henrich2004a) ensures that in our societies, the diffusion of extant cultural traits goes hand-in-hand with the continual generation of new cultural traits. Whether emerging through serendipitous discovery, iterated trial-and-error, or recombination of ideas, some portion of new technologies and behaviors manages to spread across a society, displacing other competing variants. The dynamics of innovation are a significant component of the cultural evolutionary framework (Henrich, Reference Henrich2004a; Muthukrishna & Henrich, Reference Muthukrishna and Henrich2016).

Cultural evolution predicts that societies will vary in their rate and type of cultural innovation (e.g., number of patents and incremental vs. revolutionary invention) because of differences in, for example, the size and interconnectedness of their social networks (sociality), effectiveness of cultural transmission (e.g., education), and tolerance for diversity; for a review see Muthukrishna and Henrich (Reference Muthukrishna and Henrich2016). Tolerance for diversity can drive differences in heterogeneity. Some societies such as Pakistan and Indonesia maintain relatively low levels of cultural heterogeneity, whereas others such as Brazil and Australia maintain relatively high levels of heterogeneity. One metric that can serve as a proxy for this tolerance is cultural tightness/looseness, which represents the degree to which societies tolerate deviation from social norms, and is expected to correspond to their rate of innovation (Gelfand, Reference Gelfand2018; Gelfand, Nishii, & Raver, Reference Gelfand, Nishii and Raver2006; Gelfand et al., Reference Gelfand, Raver, Nishii, Leslie, Lun, Lim and Duan2011). Societies that are more tolerant of cultural variation allow for more individual-level exploration, and if the best of these cultural mutants can be selectively incorporated into the mainstream, those societies will tend to undergo higher rates of increase in cultural complexity (Henrich, Reference Henrich2004a). Cultural variation is the engine of cultural change. Aside from variables such as cultural looseness, policies such as social safety nets and forgiving bankruptcy laws can also create the incentive structure for promoting exploration and innovation (Muthukrishna & Henrich, Reference Muthukrishna and Henrich2016).

The initial spread of an innovation will be tied to an increase in environmental variance if that innovation is disrupting an earlier, relatively homogeneous state. To the degree that innovation is masking genes, heritability will begin to decrease. To the degree that innovation is unmasking genes, heritability will begin to increase.

2.2.3 Predicting differences in heritability across societies

Over long historical durations, cultural evolution tends toward a compressive environmental effect, as ecological problems are solved and more favorable environmental conditions spread. But on shorter timescales, diffusion and innovation have opposing effects on environmental variance. Although indoor plumbing, antibiotics, and formal education have diffused to the point of being ubiquitous in the Western world, new innovations are also constantly emerging. During their early spread, such innovations increase environmental variance and between-group differences, thereby bringing forth new forms of diversity and inequality. But, as these new traits diffuse further, some of them become gradually established within the cultural corpus, reducing environmental variation through homogenization. The dynamics of diffusion and innovation thus differ in their effects, although the magnitude and direction of these effects also depend upon whether the trait in question masks or unmasks genetic variation. Moreover, across societies, the balance between diffusion and innovation differs as well.

If we imagine a society in which innovations are rapidly diffused across the whole society shortly after they emerge, the society will have a low level of cultural variation at most times. This is a society in which a high rate of diffusion quickly overpowers the environmental variance-increasing potential of cultural innovation. Now imagine a contrasting society in which innovations take a much longer time to diffuse. Perhaps, the society is more culturally clustered with less interaction between individuals in different subgroups, or perhaps there is a subset of individuals who are well-connected with the inventor and become early adopters but the novel trait diffuses only gradually beyond this inner circle. This would be a culturally unequal society that is characterized by high environmental variability at any given point in time. This kind of stunted diffusion may suggest cultural clustering, with relatively dense connections within subgroups and relatively sparse connections between subgroups. In such societies, even highly useful forms of cultural knowledge may not easily permeate social barriers, and the waves of environmental change are correspondingly “loose.” These barriers are not necessarily ethnic boundaries: They could be organized around class, wealth, occupation, political alignment, religion, or incidental geographic layout. A society may be clustered for reasons that stem from within the society itself, or it could simply be a matter of nominal mismatch between the political boundaries that we use to refer to a society and the actual organization of cultural groups. Greater differential clustering can lead to a cultural Simpson's paradox (discussed in sect. 3.4).

We can derive testable hypotheses from these dynamics: Culturally homogeneous societies will have higher heritability for culturally transmissible phenotypic traitsFootnote ⁶ compared to culturally diverse or clustered societies. We model this prediction in the Appendix. One way to measure cultural diversity would be to use cultural tightness/looseness as a proxy (Gelfand et al., Reference Gelfand, Nishii and Raver2006, Reference Gelfand, Raver, Nishii, Leslie, Lun, Lim and Duan2011). One way to measure cultural clustering would be to use the cultural fixation index (CF _ST) as a measure of cultural distance between groups within a society (Muthukrishna et al., Reference Muthukrishna, Bell, Henrich, Curtin, Gedranovich, McInerney and Thue2020) – we discuss this in more detail in section 3.2. Whatever the measure, homogeneity of cultural traits should be associated with higher heritability on average, but with respect to specific traits, the relationship between homogeneity and heritability will also depend upon the genetic masking or unmasking effect of the trait.

For cultural traits that neither mask nor unmask genetic effects, heritability will generally be higher in culturally homogeneous societies than in culturally diverse societies because of the reduced environmental variance that is a consequence of behavioral uniformity, as described above. For cultural traits that unmask genes, heritability will be higher overall compared to traits that neither mask nor unmask, and on average higher in homogeneous societies than in diverse societies. For cultural traits that mask genes, heritability will be lower overall compared to traits that neither mask nor unmask, but whether it is higher in homogeneous or diverse societies will depend upon which has the stronger effect: genetic masking or the reduction in environmental variance accompanying homogeneity. If the effect of genetic masking is stronger, heritability will be lower in culturally homogeneous societies than that in culturally diverse societies; if the reduction in environmental variance is stronger, heritability will be higher in homogeneous societies than that in diverse societies.

