Another hobbyhorse of internet hereditarians is to cite the “Wilson Effect” to dismiss contrary evidence. For instance, when one observes that gaps between African-Americans and whites are about null in some studies (Rothestein and Wozny 2009) or even that they have decreased (Rindermann & Thompson 2013, Rindermann & Pichelmann 2015), the hereditarian simply responds that this is an artefact of the Wilson effect and that if the children grow up, the gap will grow again (whether or not this is true is unrelated to the question of whether the gap is genetic). Effectively, what the Wilson effect describes is the tendency for some types of family studies to estimate an increase in the h2 coefficient longitudinally, while shared environment goes to zero. The typical studies cited on this issue are Haworth et. al (2010) and Bouchard (2013). However, more robust and comprehensive reviews of the literature show the effect is non-linear and that the shared environment component may not reach zero in adulthood (Dickens & Protzko 2015) using twin methodologies.
The issues with the proposition are innumerable. Firstly is the very conceptual idea of “heritability”. As I have long argued, “heritability” is a confused concept as such. It does not measure “geneticity” or a “genetic component” or anything like that. Having a higher heritability doesn’t mean something is “more” genetic and having a lower heritability doesn’t mean something is “less” genetic because the very question of “less” or “more” genetic doesn’t make sense in and of itself (Tabery 2014, Taylor 2016, Fox Keller 2010, Oyama 1985). That this heritability coefficient increases over time does not mean something becomes ‘more genetic’ and/or ‘less environmental’. Even more, there is the issue that heritability coefficients only have interpretations for a given population, in a given environment, at a given time. Introducing the longitudinal component violates all three of these requirements, as the population itself changes due to attrition, the environment changes due to age-related cohort effects and niche construction alongside development, and of course, the time has changed. A change in the heritability statistic can reflect any of those factors, or others. The fact is that not a single gene (to my knowledge) that is purportedly “for” intelligence has been shown to differentially express itself by age to demonstrate the coherence of the Wilson effect . Moreover, there is a question of whether IQ scores are measurement invariant across age cohorts and are commensurable as such, which has been shown to be violated in some samples (Wicherts et. al 2004, Hertzog & Bleckley 2001), but not others (Niileksela et. al 2013, Bowden et. al 2006), though the full aspects of measurement invariance (factorial, intercepts, etc) are rarely all tested and suffer from methodological, statistical and data issues.
Secondly, there is the contrary evidence. Despite the so-called “evidence” Bouchard marshals in support of the Wilson Artefact, there is a growing literature (and actually historical!) showing that the assumptions behind the methodology used to “observe” the Wilson effect are violated. The first paper published in this vein comes all the way from the 1930s; Wright (1931) fit a path analysis model to Burks IQ data and estimated that the heritability of IQ in childhood is .50, while it is .30 in adulthood. Rao et. al (1976) fit large corpuses of IQ data to path analysis models in adulthood and childhood and found that the heritability of intelligence is smaller in adulthood than in childhood. They also clarified the confusion that twin models and adoption models (relatively rare and non-representative; more to come on this) are better sources of information than larger familial relationships on this question. Another study from the same group, Rao et. al (1982), found that the heritability of IQ increased in adulthood for phenotypic homogamy, and decreases in adulthood for social homogamy (social homogamy seems to be the case for IQ, Keller et. al (2019)). A later study, Devlin et. al (1997) used Bayesian inference to test age-effects models against their maternal effects model and found that all age-effects models were inferior (in terms of the Bayes factor). A very recent adoption study has actually observed increasing correlations of adoptive parents to adopted children over time, meaning an increasing effect of shared environment over time rather than the decrease that the Wilson artefact posits.
Third is another explanation for the “Wilson effect”, namely that of test construction. Presuppositions about the way that intelligence develops over time feed item selection for IQ tests, which then inform the “estimates” of the heritability of IQ that can be found in the literature. Alternative item selection methods could show opposite trends; decreasing heritability over time, or no heritability at all. The fact is that “IQ” is not a single metric (but rather hundreds to thousands), nor is there a theory of “intelligence” (Richardson 2002, 2017) that one can verify “increasing heritability of IQ” with.
And finally, there are gene-environment interaction and gene-environment (evocative and active) correlation explanations of the increase. For example, it could be that small (heritable) differences in IQ at the beginning of development are magnified through processes of niche construction and environmental choice, inducing gene-environment correlations that bias heritability estimates (Dickens & Flynn 2001, 2002; Dickens 2019). This has been confirmed a few times empirically, by Beam et. al (2015), De Kort et. al (2014), using Phenotype→Environment models, as well as with a meta-analysis of longitudinal twin modeling (Briley & Tucker-Drob 2013). Despite the statistical and methodological limitations for these types of studies, they give strong confirmation to actual developmental and reciprocal effects models of intelligence, rather than ones that merely posit that the h2 statistic has increased “because of genetics” (as assumed in the typical interpretation of the Wilson artefact). Other developmental explanations include the accumulation of the violation of the EEA over time, in addition to the unique sociocognitive effects of dizygotic twin relationships (Richardson & Norgate 2005).
In sum, this is yet again more evidence that resemblance-based correlations are no substitute for the actual analysis of environments, genes, and the interaction between them.
 Such a differential association could also be the result of emergent developmental processes, environmentally mediated genetic influences, or other non-genetic causes.