The Scarr-Rowe effect (also “interaction” or “hypothesis”) is a gene-by-environment interaction characterized by a reduction in the additive genetic variance of cognitive ability among those with relatively low socioeconomic status (Tucker-Drob and Bates, 2015), often thought to be due to the adverse developmental effects associated with such status. The existence of this effect was first theorized in the 1960s (Jensen, 1968), with Sandra Scarr (1936–2021) initially providing evidence for it in the 1970s (Scarr-Salapatek, 1971). Scarr argued on the basis of her findings that exposure to poverty, which is associated with restricted access to beneficial environmental influences on development, might be an important determinant of between-social-class and between-socially-identified-racial-group mean differences in cognitive performance within the US, insofar as variance in developmentally relevant environmental exposures across such social classes and groups would alter the heritability of cognitive ability across them (Scarr, 1981). David Rowe (1949–2003) led a team of researchers that replicated the effect 28 years later using data from the US Add Health cohort (Rowe et al., 1999). Subsequent researchers have used the term Scarr-Rowe “effect” and related terms (such as “interaction” or “hypothesis”) (e.g., Turkheimer et al., 2009; Woodley of Menie et al., 2018) in recognition of the pivotal role that these two behavior geneticists played in gathering evidence for the existence of the phenomenon.

Since the publication of Rowe et al. (1999), numerous studies have been published examining the effect in a number of different countries employing twin-study-based estimates of the heritability of cognitive ability. The picture that has emerged is complex, however. The meta-analysis of Tucker-Drob and Bates (2015) found, via re-analysis of previously published data, that statistically significant Scarr-Rowe effects (operationalized as SES-by-additivity interactions) were only present in the US. In their European sub-sample, the effect was negatively signed and non-significant. A subsequent study by Figlio et al. (2017) found that the effect is absent in a very large sample of Floridian twins and siblings representing more recent birth cohorts (those born in the 1990s and 2000s), although it should be noted that these researchers were not able to directly assign zygosity in their twin subsample. An Australian twin study found no evidence for the effect (Bates et al., 2016), and a more recent study employing Nigerian twins similarly failed to find evidence for the effect (Hur and Bates, 2019).

Other studies of the Scarr-Rowe effect have used different behavior-genetic techniques, including adoption designs (Loehlin et al., 2022), parent-offspring resemblance measures (Nagoshi and Johnson, 2005; Flores-Mendoza et al., 2017), and molecular methods such as single nucleotide polymorphism heritabilities and polygenic scoring (PGS) (Woodley of Menie et al., 2018; Rask-Andersen et al., 2021; Woodley of Menie et al., 2021; Peñaherrera-Aguirre et al., 2022). The results of these studies have also been mixed, with some (e.g., Woodley of Menie et al., 2018, 2021; Peñaherrera-Aguirre et al., 2022) finding evidence for the effect in both younger and older US cohorts, some finding evidence for the effect in (younger) Brazilian cohorts (Flores-Mendoza et al., 2017), some finding no evidence for the effect in (older) US cohorts (Nagoshi and Johnson, 2005; Loehlin et al., 2022), and another finding evidence for the opposite effect in a large middle-aged UK cohort (Rask-Andersen et al., 2021).

A number of studies have also investigated the Scarr-Rowe effect in relation to educational attainment (EA), and associated scholastic achievement criteria, utilizing both molecular-genetic and classic behavior-genetic approaches [see the review in Ruks (2022)]. Parental education (which includes formal qualifications as well as practiced education in families) is a good predictor of children’s development, both cognitive and academic. The typical effect size (r/β) is about 0.45 (Rindermann and Baumeister, 2015; Rindermann and Ceci, 2018). EA has been found to share approximately 60% of its genetic variance in common with direct measures of cognitive ability in twin studies (Johnson et al., 2006). PGSs estimated with respect to EA also predict variation in direct measures of cognitive ability (Okbay et al., 2016, 2022; Lee et al., 2018).

As with the Scarr-Rowe effect on cognitive ability, considerable heterogeneity with respect to the effect in relation to EA and associated measures has been noted, with some studies finding signs of negative moderation (i.e., apparent anti-Scarr-Rowe effects) on at least some measures of EA (e.g., Rask-Andersen et al., 2021; Ruks, 2022).