Thus, heritability estimates reflect cultural evolutionary dynamics. Researchers such as Harden (Reference Harden2021) have highlighted how heritability can be used as a tool for measuring useful characteristics across environments such as cross-national differences in social opportunity. These interpretations are consilient with our approach, but are limited by treating the environment as exogenous rather than modeling its dynamics and interactions. Such complexities in the role of the cultural environment have historically been obscured, because, in part, of a number of methodological and epistemological problems that encumber standard behavioral genetic analysis.

3.1.1. WEIRD genetics

A comprehensive meta-analysis that claims to contain essentially all twin studies published between 1958 and 2012 (Polderman et al., Reference Polderman, Benyamin, de Leeuw, Sullivan, van Bochoven, Visscher and Posthuma2015) reveals that 94% of sampled twin pairs were from Western populations. The United States, United Kingdom, and Australia alone accounted for almost 60%, and Nordic countries accounted for another 25%. Of the non-Western countries (6%), two-thirds (4%) are from northeast Asia – specifically, China, Japan, South Korea, and Taiwan, countries that are not Western, but have most of the remaining letters of the WEIRD acronym. The remainder of the world, representing the vast majority of the human population, accounts for only 2% of the dataset.

GWASs too suffer from a myopic focus on WEIRD genomes (Need & Goldstein, Reference Need and Goldstein2009; Popejoy & Fullerton, Reference Popejoy and Fullerton2016; Sirugo, Williams, & Tishkoff, Reference Sirugo, Williams and Tishkoff2019). As of 2017, 88% of samples in GWASs were of European ancestry (Mills & Rahal, Reference Mills and Rahal2019).Footnote ⁷ Paralleling the twin studies data, 72% of participants were recruited from just three countries – United States, United Kingdom, and Iceland – with nearly 20% of the remainder being recruited from Japan, China, and South Korea.

Polygenic scores do not translate well across ancestry groups (Bitarello & Mathieson, Reference Bitarello and Mathieson2020; Curtis, Reference Curtis2018; Guo et al., Reference Guo, Bakshi, Wang, Jiang, Yengo, Goddard and Visscher2021; Kim, Patel, Teng, Berens, & Lachance, Reference Kim, Patel, Teng, Berens and Lachance2018; Martin et al., Reference Martin, Gignoux, Walters, Wojcik, Neale, Gravel and Daly2017, Reference Martin, Kanai, Kamatani, Okada, Neale and Daly2019). For example, European ancestry-derived polygenic scores have only 42% of the effect size in African ancestry samples (Duncan et al., Reference Duncan, Shen, Gelaye, Meijsen, Ressler, Feldman and Peterson2019). From a cultural evolutionary perspective, this is not unexpected given the cultural environment, coevolution between culture and genes, and cultural differences between populations.

Polygenic scores are also highly sensitive to inadequately controlled population stratification (Berg et al., Reference Berg, Harpak, Sinnott-Armstrong, Joergensen, Mostafavi, Field and Boyle2019; Morris, Davies, Hemani, & Smith, Reference Morris, Davies, Hemani and Smith2020; Sohail et al., Reference Sohail, Maier, Ganna, Bloemendal, Martin, Turchin and Chiang2019). Even within a single ancestry group, the predictive accuracy of polygenic scores can be dependent on age, sex, and socioeconomic status (Mostafavi et al., Reference Mostafavi, Harpak, Agarwal, Conley, Pritchard and Przeworski2020) – this too, from a cultural evolutionary perspective, is not unexpected given the cultural variation that exists within a population (Muthukrishna & Henrich, Reference Muthukrishna and Henrich2019; Muthukrishna et al., Reference Muthukrishna, Bell, Henrich, Curtin, Gedranovich, McInerney and Thue2020). Similarly, the SNPs that contribute to the variance of a trait are different in different populations (Akiyama et al., Reference Akiyama, Ishigaki, Sakaue, Momozawa, Horikoshi, Hirata and Matsuda2019; Gurdasani et al., Reference Gurdasani, Carstensen, Fatumo, Chen, Franklin, Prado-Martinez and Bouman2019; Pemberton et al., Reference Pemberton, Verdu, Becker, Willer, Hewlett, Le Bomin and Froment2018; Rotimi et al., Reference Rotimi, Bentley, Doumatey, Chen, Shriner and Adeyemo2017) and it is difficult to disentangle the genetic, environmental, and cultural contribution to differing polygenic scores between populations (Rosenberg, Edge, Pritchard, & Feldman, Reference Rosenberg, Edge, Pritchard and Feldman2019). Recent projects have aimed to capture a greater degree of global human genetic diversity (e.g., Simons Genome Diversity Project, Mallick et al., Reference Mallick, Li, Lipson, Mathieson, Gymrek, Racimo and Zhao2016; the exome analysis of Lek et al., Reference Lek, Karczewski, Minikel, Samocha, Banks, Fennell and O'Donnell-Luria2016; and the GenomeAsia project, Wall et al., Reference Wall, Stawiski, Ratan, Kim, Kim, Gupta and Suryamohan2019), but we are far from proportionately representing the genetic diversity of the global population.

3.1.2. WEIRD culture

When we restrict the scope of genetic samples, the cultural environment against which genetic effects are evaluated also becomes skewed, and this greatly reduces the interpretability of genetic effects. Because of a combination of cultural-group dynamics (Chudek & Henrich, Reference Chudek and Henrich2011; Henrich, Reference Henrich2004b; Richerson et al., Reference Richerson, Baldini, Bell, Demps, Frost, Hillis and Mathew2016) and cumulative cultural evolution (Dean et al., Reference Dean, Vale, Laland, Flynn and Kendal2014; Henrich, Reference Henrich2004a), the human species is characterized by large amounts of cultural and hence environmental variation between societies, which exceeds genetic variation by orders of magnitude (Bell, Richerson, & McElreath, Reference Bell, Richerson and McElreath2009). The WEIRD countries that are over-represented in genetic samples are clustered together along multiple cultural dimensions (Hofstede, Reference Hofstede2001; Inglehart & Welzel, Reference Inglehart and Welzel2005; Muthukrishna et al., Reference Muthukrishna, Bell, Henrich, Curtin, Gedranovich, McInerney and Thue2020), and are perhaps an extreme unrepresentative outlier on many psychological and behavioral measures, with these countries registering the highest scores for traits such as individualism, analytical thinking, and prosociality toward strangers, and the lowest scores on opposite constructs such as collectivism, holistic thinking, and prosociality toward relatives but not strangers (Henrich, Reference Henrich2020; Henrich et al., Reference Henrich, Heine and Norenzayan2010; Muthukrishna et al., Reference Muthukrishna, Bell, Henrich, Curtin, Gedranovich, McInerney and Thue2020; Schulz et al., Reference Schulz, Bahrami-Rad, Beauchamp and Henrich2019).

