Misguided model of human behavior: Comment on C. H. Burt: “Challenging the utility of polygenic scores for social science…”

Abstract

This commentary emphasizes two problem areas mentioned by Burt. First, that within-family designs do not eradicate stratification confounds. Second, that the linear/additive model of genetic causes of form and variation is not supported by recent progress in molecular biology. It concludes with an appeal for a (biologically and psychologically) more realistic model of such causes.

Behavior geneticists tend to think that their field is unfairly controversial because of past associations with racism and eugenics. But there's more to it than that. Over the history of BG many scholars have commented on its seeming existence in a parallel universe, demanding relaxed scientific standards, building castles in the air with much reliance on “promissory notes,” as Burt puts it. Regarding BG's grounding in unlikely assumptions, Kempthorne (1978, p. 18) asked “How naive can you get?” An illustration is how variations in cognition, educational attainment (EA), height, weight, and so on, are considered to be equally “complex,” with similar causal patterns of form and variation, as if eons of evolution and gulfs of biological necessity had never happened.

Another example, of course, is how genome-wide association studies/polygenic scores (GWASs/PGSs) appeal to vague “phenotypes,” using poorly validated measures, “surrogates” and “proxies,” inferring causes from mountains of correlations that largely wash-out over time (Richardson & Norgate, 2015). Noting such thin evidential gruel Fletcher (2021, p. 256) refers to the “sleights of hand and folk wisdom from behavioral genetics.” Burt expertly exposes such problems in BG's contemporary search for validation at the molecular level. This commentary enlarges on two aspects.

Population stratification within-families

Burt rightly notes that the problem of population stratification haunts all interpretations of GWASs/PGSs and cannot be easily eliminated. BG researchers, however, seem to agree with Zaidi and Mathieson (cited by Burt) that “family-based studies are immune to stratification,” and present the equivalent of a randomized controlled trial (RCT; Harden, 2021). However, children are not merely passive vehicles of additive G and E effects. They actively constructive perceptions of their worlds and creatively react to them.

Burt (target article, sect. 5.1.2, para. 7) mentions family dynamics that create “micro-stratification” within families. However, such dynamics also interact with wider social contexts to further generate spurious SNP–trait correlations in GWASs/PGSs. Facial appearance; height; weight; body shape; subclinical medical conditions such as myopia; hair form; and skin color (e.g., Hall, 2017), are all subject to positive/negative feedback from peers and teachers, as well as siblings and parents, acting to cultural norms. They create unequal psychological effects and reactions that course through individuals' school and occupational careers (Kraft, Kraft, Hagen, & Espeseth, 2022; Wilkinson & Pickett, 2018). Yet all will involve hundreds or thousands of SNPs covarying non-causally with psychological traits, and producing spurious GWAS/PGS results. They completely confound the preconditions for an RCT that Harden (2021) and others recommend.

Genetic causes

Burt says that “no serious scientist can suggest that genetic differences do not influence – in some complex, context-dependent way – developmental differences” (target article, sect. 6, para. 6). But we need to be clear that “influence” is not necessarily tractable as prediction. Unfortunately, prediction in BG is still dominated by the Galton/Fisher model. Despite acknowledging roles for the “environment,” “interactions,” and so on, mere attenuation of linear/additive genetic effects is assumed. So we get genomes described as “blueprints” (Plomin, 2019); or even as “cookbooks” (Harden, 2021); whereas Madole and Harden (2023) assert that PGSs reveal an individual's “genetic propensity for a trait” (Madole & Harden, 2023; sect. 3.1, para. 4); and that “the parental genotype causes an increase in their education” (Madole & Harden, 2023; note 4). Burt (e.g., target article, sect. 6, para. 6) reveals the fallacies, and the dangers, in such logic.

But so do waves of recent research in molecular biology. The creative, anticipatory dynamics mentioned above, at the socio-cognitive level, have evolutionary precursors in learning/cognitive functions in cells, physiology, brain, and behavior (Lyon, Keijzer, Arendt, & Levin, 2021; Richardson, 2020; Shapiro, 2020). Development of form and variation does not start with gene transcription. DNA can do nothing until activated by the organism. That arises from vast signaling and metabolic networks monitoring the dynamic complexity and changeability of most natural environments. Their precursors in cytoplasm are inherited with genes, and their developmental fates are best described as emergent intelligent systems in which statistical patterns predict impending states by assimilating the covariance structures of the past and present (Richardson, 2021; Shapiro, 2020).

That fundamental, multi-level, intelligence is seen in the context-dependent recruitment of transcription factors, cofactors, enhancers, promoters, and so on (Isbel, Grand, & Schübeler, 2022). Alternative splicing produces a diversity of proteins from the same gene (Wright, Smith, & Jiggins, 2022). A single gene can be associated with the development of a variety of structures and functions (Watanabe et al., 2019). And multiple alternative pathways to desirable structural/functional endpoints are constructed in spite of genetic variation (Biddle, Martinez-Corral, Wong, & Gunawardena, 2021; Wagner & Wright, 2007).

Not even mutations comprise the random genetic lottery that Harden (2021) imagines (Monroe et al., 2022). There are also processes – what Shapiro calls natural genetic engineering (NGE) – through which intelligent cells can themselves change genetic information: “NGE is shorthand to summarize all the biochemical mechanisms cells have to cut, splice, copy, polymerize and otherwise manipulate the structure of internal DNA molecules…Totally novel sequences can result from de novo untemplated polymerization or reverse transcription of processed RNA molecules” (Shapiro, 2013, p. 287).

In other words, genes are best described as intermediary resource-providers for the organism as a whole: Servants to intelligent systems, not autonomous instructors. With the exception of relatively rare disorders or single-gene (Mendelian) variations, there are no independent “effects” of genomes. This is why Noble (2016) says we've had things the wrong way around in our descriptions of genetic causes. Baverstock (2021) uses the analogy of genes as the merchants that provide the necessary materials to build a house, but are neither the architect nor the builder. He calls for a “Copernican revolution” in our geno-centric view of living things.

Help or hindrance?

Burt asks if PGSs will ever be useful and urges caution. Fundamentally, though, it's the scientific framework – the Galton/Fisher model of heredity – that is the root of the problem. It persists, attracts funding and research effort, in spite of the logical, epistemological, and statistical errors described by Kempthorne (1978) and others, because it affirms prior socioeconomic structures. Until that model is replaced by a more biologically realistic one it will be more of a hindrance to the advancement of knowledge than a help.