Heterogeneity among Scarr-Rowe effects with respect to time and region may stem from a variety of factors. For example, unusually robust institutions designed to promote greater equality and inter-generational mobility may explain the absence of the effect in certain European nations, as well as Australia, compared to the US. Highly generalized extreme poverty is reasonably expected to broadly limit life opportunities across social strata, possibly accounting for the absence of the effect in Nigeria (see Tucker-Drob and Bates, 2015; Hur and Bates, 2019; Baier et al., 2022 for further discussion). Differences in factors such as behavior-genetic measurement model, statistical power, outcome variable choice (e.g., narrow cognitive abilities, full-scale IQ, or educational attainment), and even the Wilson effect (i.e., the increase in heritability of IQ with age) might contribute to the notable variability in study results (see Tucker-Drob and Bates, 2015; Baier et al., 2022; Peñaherrera-Aguirre et al., 2022 for relevant discussion).

Anti-Scar-Rowe effects (e.g., Rask-Andersen et al., 2021) are more difficult for the classic Scarr-Rowe hypothesis to accommodate. Ruks (2022) proposed a recent behavior genetic extension of the sociological compensatory advantage hypothesis (CAH; Bernardi, 2014), which maintains that high-SES families compensate for intrinsic disadvantages of their members by provisioning environments that can cultivate cognitive growth and promote educational attainment. Such disadvantages may include a genetic propensity toward a lower level of EA. In so far as such families can overcome these disadvantages via compensatory provisioning of resources this may in some cases lead to an apparent of the Scarr-Rowe effect. This phenomenon is captured by the broader mechanism of compensation within the gene-by-environment interaction mechanism typology of Shanahan and Boardman (2009). Compensation effects arise when the influence of environmental enrichment on development is sufficiently strong as to prevent the realization of a phenotype to which “disadvantageous” genetic endowments predispose an individual (such endowments relevant to a given phenotype are sometimes collectively referred to as genetic diathesis [although this term is more often used in medical genetics]). Utilizing the German Twin Life sample, Ruks (2022) found evidence for the CAH on an EA measure (tertiary enrollment). It was noted that the effect was primarily associated with the moderating effect of SES on genes for cognitive ability.

In the current analysis a globally representative and very large sample of individuals sourced from the latest data release of the World Values Survey is used to test for the presence of both the Scarr-Rowe effect and the CAH at the cross-national level. Despite lacking twin data, the WVS nevertheless contains internationally harmonized data on both parental and offspring (participant) EA, and also participant (self-reported) social class. These data can therefore be used to detect a moderating effect of class on the degree of parent-offspring resemblance for EA. Parent-offspring resemblance cannot differentiate between genetic and environmental influences on cognitive ability (Devlin et al., 1997). But, as already noted, parent-offspring resemblance has been used in previous studies of the Scarr-Rowe effect in the absence of twin or other types of familial data that would permit the direct estimation of behavior-genetic variance components (e.g., Nagoshi and Johnson, 2005; Flores-Mendoza et al., 2017). A finding of moderation involving a simple parent-offspring resemblance model must therefore be interpreted with caution, as moderation stemming from SES on additive heritability is only one pathway that can influence the degree of parent-offspring resemblance.

As indicated in Table 2, the Hierarchical GLM multiple R reached statistical significance and explained 34.8% of the variance. The analysis revealed a positive and significant effect (these are scaled as semi-partial regression coefficients [sr], which can be interpreted analogously to standardized regression coefficients [β] in standard multilinear regression models) of the BGR set on participant EA. Age negatively and significantly predicted the criterion variable (sr = −0.18). Parental EA positively and significantly predicted participant EA (sr = 0.42), explaining a sizable proportion of the model’s variance. Social class was also a positive but small predictor of the criterion variable (sr = 0.07). The model furthermore revealed positive and significant interactions between biogeographic regions with age, parental EA, and social class. The interaction between age and parental EA did not predict participant EA. In contrast the age-by-social class interaction reached statistical significance. Consistent with the CAH, the model estimated a negative and significant interaction between parental EA and social class, indicating the presence of an anti-Scarr-Rowe effect (sr = −0.04). This interaction is illustrated with a regression plane plot (Figure 1). These correspond to the quantile weights used to estimate the regression coefficients.