WEIRD societies are also roughly aligned on basic environmental factors such as technology, civil infrastructure, healthcare, schooling, and transportation, with typically advanced levels in each of these. Within each of these countries as well, inequality is low compared to non-WEIRD countries, as can be inferred from the comparatively low Gini coefficients of Western countries, with the United States being somewhat of an unequal outlier (Selita & Kovas, Reference Selita and Kovas2019).

Just as this restricted cultural range limits our understanding of human psychology (Henrich et al., Reference Henrich, Heine and Norenzayan2010), it also limits the inferences that can be made about human genetics. Behavioral geneticists understand that genes can have different effects under different environmental conditions (Assary, Vincent, Keers, & Pluess, Reference Assary, Vincent, Keers and Pluess2018; Moffitt, Caspi, & Rutter, Reference Moffitt, Caspi and Rutter2005), but there is less appreciation of how those environmental conditions and interactions change through cultural evolution. Cultural evolution can mask or unmask genetic effects, such that gene function becomes confounded with the effect of the cultural environment in a manner that is dependent on the specifics of the underlying dynamics. Cutting through this confound requires a theoretical understanding of the cultural dynamics as well as empirical data on genetic effects across a wide variety of cultural environments. The WEIRD sampling problem, therefore, harms accurate inference of genetic effects.

3.2.1 Cultural clustering

Whether the method is a twin study or GWAS, behavioral genetic studies typically draw their data from databases such as twin registries or biobanks, whose coverage spans some circumscribed geographic range such as a country or a subnational region. Because these samples trace over political or administrative boundaries, they may not match up with the actual structure of environmental variation, which will often be dependent upon the organization of cultural clustering. Culture, generally, tends to agglomerate, such that groups of individuals who are bound together by dense links of cultural influence share a substantially greater number of features of their cultural environment than do individuals who belong to different groups (Richerson et al., Reference Richerson, Baldini, Bell, Demps, Frost, Hillis and Mathew2016). As has been discussed extensively in the cultural evolutionary literature, this in-group similarity is generated and maintained by processes such as conformist learning (Boyd & Richerson, Reference Boyd and Richerson1985), norm enforcement (Chudek & Henrich, Reference Chudek and Henrich2011; Fehr & Fischbacher, Reference Fehr and Fischbacher2004), symbolic markers of in-group membership (Boyd & Richerson, Reference Boyd and Richerson1987), and cultural-group selection (Henrich, Reference Henrich2004b; Richerson et al., Reference Richerson, Baldini, Bell, Demps, Frost, Hillis and Mathew2016). The structure of environmental variation is, thus, shaped by networks of cultural interaction, and the topology of these networks is assembled over time by cultural-group dynamics and cultural evolution.

A highly clustered population is one that can be easily partitioned into multiple subgroups whose members share among each other cultural traits that are substantially different from those found in other subgroups. When we know the underlying social network, network clustering algorithms (Emmons, Kobourov, Gallant, & Börner, Reference Emmons, Kobourov, Gallant and Börner2016) can help identify these clusters. When we have measures of a range of cultural traits, we can see the effects of clustered cultural influence by measuring the cultural distance between subgroups, for example by using CF _ST (Muthukrishna et al., Reference Muthukrishna, Bell, Henrich, Curtin, Gedranovich, McInerney and Thue2020). CF _ST applies the “fixation index” (Cavalli-Sforza, Menozzi, & Piazza, Reference Cavalli-Sforza, Menozzi and Piazza1994) to the World Values Survey of cultural beliefs and behaviors (Inglehart et al., Reference Inglehart, Haerpfer, Moreno, Welzel, Kizilova, Diez-Medrano and Lagos2014), and can quantify the cultural differentiation between any two groups. When applied to subgroups within societies, this measure can tell us, for example, the degree to which Singaporean Catholics differ from Singaporean Protestants, thus identifying the degree to which a nation state or other political grouping contains large clustering (note that this is different than diversity – a country may be diverse, but uniformly so, showing no large clustering).

Cultural clusters are typically created by barriers that impede cultural interaction, such as topography (e.g., mountain ranges or bodies of water separating populations) or cultural conflict (e.g., conflicting religious beliefs). Mutually unintelligible languages are an example of a social barrier that impedes cultural and even genetic mixing. Thus, the density of languages within a population will also predict clustering. A topographically fractured country like Papua New Guinea with its 839 often mutually unintelligible languages is likely to be a much more clustered population than an equally populated but less multilingual society like Austria. Populations can also become clustered because of more extrinsic factors. Countries that grew through recent immigration, such as Canada, are likely to have more clustering than countries that are characterized by relative cultural homogeneity, such as Japan (Fearon, Reference Fearon2003). Countries whose borders are drawn arbitrarily with respect to the geographic arrangement of cultural groups, for example as a product of past colonial administration (many countries in Africa), are also likely to have high clustering (Michalopoulos & Papaioannou, Reference Michalopoulos and Papaioannou2020). Note the parallel between the present argument and assortative mating (Schwartz, Reference Schwartz2013): whereas assortative mating results in phenotypic clustering because of reproduction, cultural transmission results in phenotypic clustering because of segregated learning.

