Genetics and Social Inquiry

Social science and genetic science still have fairly little engagement with one another, but the continued swift development of genetic science has certainly gained social scientists’ attention. First, some social scientists are incorporating techniques from quantitative and molecular genetics into their work. Genetic data is increasingly recognized as providing valuable leverage even for research animated by strict interest in social environmental causes. Second, social scientists have been interested in understanding aspects of genetic science as a social phenomenon. This literature identifies possible noxious effects of uncritical public acceptance of genetic science, although how consistent these speculations are with public opinion and other available data is less clear. Because public understanding of genetics can influence behavior and social policies in ways that affect the ultimate causal potency of genes themselves, adequately theorizing genes as causes requires integration of these two lines of inquiry. 107 Review in Advance first posted online on April 2, 2009. (Minor changes may still occur before final publication online and in print.) A nn u. R ev . S oc io l. 20 09 .3 5. D ow nl oa de d fr om a rj ou rn al s. an nu al re vi ew s. or g by B R A N D E IS U N IV E R SI T Y o n 07 /0 2/ 09 . F or p er so na l u se o nl y. ANRV381-SO35-06 ARI 23 March 2009 16:53 The price to assay a basic bit of genetic information—the single nucleotide polymorphism (or SNP)—is presently declining at rates faster than the fabled decline in the price per bit of computer memory. Consequently, we are at the beginning of what will be a radical increase in the data available to researchers for drawing connections among genetic endowments, life experiences, and observable life outcomes. Direct study has so far been broadly consistent with what indirect study has suggested for decades: that genetic differences are pervasively associated with outcomes of longtime social science interest (Udry 1995, Hernandez & Blazer 2006, Benjamin et al. 2007). Twenty-five years ago, serious scholars could credibly speculate that genetic differences might prove of only trifling importance for understanding individual differences in psychological characteristics, behaviors, and life attainments (Lewontin et al. 1984). Now, the terrain has shifted to arguments that genetic influence should not be overstated and can only be understood in a context of extensive causal interdependence with the social environment (Pescosolido 2006, Rutter 2006). Even though no one can foresee the particulars ahead, every indication is that advances in genetic science will continue to reveal how DNA differences combine with environmental differences to yield different life course trajectories and outcomes. Some sociologists see nothing so opposed to the spirit of their craft as genetics. Deterministic genetic explanations have long served as textbook staples for illustrating what a sociological imagination is not. A common stance has been first to doubt that genetic differences actually are important for outcomes that sociologists study, but then also to assert that, if genetic differences do matter, they would still be irrelevant to the sociological study of those outcomes (Anderson 1967). The dramatic rise of human genetic science has happened despite the skepticism, hostility, and studied indifference of many sociologists. Sociologists have been thinking much more about genetics lately, however, as genetic science has become not just theory and findings but an ascendant social phenomenon. The rapid maturation of this science and its human consequences seem likely to stand as one of the most important developments of our age. As demonstrated by this review, the ways sociologists are now focused on genetics reflect the vast intellectual diversity of the discipline. Some are working with genotypic data, some are examining public beliefs and attitudes about genetics, some are interviewing the scientists, and some are forecasting the long-run implications of genetic innovation for culture. The diversity of sociology’s engagement with genetics also reflects the complex and dynamic character of genetic causes. The classic caricature is one of genetic differences unfolding into phenotypic differences, augmented perhaps by appreciation that this unfolding may be highly contingent upon characteristics of an organism’s environment. Increasingly, social scientists are appreciating that genetics also provide a vital resource for understanding the consequences of environmental differences, and, indeed, genetic and environmental causes likely interpenetrate in ways that thoroughly defy our current capability for causal description. More than this, however, human genomic causality is not like causality in physics or even genomic causality in other species. What people believe about genetics and about their own genomes can influence their decisions and behaviors. As we develop more nuanced understandings of the biochemical causality of genetics, we create new opportunities for the psychosocial causality of genetic information (Freese 2006). Today, we have healthy individuals with genetic diagnoses lobbying the state to fund scientists to discover knowledge that can be translated into new technologies that can be used to prevent their genes from causing pathological consequences. Such possibilities are at the center of much of sociology’s enduring unease with genetics, which reflects not just skepticism about the actual importance of genetic differences but also concern about deleterious social consequences that might follow from transformations in how people think about genes as causes of human difference. Prominent among these concerns has been that 108 Freese · Shostak A nn u. R ev . S oc io l. 20 09 .3 5. D ow nl oa de d fr om a rj ou rn al s. an nu al re vi ew s. or g by B R A N D E IS U N IV E R SI T Y o n 07 /0 2/ 09 . F or p er so na l u se o nl y. ANRV381-SO35-06 ARI 23 March 2009 16:53 the rising public enthusiasm for genetics will divert resources from research and policy oriented toward social structural causes of difference (Duster 2003). Our review is divided into two complementary parts. First, we examine the possibilities for genetic information to elaborate and reshape how social scientists understand how social environments affect individuals. Second, we examine social science research on the possibilities for genetic information to elaborate and reshape individual beliefs and social relations. In both cases, we find that existing work is highly suggestive but also sharply limited, and the extent to which either social science or society will be reshaped by the increased availability of molecular genetic information remains unknown. GENETICS AND THE STUDY OF SOCIAL ENVIRONMENT As noted, sociologists who acknowledge the possible importance of genetic influence for outcomes they study might nonetheless regard such influences as simply outside their discipline’s purview. Although much could be said about the historical “boundary work” of sociologists in regard to their relationship with the biological sciences (Gieryn 1999, Pescosolido 2006), it is clear today that genetic information has much potential value for research oriented toward understanding social environmental causes. First, given accumulating evidence that genetic differences may be a source of pervasive confounding in estimating effects of social environmental causes on outcomes, genetic information is being used to try to strengthen causal inference. Second, heritability estimates attempt to measure what percentage of population variation in an outcome is accounted for by genetic differences. Heritability estimates differ across populations, and probing these differences may provide unique information for assessing macroscopic characteristics of societies, especially regarding the intergenerational dynamics of social stratification. Third, a ubiquitous problem in studying social environmental causes has been why such causes almost invariably affect similar people very differently, and studies of geneenvironment interaction and correlation may help illuminate some of this heterogeneity. Improving Estimates of Effects of Environmental Causes Findings of the substantial heritability of traits have been used to argue against any presumption that observed differences among persons can be wholly explained by divergent experiences. For example, the heritability of political attitudes contradict preceding literature on political socialization in which the possible importance of genetic influences on attitudes went virtually unmentioned (Alford et al. 2005). If an environmental cause and outcome each have substantial heritability, then it seems tenuous to presume that an ordinary observational research design will accurately estimate the true effect of the environmental cause.1 For outcomes of interest to social scientists, genes may often be best conceptualized not as confounders per se but as causes of confounding phenotypic characteristics (Freese 2008). That is, the relationship between genetic differences and life circumstances seem likely to be mediated by more immediate individual characteristics. For example, genes may influence personality traits that affect choices (e.g., impulsivity and school continuation), early displays of aptitude that affect opportunities for development (e.g., early cognitive ability and tracking), and characteristics of appearance that affect treatment by others (e.g., skin tone variation within ethnic groups and discrimination). Substantial heritability estimates have been observed for a wide enough range of outcomes so as to prompt otherwise moderate behavioral geneticists to declare that virtually 1In this scenario, for an ordinary observational study to yield an unbiased estimate, either the genetic influences on the environmental cause and on the outcome would need to be entirely independent of one another, or the intervening phenotypic measures would need to be measured fully and without error. www.annualreviews.org • Genetics and Social Inquiry 109 A nn u. R ev . S oc io l. 20 09 .3 5. D ow nl oa de d fr om a rj ou rn al s. an nu al re vi ew s. or g by B R A N D E IS U N IV E R SI T Y o n 07 /0 2/ 