A subset analysis was conducted to determine the magnitude of the interaction between parental EA and social class on the criterion variable for each sex separately. The hierarchical GLMs were specified following the same order as that described in Table 2. Both hierarchical GLMs revealed negative and significant interactions between parental EA and social class. As indicated in Table 3, however, no significant sex differences were detected when the outputs of the two models were compared. A quantile regression analysis revealed a positive and significant influence of parental EA on participant EA. It also found that the greater the social class the higher the participant EA (using the lowest social class ordinal as a baseline). Lastly, the model detected negative and significant interactions between parental EA with each social class ordinal on participant EA (using the lowest social class ordinal as a baseline). These results are further summarized in Table 4. A quantile regression scatterplot illustrates the slope variation as a function of various tau values (Figure 2).

A relatively small-magnitude negative effect of social class on parent-offspring resemblance for EA was identified, consistent with predictions from the compensatory advantage hypothesis (CAH). It should be noted that in psychological science the mean effect size [r] is approximately 0.20 (Gignac and Szodorai, 2016). Therefore, values of <0.10 can be taken to indicate the presence of relatively small magnitude effects (Gignac and Szodorai, 2016).

The effect is not confounded with (biogeographic) regional influences, age (which is a negative predictor of participant EA), parental EA (which unsurprisingly is a relatively large magnitude positive predictor of offspring EA), or social class (which is also an independent, albeit relatively small-magnitude, positive predictor of participant EA). The (relatively small magnitude) impact on participant EA of age indicates that older participants have lower EA, consistent with the expectation that younger participants are likely to have benefited from expansion of access to education through modernization (Barro and Lee, 2013). Reduced parent-offspring resemblance for EA among those with “higher” social class levels is not confounded with the various interactions among these main effects either. One notable interaction concerned the positive influence of participant age on social class as a predictor of participant EA. This indicates that while social class is a stronger predictor of EA among older participants, there appears to be increased social mobility with respect to educational opportunities among younger participants.

The quantile regression also found that among those with lower social class ordinals, the strength of the association between parental and participant EA was stronger relative to those with the highest social class ordinal. This result indicates that the finding of an effect consistent with the CAH in these data is robust to the use of different measurement models.

This apparent support for the CAH using a globally representative individual-differences dataset suggests that at the macrosocial scale, access to EA-enhancing resources associated with “higher” social class has (at least historically) had the effect of positively amplifying EA across generations in such a way that reduces parent-offspring resemblance for this outcome.

The precise mechanism through which the CAH may be operating in these data is unclear, however, as parent-offspring resemblance cannot distinguish between environmental and genetic influences on trait development (Devlin et al., 1997). Ruks’ (2022) behavior-genetic study of the CAH in a German twin sample found that the moderating effect of socioeconomic status on the heritability of an EA measure (specifically tertiary enrollment) is mediated by genetic variance associated with cognitive ability. This could explain results of the current study that are consistent with the action of gene-by-environment interactions.

It is important to note that gene-by-environment interactions are not the only pathway through which the CAH might operate. Interactions between and among bioecological factors (such as social class) and behavior-genetic variance components, capturing different aspects of environmentality, might also reduce parent-offspring resemblance for EA. Significant environment-by-environment interactions (Hanscombe et al., 2012) on cognitive ability and related phenotypes are seldom reported in the literature, however Hur and Bates (2019) found that despite the absence of a Scarr-Rowe effect in their study of Nigerian twins, there was nevertheless a significant interaction between a measure of familial chaos and shared environmentality (C) predicting cognitive ability. A small number of prior studies have also noted similar effects. In a large behavior-genetic study from the UK involving twins aged from 2 to 14 years, Hanscombe et al. (2012) noted an apparent C-by-SES interaction, such that among those with lower SES, C variance was greater for cognitive ability than among those with higher SES, even though, as with Hur and Bates (2019), no Scarr-Rowe effect was found. In a study involving German child twins, a significant interaction between non-shared environmentality (E) and SES (proxied by parental education) was noted predicting participant verbal performance (Spengler et al., 2018). The meta-analysis of Tucker-Drob and Bates (2015), on the other hand, found no evidence for these environment-by-environment interaction effects, when these were estimated with respect to both environmental variance components for the full set of studies.