Moreover, cultural clusters may cut through divisions of society that we don't always think of as being “cultural,” such as stratification by social class or wealth. The strength of this stratification regulates the amount of within-stratum versus between-stratum cultural interaction. Although this picture may be complicated by factors such as prestige bias engendering asymmetric influence from upper to lower classes (Henrich & Gil-White, Reference Henrich and Gil-White2001), such forms of social hierarchy can create clusters. These examples show that cultural clustering is not necessarily bound by geographic contiguity. Although spatial proximity is a good predictor of cultural influence, cultural influence may also be decoupled from geography, particularly with modern forms of mass media and information technology (Anderson, Reference Anderson2006; Martin & Yurukoglu, Reference Martin and Yurukoglu2017). In section 4.1, we discuss in more detail how cultural clustering may contribute to the phenomenon of differential heritability across socioeconomic levels.

In sum, the scope of a genetic sample is often defined by political and administrative boundaries, but these boundaries may be mismatched to the organization of cultural clusters within a society, and the extent of this mismatch will vary greatly across populations. So, in addition to measuring cultural variation (diversity) using indices such as cultural looseness, we need to also measure cultural clustering (“segregated diversity”) to fully understand the environment. Once we are able to do this, the next intellectual step would be to achieve an understanding of the superimposition of cultural and genetic clusters as well as their interaction.

3.2.2 Contrasts between cultural and genetic clustering

Behavioral geneticists have devoted much effort to the study of genotypic clustering, usually referred to as population stratification or population structure. Populations can become stratified in the presence of genetic variation arising from systematic differences in subpopulation ancestry or from other forms of structured, non-random mating (Brumpton et al., Reference Brumpton, Sanderson, Hartwig, Harrison, Vie, Cho and Howe2020). This kind of genetic clustering can introduce spurious associations between genotypes and traits. As such, researchers continue to develop an array of methods for dealing with this problem (Hellwege et al., Reference Hellwege, Keaton, Giri, Gao, Velez Edwards and Edwards2017; Price, Zaitlen, Reich, & Patterson, Reference Price, Zaitlen, Reich and Patterson2010).

Genetic clustering and cultural clustering are linked – a well-known example is the close alignment of genes and languages in historical populations (Cavalli-Sforza, Reference Cavalli-Sforza2001; Cavalli-Sforza, Piazza, Menozzi, & Mountain, Reference Cavalli-Sforza, Piazza, Menozzi and Mountain1988; Pagani et al., Reference Pagani, Kivisild, Tarekegn, Ekong, Plaster, Romero and Ayub2012; Tishkoff et al., Reference Tishkoff, Reed, Friedlaender, Ehret, Ranciaro, Froment and Hirbo2009). This gene–language alignment occurs because languages, similar to genes, have traditionally been transmitted vertically – that is, from parents to children or within the family. This is because of critical periods for language learning that close as early as 6 months for phonology (Kuhl, Williams, Lacerda, Stevens, & Lindblom, Reference Kuhl, Williams, Lacerda, Stevens and Lindblom1992) and year 7 for grammar (Johnson & Newport, Reference Johnson and Newport1989), because of plasticity in the relevant cortical networks being progressively staunched by “molecular brakes” (Werker & Hensch, Reference Werker and Hensch2015). Early learning of this kind is primarily dispensed by parents or other immediate caretakers (Hewlett, Fouts, Boyette, & Hewlett, Reference Hewlett, Fouts, Boyette and Hewlett2011; Kline, Boyd, & Henrich, Reference Kline, Boyd and Henrich2013), and it is unsurprising that genes and languages should be largely aligned in reconstructions of population history.Footnote ⁸

However, beyond early childhood, children become increasingly exposed to other children from different families as well as to non-kin adults, and the range of sources for cultural learning widens accordingly: a two-stage model of cultural learning (Kline et al., Reference Kline, Boyd and Henrich2013). Because of transmission through these horizontal (i.e., age-peers) and oblique (i.e., non-parent adults) channels, cultural learning becomes untethered from strict vertical transmission and extends laterally across genetic lineages, similar in structure to horizontal gene transfer that is prevalent in bacteria and archaea (Soucy, Huang, & Gogarten, Reference Soucy, Huang and Gogarten2015). Similar to horizontal gene transfer, horizontal cultural transmission is rapid, because of it being unconstrained by the generational cycle of vertical transmission. Oblique and horizontal transmission play a large role in cultural diffusion both within and across extant small-scale societies (Henrich & Henrich, Reference Henrich and Henrich2010; Hewlett et al., Reference Hewlett, Fouts, Boyette and Hewlett2011; Kline et al., Reference Kline, Boyd and Henrich2013), just as they do in large-scale industrialized societies where innovation, knowledge, and practices across domains such as technology, art, education, political systems, and supernatural beliefs commonly diffuse horizontally (Rogers, Reference Rogers2003). Indeed, with only vertical transmission, culturally evolution as it is currently understood would not work (Enquist et al., Reference Enquist, Strimling, Eriksson, Laland and Sjostrand2010). The importance of oblique and horizontal transmission in human cultural evolution may have been amplified by adaptations such as the timing of weaning relative to brain maturation, which is substantially earlier in our species than would be expected on the basis of our close primate relatives, thereby expediting the human child's exposure to cultural models other than parents (Finlay & Uchiyama, Reference Finlay, Uchiyama and Kaas2020; Hawkes & Finlay, Reference Hawkes and Finlay2018).

Because, in part, of horizontal transmission, the diffusion of culture can occur much more rapidly than the diffusion of genes, or “demic” diffusion. In historical populations, the speed of demic diffusion has been limited by rates of reproduction and migration (Ammerman & Cavalli-Sforza, Reference Ammerman and Cavalli-Sforza1984; Fort, Reference Fort2012). Horizontal transmission is the diffusion of ideas rather than people and has no such limits; cultural clusters are, therefore, more fluid and malleable than genetic clusters. For example, if archeological data show cultural commonality between two contemporaneous populations but genomic analyses suggest disjunction, we can often infer that the shared cultural traits between the two groups were because of cultural transmission rather than migration or admixture (e.g., Fu et al., Reference Fu, Posth, Hajdinjak, Petr, Mallick, Fernandes and Furtwängler2016; Olalde et al., Reference Olalde, Brace, Allentoft, Armit, Kristiansen, Booth and Rohland2018).