09 . F or p er so na l u se o nl y. ANRV381-SO35-06 ARI 23 March 2009 16:53 “everything is heritable” (Turkheimer 2000). These include such outcomes as educational attainment, earnings, divorce, delinquent behavior, and voter turnout (Behrman et al. 1980, McGue & Lykken 1992, Fowler et al. 2008). Even though specific heritability estimates are easily criticized, they nonetheless provide a diagnostic for the possibility of confounding by genetically influenced characteristics (Freese et al. 2003). If the received findings from behavioral genetics are even approximately correct, then conventional sociological techniques for estimating effects of social environmental causes likely yield pervasively biased results. We may expect this bias usually to overstate the consequences that would result from a targeted change of the specific social condition in question, although effects of environmental changes may also be underestimated (Martin 2008). Longitudinal data have provided the traditional first line of defense against confounding from unmeasured differences among individuals. Longitudinal data work best for causes that are specific events, and even then the limitations of inference are well documented (e.g., Winship & Morgan 1999). In contrast, attempts specifically to reduce confounding owing to genetically influenced causes have long relied upon constructing samples of sets of persons with particular genetic relationships, most importantly twins and adoptees. For instance, J. Schnittker & J. Behrman (unpublished manuscript) raise the possibility that estimates of the strong positive effect of schooling on social connectedness may be confounded by unobserved characteristics that cause differences in educational attainment. They find greatly diminished and even sometimes negative effects when attention is restricted to monozygotic (MZ, identical) twins. Because MZ twins share not only genetic endowments but also generic features of family background and cohort, the strategy of estimating effects by comparing MZ twins eliminates a wide variety of potential confounders. Even so, MZ twins do not differ in educational attainment randomly, and estimates following this strategy may still be confounded by psychological and other characteristics that are not fully determined by genes and shared family environment. In other words, if one believes that cognitive ability confounds an estimated effect of educational attainment and also that genes are only one determinant of cognitive ability, then genetically informed designs still do not account fully for cognitive ability as a confound. When samples include pairs with different genetic relatedness [e.g., MZ versus dizygotic (DZ, fraternal) twins, or adopted versus biological siblings], structural equation models can be used to estimate the relationship between an environmental cause and outcome net of overlapping genetic and shared family ætiology.2 Using such models, Prescott & Kendler (1999) report that genetic and other confounding factors account for all of the strong association between age at first birth and alcoholism. Schnittker (2008) reports that genetic endowments account for most of the relationship between socioeconomic success and self-reported happiness. Rodgers et al. (2008) report that the apparent effect of education on age at first birth is accounted for by environmental causes shared by siblings in the same family. Thus, studies using genetically informed designs to assess confounders can yield much different conclusions than the samples of unrelated respondents more commonly used within sociology. Sociological research has invested heavily in data resources that are excellent for descriptive inference about populations (e.g., the General Social Survey); sociology has been slower to recognize how limited such data are for many kinds of causal inference. The assumptions of behavioral genetics models have been heavily scrutinized (Goldberger 1979, Lewontin et al. 1984, Freese et al. 2003, Schaffner 2006). Unfortunately, however, such scrutiny has focused predominantly just on whether violations could result in heritability estimates being so overstated as to find consistently high heritabilities even 2Bayesian techniques are also being increasingly used in quantitative behavioral genetics (see Fowler et al. 2008). 110 Freese · Shostak A nn u. R ev . S oc io l. 20 09 .3 5. D ow nl oa de d fr om a rj ou rn al s. an nu al re vi ew s. or g by B R A N D E IS U N IV E R SI T Y o n 07 /0 2/ 09 . F or p er so na l u se o nl y. ANRV381-SO35-06 ARI 23 March 2009 16:53 when none exist. That possibility as grounds for generalized skepticism toward twin studies now seems quite unlikely (e.g., Kendler & Prescott 2006, Rutter 2006). Far less consideration has been given to assessing the consequences of violations of model assumptions for assessing multivariate relationships, such as assessing the relationship between a measured environmental cause on an outcome (like educational attainment and age at first birth). The model-fitting strategies that are used often imply relatively lower statistical power for detecting such effects. This can make