A point needs to be made concerning the relatively small magnitude of the interaction terms identified in the current study, which has been used by some to dismiss the theoretical significance of prospective gene-by-environment and related effects more generally (for discussion see McGue and Carey, 2017). This situation is, however, consonant with the expectation that, when present, such interactions should be smaller than their associated main effects as the interactions will likely only influence trait variance to a small degree when considered independently, but may have additively larger impacts when grouped (McGue and Carey, 2017). As the CAH effect was estimated with respect to a single bioecological factor (social class) in the current study, its relatively low magnitude is therefore in line with theoretical expectations. It is furthermore increasingly clear that the nature of these interactions, both in relation to cognitive ability and related traits or outcomes such as EA, is much more complex than Sandra Scarr and other researchers initially thought. Theoretical work such as that of Shanahan and Boardman (2009) helps in making sense of the broader pattern of findings that have now emerged in the half-century since the publication of Scarr-Salapatek (1971). Nonetheless, it is apparent that bolder general theorizing concerning these interactions, and predictions about where and when they will be present (in either direction) or absent altogether are needed in order for scientific progress to continue in this area.

Given this need, presented here is some speculative theorizing about the nature of the effects under consideration as well as some testable predictions that emerge from this. Scarr-Rowe and CAH research might best advance if it is integrated within a broader behavioral and evolutionary ecology framework describing how and why organisms vary along a preparedness-plasticity axis. Preparedness is the “degree to which an organism is genetically predisposed toward a particular developmental trajectory, whereas [plasticity] constitutes the degree to which gene–environment interaction induced phenotypic changes during development may alter that prepared trajectory” (Woodley of Menie et al., 2015, p. 2). Properly understanding the Scarr-Rowe and CAH effects likely requires that the environmental factors modulating the degree of plasticity-preparedness a person will exhibit in development are taken into account.

A reasonable prediction is that human populations contending, to a rather uniform degree, with especially harsh environments—due, for example, to very low GDP per capita—will be epigenetically biased in development to exhibit lower plasticity and higher preparedness, as reduced sensitivity to environmental insults in ontogenetic time reduces the potential fitness costs of such insults. Societies with very and pervasively low absolute material quality of life might then fail to exhibit either Scarr-Rowe or CAH effects, due to unfavorable environments generally constraining realization of developmental plasticity (such environments likely also constrain the development of many traits to the full genetic potential of individuals who are raised therein). Conversely, societies high in inequality of material quality of life, but whose poor have benefited substantially from modernization, may exhibit both the Scarr-Rowe and CAH effect simultaneously, with the highest heritability of traits and outcomes such as cognitive ability and EA observed in those with mid-level SES. This is because in such societies, environments may be good enough for even those quite low in SES to be epigenetically predisposed to manifest developmental plasticity to such a degree that the heritability of certain traits is reduced for them relative to other SES groups within their society, via gene-by-environment interactions associated with the harmful influences of low-SES environments, giving rise to the Scarr-Rowe effect; and concomitantly, those with very high SES will likely tend to exhibit higher plasticity due to their favorable environments, which would be leveraged to compensate for diathesis at the level of genetic potential, again through gene-by-environment interactions, giving rise to the CAH effect. A society that prospectively meets these criteria, and would be predicted to exhibit both Scarr-Rowe and CAH effects, would be that of India, owing to the presence of an increasingly Westernized middle class, coupled with extremes of both poverty and wealth. Very large samples would be needed to properly saturate models, however (as a large number of non-linear, in addition to linear interactions would need to be estimated in addition to the effect of interest).

Publicly available datasets were analyzed in this study. This data can be found here: https://www.worldvaluessurvey.org/WVSNewsShow.jsp?ID=428.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent from the patients/participants or patients/participants legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements.

MW: Conceptualization, Writing – original draft. MS: Writing – original draft. MP-A: Writing – original draft. HR: Writing – review & editing.