Cultural clusters are not only more fluid than genetic clusters, but they also explain intergroup differences better than genes do. Cultural distances between neighboring countries are an order of magnitude larger than genetic distances (Bell et al., Reference Bell, Richerson and McElreath2009). Among small-scale societies in Kenya, pastoral clans are differentiated by cultural traits considerably more than they are by genes, and cooperation among these clans is predicted by their cultural – but not geographic – distance to each other (Handley & Mathew, Reference Handley and Mathew2020). These findings are consistent with the predictions of cultural-group selection (Henrich, Reference Henrich2004b; Richerson et al., Reference Richerson, Baldini, Bell, Demps, Frost, Hillis and Mathew2016), particularly in how competition between cultural trait-groups helps explain the evolution of human cooperation (Henrich & Muthukrishna, Reference Henrich and Muthukrishna2021). Cultural-group selection requires large differences between groups of cultural traits relative to differences within groups (Apicella, Marlowe, Fowler, & Christakis, Reference Apicella, Marlowe, Fowler and Christakis2012; Fehr & Fischbacher, Reference Fehr and Fischbacher2003).

Reconciliation between cultural evolution and behavioral genetics requires an update in the way we think about culture. An ethnolinguistic conception of culture that revolves around vertical transmission roughly in alignment with genes is insufficient. For example, linguistic, ethnic, religious, or caste boundaries can create genetic population structure through endogamous marriage practices, whereby individuals marry others who are culturally similar. In the modern world, such assortative mating may be exaggerated as people become more mobile and better connected, making it easier to mate with partners who share cultural traits across many dimensions. Efficiency of assortment may be boosted by cultural matching through dating apps, stretching the tails of the genetic distribution. These are all examples of human culture but they represent only one particular aspect of it, namely cultural effects that channel mating and thereby regulate genetic clusters. This aspect of culture is what appears as signal in population genetic data and is thus often the focus of attention for geneticists insofar as culture is concerned. This ethnolinguistic conception of culture that reduces culture to its effects on genetic clustering is inadequate for comparing genetic effects to environmental effects, given the many non-reproductive cultural effects that shape human phenotypes and environments. Indeed, an understanding of cultural clustering may elucidate novel aspects of genetic population stratification.

Abdellaoui and colleagues (Abdellaoui, Verweij, & Nivard, Reference Abdellaoui, Verweij and Nivard2021; Abdellaoui et al., Reference Abdellaoui, Hugh-Jones, Yengo, Kemper, Nivard, Veul and Holtz2019) offer evidence for geographic clustering of polygenic scores in the United Kingdom that is statistically independent of genetic ancestry, and explained instead by recent migration. For example, coal mining regions experienced “brain drain” and other forms of trait-conditional departure because of changes in the UK economy over the twentieth century. The outcome is detectable in genetic assortment at both the source and destination regions. This is an example of gene–environment correlation, where the environmental factor is a macroeconomic variable with a relatively clear geographic distribution. But this kind of genetic sorting is likely to occur also with respect to environmental factors that are shaped by cultural transmission yet not as readily localizable along geography or other conventional social scientific dimensions. Just as we can use genomic methods to reconstruct aspects of cultural history (e.g., Dai et al., Reference Dai, Vazifeh, Yeang, Tachet, Spencer Wells and Vilar2020; Peter, Petkova, & Novembre, Reference Peter, Petkova and Novembre2020; Petkova, Novembre, & Stephens, Reference Petkova, Novembre and Stephens2016), a research strategy that builds upon cultural transmission may help us reconstruct and even predict the emergence of genetic clusters that coalesce around – and perhaps feed back onto – cultural clustering of human behavior, psychology, and environments.

3.3.1 Genes that break and genes that make

The more complex a system, the more ways it can fail. Take the history of lighting: Compared to the two ways in which a wood-fueled fire can be extinguished (smothering and exhaustion of fuel), there are seven known failure modes for a fluorescent bulb and more than 30 for the newer LED bulb (de Groot, Vos, Vogels, & van Driel, Reference de Groot, Vos, Vogels, van Driel, van Driel and Fan2013). A faulty rubber O-ring caused the space shuttle Challenger to explode, and a severed fiber-optic cable knocked out internet access for a large swath of people across India and the Middle East. There is a fundamental asymmetry between the identification of elements that support a system and those that undermine it. A well-functioning system is the product of a design process that has solved many problems and closed many paths that do not work. For such a system, “something going wrong” can be caused by singular aberrations, but “going right” and “going well” are properties of the integrated system as a whole. The notions of success and failure in this context, therefore, point to very different things. For complex functional systems such as machines and organisms, it is easier to identify ways to break the system than ways to explain or improve it.

Gene function can be viewed through this lens. Organisms are the outcomes of complex, emergent interactions involving many genes and their surroundings (Davies, Reference Davies2014), but there are many ways these interactions can go wrong. It is easier to identify deleterious genetic mutations than beneficial mutations, as deleterious mutations are more common. The space of failure is larger than the space of success, making genes that break more detectable than genes that make. For example, a single mutation can cause Mendelian disorders such as cystic fibrosis and Huntington's disease, but no single mutation creates genius. Over 1,000 genes have been linked to intelligence (Davies et al., Reference Davies, Lam, Harris, Trampush, Luciano, David Hill and Hagenaars2018; Lee et al., Reference Lee, Wedow, Okbay, Kong, Maghzian, Zacher and Nguyen-Viet2018; Savage et al., Reference Savage, Jansen, Stringer, Watanabe, Bryois, de Leeuw and Nagel2018). Each gene only explains a miniscule fraction of variation in intelligence, and the causal mechanisms are unlikely to be straightforward (Gottlieb, Reference Gottlieb2003; Turkheimer, Reference Turkheimer2000). In contrast to these genes that make, the causal mechanisms behind single-gene mutations that cause intellectual disability – for example, BCL11A (Dias et al., Reference Dias, Estruch, Graham, McRae, Sawiak, Hurst and Joss2016), PHF8 (Bathelt, Astle, Barnes, Lucy Raymond, & Baker, Reference Bathelt, Astle, Barnes, Lucy Raymond and Baker2016), ZDHHC9 (Schirwani, Wakeling, Smith, Study, & Balasubramanian, Reference Schirwani, Wakeling, Smith, Study and Balasubramanian2018) – are relatively well understood.