conclusions that genetic overlap entirely accounts for the difference between two observed variables premature (Rutter 2006). Despite their advantages, then, multivariate behavioral genetics models surely should not be mistaken for a panacea for causal inference problems. More work is needed to understand how to combine findings from twin and adoption studies with those from other designs. Compared with conventional models from twin and adoption studies, molecular genetic information allows for the possibility of more direct and more convincing inferences. As we discuss shortly, direct genotypic measures have enormous utility for assessing specific geneenvironment interactions. As control variables in attempts to assess the importance of environmental causes, however, molecular genetic measures are limited in that measuring selected genes does not equate to measuring the whole genome [all genes and gene × gene ( × gene) interactions]. Indeed, a humbling finding for behavioral genetic studies has been the small, often hard-to-replicate nature of estimated specific gene effects for outcomes with substantial heritability (Balaban 2001). Some speculate that one of the strongest ultimate lessons of the new genetics will be greater appreciation of how human development is dynamic and highly contingent in ways that contradict a regression-analysis-oriented conceptualization of either genetic or social environmental causes (Turkheimer 2006, Freese 2008). Even so, molecular genetic information may help illuminate complex causal relationships among variables of enduring social science interest because the genome represents perhaps the primordial natural experiment: For each gene, which of the two copies we receive from each parent is effectively random and independent except for proximal locations on the genome. If a mother has copies of the ε2 and ε4 allele of the APOE gene, the latter of which is associated with increased risk of Alzheimer’s disease, then it is a (quasi-)random matter whether the child inherits the ε2 or ε4 copy, and siblings have an equal chance of being concordant or discordant. Consequently, for example, if particular genes could be identified that have strong effects on lifetime health but did not affect socioeconomic status (SES)— except indirectly, through whatever their effect on health—then variation on these genes among full siblings potentially offers a powerful approach to disentangling the effect of health on SES from the effect of SES on health (Ding et al. 2006; see also Benjamin et al. 2007, p. 305; Ebrahim & Davey Smith 2008). Comparing genetically discordant full siblings using this design also eliminates the possibility of confounders from consequences of parental genes on parental health and subsequent parental SES (Fletcher & Lehrer 2008). This approach depends not just on identifying genes associated with the cause in question (e.g., health), but also establishing that these genes only influence the outcome (e.g., SES) through their influence on health. In sum, genetic information offers many possibilities for strengthening the work of even those researchers who are exclusively interested in social environmental causes. However, the strategies for doing so require both better understanding of the assumptions of the statistical models and further advances in our understanding of genetics. Heritability and Macrostructure The heritability of traits, often presented as statements that x% of variation in a trait is genetic, are not transcendent facts of nature but population-dependent statistics. As such, when heritability estimates vary across populations, www.annualreviews.org • Genetics and Social Inquiry 111 A nn u. R ev . S oc io l. 20 09 .3 5. D ow nl oa de d fr om a rj ou rn al s. an nu al re vi ew s. or g by B R A N D E IS U N IV E R SI T Y o n 07 /0 2/ 09 . F or p er so na l u se o nl y. ANRV381-SO35-06 ARI 23 March 2009 16:53 that variation may indicate something meaningful about the populations. Boardman (2009) finds that the heritability of smoking is reduced in states with more aggressive public health measures against smoking. Guo & Stearns (2002) find that the heritability of cognitive ability is higher among children from higher status families (see also Rowe et al. 1999, Turkheimer et al. 2003). Guo & Stearns interpret the finding as indicating that socioeconomic advantage facilitates children reaching their genetic potential (for critique of this view, see Perrin & Lee 2007). Put differently, the results strongly suggest gene-environment interactions whereby socioeconomic advantage amplifies the positive effects of some genetic variants on cognitive ability. Nielsen (2006) follows similar reasoning to argue that crosssocietal comparisons of heritability estimates of status attainments can be used to infer variation in opportunity for achievement (versus ascription) in a society. After all, if the only systematic determinant of status was the status of one’s parents, then the correlation of status among MZ twins would be the same as that among DZ twins, and the heritability would be zero. At the same time, inferences based on differences across