Educational attainment: Genetic variance accounted for 51% of the total variance, but within women and men, 40% and 58% of the total variance respectively

Nature, Nurture, and the Meaning of Educational Attainment: Differences by Sex and Socioeconomic Status. Thalida Em Arpawong et al. Twin Research and Human Genetics, March 13 2023. https://doi.org/10.1017/thg.2023.6

Abstract: Estimated heritability of educational attainment (EA) varies widely, from 23% to 80%, with growing evidence suggesting the degree to which genetic variation contributes to individual differences in EA is highly dependent upon situational factors. We aimed to decompose EA into influences attributable to genetic propensity and to environmental context and their interplay, while considering influences of rearing household economic status (HES) and sex. We use the Project Talent Twin and Sibling Study, drawn from the population-representative cohort of high school students assessed in 1960 and followed through 2014, to ages 68−72. Data from 3552 twins and siblings from 1741 families were analyzed using multilevel regression and multiple group structural equation models. Individuals from less-advantaged backgrounds had lower EA and less variation. Genetic variance accounted for 51% of the total variance, but within women and men, 40% and 58% of the total variance respectively. Men had stable genetic variance on EA across all HES strata, whereas high HES women showed the same level of genetic influence as men, and lower HES women had constrained genetic influence on EA. Unexpectedly, middle HES women showed the largest constraints in genetic influence on EA. Shared family environment appears to make an outsized contribution to greater variability for women in this middle stratum and whether they pursue more EA. Implications are that without considering early life opportunity, genetic studies on education may mischaracterize sex differences because education reflects different degrees of genetic and environmental influences for women and men.

Discussion

In this article, we addressed a long-standing question on the importance of nature, via genetic endowment, and nurture, via shared and unique environmental influences, for EA. We found that the balance of nature and nurture underlying EA is not uniform between sexes. First, men and women who were raised in homes with higher household economic status had more years of EA and greater variability in EA than those raised in homes with lower household economic status. Second, we found that overall, there was larger genetic and total variance underlying EA for men than women. Third, nature makes the largest contribution for individuals from the highest family-of-origin economic backgrounds for both men and women. When sex and household economic stratum are both considered, absolute genetic variance contributes similar amounts for men across the economic strata, as well as for women from only the highest economic strata. For women in the lowest and middle economic strata, genetic variance contributes much less to the variability in EA compared to women in the highest economic stratum and to male counterparts across all economic strata. Unexpectedly, for women from the middle economic stratum, it appears that family rearing environment, which may in part reflect parents’ own genetic endowments, may take on an outsized role in contributing to EA. These results confirm that critical interrelationships exist, with nurture moderating effects of nature to alter the range of influence possible on EA. Greater total variance and expression of genetic potential for EA is afforded differently by degree of household economic resources when growing up and in combination with sex or gender.

Findings from this study help us understand the etiology of EA and that EA does not mean the same thing across people, especially for older women today, who were born in the early 1940s. Differences in household economic resources contribute to a disparity in total years of education attained, evidenced by lower overall means and less variability in years of EA for both men and women in the lowest household economic brackets. The finding on differing sources of variance for EA for women by socioeconomic strata was not detectable when examining phenotypic HES × sex effects in predicting EA. This points out that null results in phenotypic models that test sex interactions do not preclude there being differences in etiologies for men and women, particularly with regard to outcomes like EA that are likely influenced by complex processes related to gender socialization and expectations.

Overall, genetic variance accounted for more variation in men’s EA than women’s, at 58% and 40% of the total variance respectively. This is consistent with the ranges for sex-specific calculations reported in prior literature (Baker et al., Reference Baker, Treloar, Reynolds, Heath and Martin1996; Branigan et al., Reference Branigan, McCallum and Freese2013; Heath et al., Reference Heath, Berg, Eaves, Solaas, Corey, Sundet, Magnus and Nance1985; Nielsen & Roos, Reference Nielsen and Roos2015). In turn, the role of nurture was greater for women than for men. This finding supports the interpretation that socio-cultural factors and opportunities shape different trajectories of expression of genetic endowment for men and women (Allan, Reference Allan2011; Klein et al., Reference Klein, Ortman, Campbell, Greenberg, Hollingsworth, Jacobs, Kachuck, McClelland, Pollard, Sadker, Sadker, Schmuck, Scott and Wiggins1994). In this cohort, men were able to pursue genetically driven talents for EA irrespective of socioeconomic strata of their families of origin, but women did not have the same benefit unless in the highest HES group. These findings are in line with prior research that has not detected SES-by-sex differences in the heritability of EA (Branigan et al., Reference Branigan, McCallum and Freese2013; Silventoinen et al., Reference Silventoinen, Krueger, Bouchard, Kaprio and McGue2004) in countries that implement social policies to promote equity in access to educational opportunities (Ahola et al., Reference Ahola, Hedmo, Thomsen and Vabø2014; Gorard & Smith, Reference Gorard and Smith2004; Kyrö & Nyyssölä, Reference Kyrö and Nyyssölä2006). Our findings also support evidence thus far on sex differences by country and birth cohort that show EA likely reflects accumulated genetic sensitivities to the environment (gene-by-SES and gene-by-sex) that are different depending on environmental circumstances (Heath et al., Reference Heath, Berg, Eaves, Solaas, Corey, Sundet, Magnus and Nance1985; Silventoinen et al., Reference Silventoinen, Krueger, Bouchard, Kaprio and McGue2004), and therefore support for G × E effects, by sex and socio-economic group. This points to the results reflecting an opportunity structure and differences in men’s and women’s lived experiences, not biological sex differences.