This spectrum of localizability ranging from Mendelian to polygenic to “omnigenic” traits (Boyle, Li, & Pritchard, Reference Boyle, Li and Pritchard2017) has been discussed extensively, but its interaction with cumulative culture has not sufficiently been appreciated. We have known for a long time that increasing nutrition (Lynn, Reference Lynn1990; Stoch, Smythe, Moodie, & Bradshaw, Reference Stoch, Smythe, Moodie and Bradshaw1982), improving schooling (Ceci, Reference Ceci1991; Davis, Reference Davis2014; Ritchie & Tucker-Drob, Reference Ritchie and Tucker-Drob2018), and removing parasites (Jardim-Botelho et al., Reference Jardim-Botelho, Raff, de Ávila Rodrigues, Hoffman, Joseph Diemert, Corrêa-Oliveira and Michael Bethony2008) have positive effects on IQ. None of this is surprising, but it means that in a society where parasite infection is kept under control, we would not notice that parasite status correlates with intelligence, because of a lack of sufficient variation in parasite load. For the same reason, a correlation between lead exposure and IQ (Needleman & Gatsonis, Reference Needleman and Gatsonis1990; Wasserman et al., Reference Wasserman, Liu, Lolacono, Factor-Litvak, Morina, Musabegovic and Vrenezi1997) will not be revealed in a society where lead is not a problem. And by corollary, genes that provide protection against malnutrition, parasites, or pollution would only be positively associated with intelligence in environments where these insults occur. In environments where these insults have been removed, the same genes would not be associated with intelligence, and can even be deleterious, as in the well-known example of sickle cell trait (Elguero et al., Reference Elguero, Délicat-Loembet, Rougeron, Arnathau, Roche, Becquart and Gonzalez2015). Similarly, alleles that protect against parasite infection (Carter, Reference Carter2013) or lead poisoning (Onalaja & Claudio, Reference Onalaja and Claudio2000) will be predictive of IQ only if the environmental risk factors are present in sufficient quantities.

3.3.2. Cumulative culture masking genes

Our living conditions have not always been the way they are today. Just 200 years ago, 89% of humanity lived in extreme poverty (Ravallion, Reference Ravallion2016), 88% were illiterate (van Zanden, Baten, Mira d'Ercole, Rijpma, & Timmer, Reference van Zanden, Baten, Mira d'Ercole, Rijpma and Timmer2014), and 43% of children died before they were 5 years old (Gapminder, 2020). Conditions have rapidly improved: rates of extreme poverty are now 10%, illiteracy is down to 14%, and deaths before 5 years of age are now 4% (UNESCO Institute for Statistics, 2013; World Bank Group – International Development, Poverty, & Sustainability, 2020). Of course, even if most are now better off, our world still suffers from immense global inequality. Given this restriction in historical range combined with the WEIRD sampling problem, it is not clear how well the genetic effects that have been cataloged thus far can be generalized beyond the particular cultural and temporal contexts in which they were studied. And even if this generalizability issue is acknowledged in principle, the bounds on generalizability are unknown.

Genes can be functionally masked by cumulative cultural evolution, and we expect that this masking is extensive and systematic. Cultural masking may help explain the limited portability of polygenic scores across populations (Kim et al., Reference Kim, Patel, Teng, Berens and Lachance2018; Martin et al., Reference Martin, Kanai, Kamatani, Okada, Neale and Daly2019, Reference Martin, Gignoux, Walters, Wojcik, Neale, Gravel and Daly2017). There is nothing too mysterious about this phenomenon: It's what happens when there are multiple evolutionary systems operating within a shared space of biological function. To build upon an example discussed by Deacon (Reference Deacon, Weber and Depew2003): Vitamin C is an essential nutrient and its acquisition is thereby an essential biological function. Endogenous synthesis of vitamin C requires a gene called GLO, and GLO is present across most of the animal kingdom. But, because vitamin C synthesis is metabolically costly, the gene is inactive in some species that have access to sufficient quantities of the nutrient in their diets (Drouin, Godin, & Page, Reference Drouin, Godin and Page2011). These include taxa such as teleost fishes, guinea pigs, many bats, some passerine birds, and anthropoid primates, that is, monkeys and apes (Chatterjee, Reference Chatterjee1973). Anthropoids, for instance, occupy a frugivorous niche, and fruits often contain sufficient vitamin C. Here, gene function is offloaded onto environmental resources. In turn, this offloading has behavioral implications. If a species becomes dependent on its environment (“auxotrophic”) for vitamin C, both its behavioral range and evolutionary trajectory become constrained by the availability of the nutrient. Humans are a nice example of this. As our species migrated across the planet, we found ourselves in environments where vitamin C was in short supply. A deficiency of vitamin C causes scurvy – the bane of seafarers until the trial-and-error discovery that certain food items such as sauerkraut and citrus could prevent ships from being packed with tired, bleeding, toothless, and eventually dead sailors (Lamb, May, & Harrison, Reference Lamb, May and Harrison2017).

Other species have other ways of obtaining vitamin C from their environments, and once these alternative pathways are established, the function of GLO becomes masked. A functionally masked gene is removed from the selection pressures that brought it about in the first place, and can therefore be culled from the genome, as evolution tends to do to unused elements (Albalat & Cañestro, Reference Albalat and Cañestro2016; Wolf & Koonin, Reference Wolf and Koonin2013). Masking does not necessarily need to be in the direction from culture to genes: We can think of genetic assimilation (Crispo, Reference Crispo2007; Waddington, Reference Waddington1953) as the same process working in the opposite direction, where a trait that is regularly acquired through learning gradually transfers its locus to the genome (see Baldwin effect; Morgan, Suchow, & Griffiths, Reference Morgan, Suchow and Griffiths2020). Genetic assimilation can eliminate the cost of learning, but only by sacrificing flexibility. The masking of genes by cumulative culture can eliminate the metabolic cost of endogenous synthesis in the case of vitamin C, and perhaps other costs in other cases, but only by sacrificing reliability of outcome.