populations in heritability estimates can easily conflate the influence of social environments with the amount of overall variation of environments. Again, heritability estimates are based on the premise that observed variation in an outcome can be partitioned into the percentage accounted for by genetic differences and the percentage accounted for by environmental differences. If the amount of overall pertinent environmental variation decreases, the extent to which the environment can explain variation in an outcome generally decreases, and estimates of heritability increase. In addition, traits like height and skin color variation are heritable, but we would not necessarily think of their contribution to the overall heritability of status attainment as indicative of the relative importance of achievement versus ascription in society (Freese 2008). That is, linking heritability to meritocracy can be misleading insofar as genetic differences influence some traits that may affect attainment in ways typically not recognized as merit. More generally, heritability estimates may provide an intriguing structural diagnostic, but these estimates are strongly limited in what can be inferred from them. Molecular genetic studies of specific gene-environment interactions may ultimately prove much more illuminating for the structural questions that heritability estimates have been used to address. Genetic Moderation of Environmental Causes Social environmental causes commonly explain only a small portion of observed variation among individuals. A corollary is that outcomes typically differ widely among individuals who experience similar events. The same stressor, for instance, can have a strongly negative effect on one person and no discernible impact on another. Despite all the attention from sociologists about the advantages parents confer to children, about three-fourths of the income inequality in the United States is within sibships (Conley 2004). Genetic differences may be implicated in the different ways that two individuals respond to the same cause. The field of quantitative behavioral genetics was once skeptical of the importance of gene-environment interaction (Rutter 2007, pp. 15–16). Now, gene-environment interaction is a hot topic in behavioral genetics, biomedical science, and social science (Shostak 2003, Shanahan & Hofer 2005, Hernandez & Blazer 2006, North & Martin 2008). For sociology, the moderating effect of genetic differences may be key to understanding why individuals exhibit so much more heterogeneity in response to similar environments than would be predicted if they were uniform, passive automatons of structure and culture. Shanahan & Hofer (2005) develop a typology of four common varieties of geneenvironment interactions (see also Ottman 1996, Boardman et al. 2008). First, an environmental cause may trigger a genetic difference in an outcome for which little or no genetic 112 Freese · Shostak A nn u. R ev . S oc io l. 20 09 .3 5. D ow nl oa de d fr om a rj ou rn al s. an nu al re vi ew s. or g by B R A N D E IS U N IV E R SI T Y o n 07 /0 2/ 09 . F or p er so na l u se o nl y. ANRV381-SO35-06 ARI 23 March 2009 16:53 difference would be observed in the absence of the cause. In the most famous study of geneenvironment interaction in recent behavioral science, Caspi et al. (2003) find that a genetic variant (MAOA) is associated with antisocial behavior in boys, but only among those who experienced abuse in childhood. Second, a positive environmental cause may compensate for a genetic liability, resulting in a more positive outcome than would otherwise be observed for those with the liability (and thus less effect of genetic difference overall). Shanahan et al. (2008) find that a dopamine-related genetic variant (DRD2) has a substantial relationship with school continuation, but this relationship is much reduced among those with high social capital. Similarly, Pescosolido et al. (2008) find that family support reduces the association between a genetic variant (GABRA2) and alcoholism. Third, social norms and limitations of opportunity can inhibit the relevance of psychological variation for outcomes and thus reduce the effect of genetic differences. Guo et al. (2008) offer a social control interpretation to the finding that a relationship between a DRD2 variant on delinquent behavior is not observed for respondents who have daily meals with their parents. Fourth, positive social contexts can increase genetic differences by providing disproportionate benefit to those whose genetic endowment already dispose them toward a more favorable outcome. As already mentioned, several studies find that genetic differences matter most for cognitive skills in advantaged environments. Studies of gene-environment interaction will undoubtedly be central to future social science study of the environment, but challenges to positively identifying interaction temper enthusiasm for single studies. Entirely apart from genetics, interaction effects are notorious for replication failure (Benjamin et al. 2007). In gene-environment interactions, the possible genes and environments to be tested can multiply