Shared family estimates from this study are substantially smaller than what is reported in the most recent meta-analysis of proportional variance, yet closer to what is expected given prior knowledge of twin and family studies of other traits (Turkheimer, Reference Turkheimer and DiLalla2004). Shared environmental variance encompasses family-level resources, including the measured component of household economic status and nonmeasured components, such as family activities and behaviors modeled at home to facilitate exposure to scholarly interests or success in academic pursuits, an living in social and built environments that promote EA. These are not entirely distinguishable from the larger community-level environmental factors that members from the same household share, such as better school quality, or access to healthcare services that promote mental and emotional health. Among women in the middle socioeconomic stratum, the shared family environment appears to make a particularly weighty contribution to greater variability in whether these women pursue higher educational attainment. Conceptually, the family-level resources can have genetic components (e.g., through genetic-environment correlation, effects of assortative mating), but given that twin correlations for both MZ and DZ women were similar and large, this implies that twin and sibling members of the family experience them as a part of their shared, social environment.

Environmental factors on EA that are resources unique to the individual could include parent expectations placed on individual children, peer encouragement, varied learning opportunities offered by teachers, or experiences after school that reinforce scholarly pursuits. When comparing estimates by sex and HES strata, differences in unique, individual-specific experiences are relatively small. Although it is possible these factors have profound influence on particular individuals, findings suggest that adolescents who grew up with more influences from both unique factors and socioeconomic resources in the family are more variable in whether they pursue higher educational opportunities.

The balance of nature and nurture components holds implications for use of EA to predict later life outcomes for different groups. While there are robust and consistent correlations in the literature between education and cognitive function (Opdebeeck et al., Reference Opdebeeck, Martyr and Clare2016; Ritchie & Tucker-Drob, Reference Ritchie and Tucker-Drob2018), our prior work showed that genetic variance underlying earlier life cognitive ability overlaps only 11% with genetic variance sources for EA (Arpawong et al., Reference Arpawong, Zavala, Gatz, Gruenewald and Prescott2018). This finding suggests that the strong relationship between education and cognition is predominantly driven by overlapping nurture components, including life experiences and resources. Relatedly, while education has shown strong predictive ability for cognitive impairment and dementia (Caamaño-Isorna et al., Reference Caamaño-Isorna, Corral, Montes-Martínez and Takkouche2006), it has also shown differential ability to predict dementia risk by sex. In particular, education has the lowest predictive value for risk of dementia among women in more impoverished countries (Sharp & Gatz, Reference Sharp and Gatz2011). Given our findings, we speculate that constrained potential in lower resourced countries and for women means that irrespective of genetic potential, there is less opportunity for these women to attain education; hence, this reduces the overlapping genetic variance between education and cognition. If access to education is driven by within-country environmental factors (e.g., access to resources to pay for education, social prioritization of academic achievement for boys vs. girls), this likely reduces the genetic correlations for EA and cognitive status. In contrast, in higher resourced countries and for men, genetic endowment has more opportunity for expression and thus greater overlap.

In this Project Talent sample, we had limited power to test effects of other social constraints, such as racial/ethnic inequalities. Additionally, we are unable to assess mechanisms by which socioeconomic bracket influences educational differences beyond the variance components quantified, or for those who would not have attended high school given the recruitment design for Project Talent and compulsory schooling laws. Furthermore, we cannot conclude causal associations. For instance, common concerns about causal inference in observational studies center on issues of reverse causation and confounding (McGue et al., Reference McGue, Osler and Christensen2010). With the present study, we use a longitudinal design where genetics and household economic status precede educational attainment, thus alleviating the first concern. With the second concern, invoking the twin design enables us to control for genetics and shared family environment, and thereby account for the degree of influences from unmeasured environmental factors, or potential confounders (McGue et al., Reference McGue, Osler and Christensen2010). Thus, although we are not able to establish causality with this study, we are able to make inferences for the direction of effects. A limitation to the design is our inability to make general inferences about siblings because those included are all siblings of twins, and siblings within the same age range of twins, and thus are not representative of the experience of all siblings within families. Lastly, results are likely cohort specific because our estimates align well with prior research evaluating variance sources in education in individuals born between 1940 and 1961 (Heath et al., Reference Heath, Berg, Eaves, Solaas, Corey, Sundet, Magnus and Nance1985). Follow-up analyses in younger cohorts will be important to compare differences in findings.