Culture and genes are tightly intertwined in this manner, but this kind of functional interaction occurs across various levels of biological organization: between genes within the same genome – intragenomic (Phillips, Reference Phillips2008), between nuclear and organellar (mitochondria and plastid) genomes – cytonuclear (Sloan et al., Reference Sloan, Warren, Williams, Wu, Abdel-Ghany, Chicco and Havird2018), and between host and microbial symbiont genomes – holobiontic (Bordenstein & Theis, Reference Bordenstein and Theis2015). Mitochondria, for example, are believed to have undergone extensive reductive evolution, transferring nearly all of their genes to the nuclear genome (Sloan et al., Reference Sloan, Warren, Williams, Wu, Abdel-Ghany, Chicco and Havird2018; Wolf & Koonin, Reference Wolf and Koonin2013). Indeed, the residual mitochondrial and nuclear genomes collaboratively assemble “chimeric” proteins (Osada & Akashi, Reference Osada and Akashi2012). This kind of coevolution is not uncommon in the history of life (Jablonka & Lamb, Reference Jablonka and Lamb2005; Laland et al., Reference Laland, Uller, Feldman, Sterelny, Müller, Moczek and Jablonka2015).

Similar to these cross-level interactions, culture and genes are interwoven in the construction of many behavioral traits, making separation effectively impossible. Because these two systems interact within a shared space of phenotypic variation, a focus on only one or the other leaves significant ambiguity in the causal underpinnings of these phenotypes. For example, is language primarily the result of culture or genes? Such a question can be answered only by recourse to both sides of our dual inheritance (Chater, Reali, & Christiansen, Reference Chater, Reali and Christiansen2009; Christiansen & Chater, Reference Christiansen and Chater2008; Deacon, Reference Deacon1997; Dediu, Reference Dediu2011; Dediu & Ladd, Reference Dediu and Ladd2007; Wong et al., Reference Wong, Kang, Wong, So, Choy and Geng2020). Epistemological biases in the interpretation of the causal loci of phenotypic traits can misinform policy decisions and impede progress in the research and development of beneficial interventions. Given the inherent ambiguity that arises because of gene–culture interaction, it is important to adopt a sufficiently integrative framework to interpret findings.

4.1.1. Discrepancies in cultural transmission across societies

A meta-analysis (Tucker-Drob & Bates, Reference Tucker-Drob and Bates2016) found the gene × SES interaction on IQ in a subset of US samples, but not in samples from Europe and Australia. Pooling the US studies, the authors found an effect size that corresponds to a heritability estimate of 0.61 at 2 standard deviations above the mean SES but only 0.26 at 2 standard deviations below the mean. In Western Europe and Australia, heritability is more uniform. The cause of this interaction is still debated.

Several researchers (e.g., Bates, Lewis, & Weiss, Reference Bates, Lewis and Weiss2013; Beam, Turkheimer, Dickens, & Davis, Reference Beam, Turkheimer, Dickens and Davis2015; Tucker-Drob, Briley, & Paige Harden, Reference Tucker-Drob, Briley and Paige Harden2013) have suggested that gene–environment correlation via phenotype-to-environment transmission, otherwise referred to as “reciprocal causation,” is the most likely explanation. By this explanation, those with genes well suited to a task can better nurture their skills in a wealthier environment compared to in a poorer environment. That is, initially small differences in genetic potential become gradually amplified over time because of the iterative matching of environments to abilities: An increase in expressed ability brings forth new environmental conditions that enable further growth along that dimension (Bronfenbrenner & Ceci, Reference Bronfenbrenner and Ceci1994; Dickens & Flynn, Reference Dickens and Flynn2001; Scarr, Reference Scarr1992). Such processes can increase genetic heritability, but through reciprocal shaping between genetic potential and environment, rather than through innately specified ability levels. The reasoning is that high-SES households are able to provide environments that do this more effectively and are thereby able to let genetic potential be more reliably associated with the corresponding outcomes, lifting heritability as a result. While such reciprocal causation may indeed be occurring, reconciling this explanation with the findings from Europe and Australia seems more challenging or at least incomplete.

Heritability is a function of the variability in culture, which is shaped by cultural-group dynamics and cultural evolutionary forces. In the United States, the differences between, for example, school and home environments among high-SES households is likely to be small relative to differences between school and home environments among low-SES households, where factors such as school lotteries can dramatically affect the cultural input. In contrast, the cultural environment is less unequal in Western Europe and Australia, where, for example, high-quality schools are available across SES. Where these two explanations make different predictions is for poorer countries. The reciprocal causation explanation would predict low heritability in poorer countries. The cultural evolution of genetic heritability explanation would instead predict high heritability where there is equal access to similarly poor schools and household conditions, but low heritability if inequality is high.

Although environmental variability is commonly construed through the framework of economic inequality, we can also look at it through the lens of cultural variability. For example, even if a society maintains economic equality, it may still contain cultural subdivisions that lead to multiple clusters each with different traditions and behaviors. When this cultural clustering creates differences in the level of some given trait between clusters, the heritability of this trait decreases in accordance with the strength of the clustering.

We predict that networks of cultural transmission are more fractured (i.e., contain a higher density of cultural clusters) in low-SES households than they are in high-SES households comparing within a country, at least for (but not limited to) transmission of cultural information and influence that impacts traits in domains such as cognitive ability. We also predict that these cultural networks are more fractured in the United States compared to those in Australia and Europe, with the greatest fracturing expected in low-SES US communities. Moreover, we predict that these differences across groups will explain at least some portion of the interaction between SES and heritability in the measurement of IQ. These effects could be tested through causal identification techniques, including randomized controlled trials or natural experiments such as school lotteries.