quickly. Additionally, some genetic variants may be rare, and observing a sizable effect in a rare subgroup relative to a much larger subgroup may prompt especial concern about replicability. Guo et al. (2008) find an interaction in which the relationship between the proportion of sexually active teens in the respondent’s high school and the respondent’s number of sexual partners is moderated by a dopaminerelated genetic variant (DAT1). While certainly intriguing, the genetic variant found to be more responsive to the environment was observed for only 5% of the 674 individuals in their study. Moffitt et al. (2005, p. 478) emphasize the importance of theoretical justification for a candidate gene-environment interaction, and even then they extol the “wisdom of awaiting the meta-analysis, while not overreacting to any single study.” Unfortunately, awaiting the meta-analysis is not easy to do in social science, where multiple studies on independent samples do not accumulate with the speed that they do in medicine. Beyond this, what constitutes a statistical interaction is more conceptually complicated than is commonly recognized within sociology, especially when guidance from theoretical models is vague at best (Ai & Norton 2003, Campbell et al. 2005). Additionally, gene-environment interaction is sometimes used to characterize what is better termed gene-environment interdependence, encompassing not just interactions but also gene-environment correlations. In geneenvironment interaction, genetic differences moderate effects for individuals who experience the same environments; in gene-environment correlation, genetic differences cause individuals to experience different environments. A poster child of gene-environment correlation would be a tall girl whose height prompts encouragement to play sports and who consequently ends up with more athletic skills than she would otherwise. The signal contribution of sociologists to studies of gene-environment interdependence may be to elaborate how environment is understood and measured (Perrin & Lee 2007). Shanahan et al. (2007, 2008) employ methods borrowed from comparative-historical analysis to evaluate how the effect of genetic differences on school continuation may be moderated by configurations of multiple indicators of www.annualreviews.org • Genetics and Social Inquiry 113 A nn u. R ev . S oc io l. 20 09 .3 5. D ow nl oa de d fr om a rj ou rn al s. an nu al re vi ew s. or g by B R A N D E IS U N IV E R SI T Y o n 07 /0 2/ 09 . F or p er so na l u se o nl y. ANRV381-SO35-06 ARI 23 March 2009 16:53 social capital. More than this, behavioral genetics sorely needs a more sociological imagination of how the experience of environmental causes is patterned and often self-reinforcing over time. Elaboration of work on cumulative disadvantage and other tools from life course research can further our understanding of how selection into different environments widen initial individual differences, so that genetic and environmental causes may often be mutually reinforcing (Dannefer 2003, DiPrete & Eirich 2006). Similarly, work on fundamental causality in medical sociology can increase appreciation of how basic conditions like SES manifest as a very broad array of specific environmental causes (Link & Phelan 1995, Freese & Lutfey 2009). This work also highlights how advances in knowledge often benefit different groups differently, underscoring the potential for technologies like genetic testing and therapies to affect health disparities. Theories of culture and symbolic interaction can help articulate how the effects of environments are contingent on the meanings attached to them, implying among other things that improving measures of environment requires improving measurement of their subjective as well as objective aspects (Schulz & Lempert 2004, Kaufman 2004). In sum, the sociological imagination seems to have much to contribute to improving the environment side of the burgeoning literature on gene-environment interdependence, but to do so, sociologists must be willing to give genetics serious attention. GENETICS AND THE STUDY OF ITS SOCIAL IMPLICATIONS Human genetics has been centrally concerned with understanding how genes work as causes of development and of disease. As already noted, however, genetic advances are not passive, but rather new knowledge claims diffuse, serve as warrants for individual action, and potentially transform social arrangements. Sociologists have done at least as much work trying to identify the consequences of genetic science for society as they have trying to make use of genetic information in their own studies. This work has expressed much concern for the possibility of genetic science remaking the public consciousness through a process of geneticization. As we discuss below, writing on geneticization has to date not made much use of studies of beliefs and attitudes regarding genetics among the general populace. In addition, work on geneticization has also been only loosely associated with a burgeoning separate literature on biosociality that emphasizes new collective identities based on genetic or other biological information.