A number of studies have examined unstandardized variance components to make sense of the forces that are driving the gene × SES interaction effects for cognitive ability, but the results have been mixed. In some studies, the interaction is apparently because of an increase in variance explained by the shared environment (the C component of the ACE model; A: additive genetic variance, C: shared environmental variance, E: nonshared environmental variance) in lower-SES households compared to higher-SES households (e.g., Hanscombe et al., Reference Hanscombe, Trzaskowski, Haworth, Davis, Dale and Plomin2012; Kremen et al., Reference Kremen, Jacobson, Xian, Eisen, Waterman, Toomey and Neale2005), while other studies suggest that it is instead because of a decrease in variance explained additively by genes (the A component of the ACE model) (e.g., Bates et al., Reference Bates, Lewis and Weiss2013; Kirkpatrick, McGue, & Iacono, Reference Kirkpatrick, McGue and Iacono2015). The shared environment explanation is more consistent with cultural dynamics, although reality may be more complicated with other processes such as reciprocal causation (Bronfenbrenner & Ceci, Reference Bronfenbrenner and Ceci1994; Dickens & Flynn, Reference Dickens and Flynn2001; Scarr, Reference Scarr1992) contributing to the same effect through other pathways, such as via moderation of the additive genetic component. Moreover, we would predict that this environmental variance would be reduced if cultural opportunities and transmission networks among low-SES households became more broadly connected, supporting greater cultural homogeneity. This would be true, even if the level of poverty or other indicators of well-being remained just as low. That is, this effect is not about poverty or deprivation per se – heritability can be high even among lower-SES groups – but more about cultural clustering. Simply moving neighborhoods can vastly improve life outcomes for low-SES Americans (Chetty & Hendren, Reference Chetty and Hendren2018a, Reference Chetty and Hendren2018b), but we would not predict such large effects for high-SES Americans.

In general, we predict that Scarr–Rowe-like discrepancies in the heritability of IQ will be found when comparing groupings that vary in their degree of within-group cultural clustering (or homogeneity) whenever this clustering imposes barriers upon the cultural transmission of cognitive ability. The more clustered society will be associated with a lower heritability for the trait; this effect would be expected across many culturally transmissible traits beyond IQ as well. When disaggregated by cultural cluster (e.g., SES), we should expect lower heritability among the more culturally diverse and/or more culturally clustered subgroup.

4.1.2 Discrepancies in environmental variation between humans and other animals

A comparison of these SES effects on humans with similar experiments in rodents offers additional insight. Sauce et al. (Reference Sauce, Bendrath, Herzfeld, Siegel, Style, Rab and Korabelnikov2018) found that mice reared in an enriched environment exhibited lower heritability of a rodent analog of general intelligence (for standard rodent-learning tasks such as odor discrimination and navigation) than did mice reared in a control environment, with a heritability of 0.15 in the enriched group versus 0.55 in the controls. The directionality of this difference is opposite of what we have described earlier for the human literature, and the opposite of what the reciprocal causation explanation would suggest (Bronfenbrenner & Ceci, Reference Bronfenbrenner and Ceci1994; Dickens & Flynn, Reference Dickens and Flynn2001; Scarr, Reference Scarr1992). If we assume that this type of environmental enrichment paradigm can be mapped onto differences between high- and low-SES environments in humans, as has been argued (Hackman, Farah, & Meaney, Reference Hackman, Farah and Meaney2010; Rosenzweig & Bennett, Reference Rosenzweig and Bennett1996), the reported discrepancy between rodent and human studies might appear surprising.

Non-genetic phenotypic variation in mice appears to be generated in part by initially small, random differences in experience being amplified over the course of development (Kempermann, Reference Kempermann2019) – complex environments enable greater amplification of these initial differences and hence “individualization,” whether the environmental complexity is of a physical (Freund et al., Reference Freund, Brandmaier, Lewejohann, Kirste, Kritzler, Kruger and Sachser2013) or social nature (Shemesh et al., Reference Shemesh, Sztainberg, Forkosh, Shlapobersky, Chen and Schneidman2013). In these studies, both genes (because of inbreeding) and the shared environment are controlled. It is, therefore, the nonshared environment, or individually unique forms of experience, that is driving this differentiation. The enriched environment enables a greater range of variation in experience and as a consequence, genetic effects are proportionally reduced.

Compare this to humans, where cultural transmission gradually homogenizes the environment within a cultural cluster (though not necessarily between cultural clusters), and to varying degrees across clusters as well. We predict that high-SES US households as well as households within Australia and European countries are less culturally fragmented compared to low-SES US households, leading to higher environmental similarity within Australians, Europeans, and high-SES Americans, respectively (but not necessarily high similarity between the groups). In other words, relatively enriched environments (high-SES or Australia/Europe) are also those that are more culturally homogeneous. This, in turn, may explain why enriched environments would be negatively associated with heritability in mice (Sauce et al., Reference Sauce, Bendrath, Herzfeld, Siegel, Style, Rab and Korabelnikov2018) but positively associated with heritability in humans at least under certain conditions (Tucker-Drob & Bates, Reference Tucker-Drob and Bates2016). Our prediction is that the opposite effect between the two species arises because in humans, the amplification of initial differences in experience by complex environments (individualization) is overwhelmed by the homogenizing effect of cultural learning and group dynamics (e.g., through oblique transmission from favored models in the previous generation and horizontal transmission from peers). The more variation is generated by enriched environments and individualization, the more incentive there is to learn from successful individuals or from those who are likely to have learned from successful individuals themselves. Cultural diffusion systematically squashes the variation created by innovation (see the Appendix for a model of such dynamics).

A cultural evolutionary approach to heritability distinguishes between humans and other animals in ways that a standard behavioral genetics approach does not. Whereas behavioral genetics provides no clear reason to differentiate analysis of genetic effects in humans from genetic effects in nonhuman animals, cultural evolutionary theory explicitly formalizes the difference between humans and other animals: our extensive dual inheritance (Boyd & Richerson, Reference Boyd and Richerson1985). The human-specific predictions that arise from a cultural evolutionary behavioral genetics offer clear tests for our hypotheses. The approach may, thus, also be validated by comparative behavioral genetic analyses between humans and nonhuman animals. This research direction may help make sense of some of the limitations of nonhuman animal models, in cases where cultural adaptations create unexpected gaps in generalizability despite sufficient phylogenetic conservation of relevant phenotypes.