Everything is heritable & shared environment largely irrelevant*; "Not by Twins Alone: Using the Extended Family Design to Investigate Genetic Influence on Political Beliefs", Hatemi et al 2010; excerpts:
* this is one of the most surprising things in all of psychology that I know of - how routinely in twin studies, when measuring anything and everything, 'shared environment' typically has tiny values
"Variance components estimates of political and social attitudes suggest a substantial level of genetic influence, but the results have been challenged because they rely on data from twins only. In this analysis, we include responses from parents and nontwin full siblings of twins, account for measurement error by using a panel design, and estimate genetic and environmental variance by maximum-likelihood structural equation modeling. By doing so, we address the central concerns of critics, including that the twin-only design offers no verification of either the equal environments or random mating assumptions. Moving beyond the twin-only design leads to the conclusion that for most political and social attitudes, genetic influences account for an even greater proportion of individual differences than reported by studies using more limited data and more elementary estimation techniques. These findings make it increasingly difficult to deny that—however indirectly—genetics plays a role in the formation of political and social attitudes.
The most extensive early studies on attitudes were Martin et al. (1986) and Eaves, Eysenck, and Martin (1989). Using large data sets drawn from twins in Australia and the United States, respectively, these scholars reported results that “undermine the na ̈ıve assumption that the resemblance of family members can be interpreted in purely social terms” (Martin et al. 1986, 4368). Even though later analyses, often with different data and in different countries (see Bouchard and McGue 2003; Eaves et al. 1999; Olson et al. 2001), produced similar results, political scientists took virtually no notice of these provocative findings, perhaps because political scientists typically assume attitudes are entirely the product of environmental forces such as parental socialization and do not take seriously the possibility that genes could be involved (for exceptions, see Merelman 1971; Peterson 1983; Segal and Spaeth 1993, 234; Zaller 1992, 23). Then in 2005, Alford, Funk, and Hibbing performed additional analyses on the same combined data set collected and employed by Eaves et al. (1999) and presented the results to the political science community. Similar to earlier results, the findings suggested a surprising degree of genetic influence for political attitudes but suggested that genetics may play less of a role in the direction of party identification.
Regardless, since 2005 interest in the heritability of political variables has increased. Hatemi et al. (2007) found that vote choice is heritable but that the majority of the genetic influence on vote choice appears to be accounted for not directly but indirectly through the heritability of political attitudes. Fowler, Baker, and Dawes (2008) made the important discovery that actual voter turnout (not self-reported turnout) also is influenced substantially by genes. Further studies found that strength of affiliation with a party (regardless of the particular party involved) is strongly heritable (Hatemi, Alford et al. 2009). The findings that genes appear to have only a modest effect on direction of party identification, mostly an indirect effect on vote choice, and a stronger effect on strength of group affiliation could make sense given that party identification and voting for specific candidates are time-bound phenomena whereas the tendency toward group attachment (regardless of the nature of the group) may run deeper...There is likely no direct genetic basis for whether or not to build a wall on the Mexican border, but there might be genes that indirectly shape perception of outgroups, sensitivity to external threat, and preference for ingroup cohesion (for evidence that this may be the case, see Oxley et al. 2008).
univariate twin-only analyses provide little opportunity to detect violations of either the equal environments assumption or the random mating assumptions. The third shortcoming is the problem of estimating and correcting for measurement error, a perennial concern for all empirical investigations but a particular concern in the CTD because measurement error creates an upward bias on estimates of the impact of unshared environment (Fisher 1918)—and in models that correct for mate assortation, as we do here, this concern is compounded (Eaves and Hatemi 2008). In this article, we address these shortcomings by employing a nuclear family design, which includes parents and nontwin siblings, as well as test-retest measures of each of the traits of interest. By applying improved methodological procedures to a valuable data set, we provide a more accurate estimate of the influence on political attitudes of genetics, shared environment, and unshared environment.
The data we utilize were also the basis for the Eaves et al. (1999), Alford, Funk, and Hibbing (2005), and Hatemi, Medland, and Eaves (2009) twin-only studies, a data set originally known as the “Virginia 30,000” or “VA30K” for short (for information on the structure of the sample and ascertainment procedures, see Lake et al. 2000; Maes, Neale, and Eaves 1997; Truett et al. 1994). The approximately 30,000 adult subjects (aged 18–84 years) were twins (N = 14,781), spouses (N = 4,391), parents (N = 2,360), relatives (N = 195), offspring (N = 4,800), and nontwin siblings of twins (N = 3,184). The inclusion of nontwin relatives is especially helpful in identifying the multiple sources of biological and cultural inheritance (Heath et al. 1985).
The social and political attitude measures were included in a 28-item contemporary attitude battery gathered as part of a larger “Health and Life Styles” inventory conducted in 1986. Item format was the same as the Wilson-Patterson Attitude Index (Wilson and Patterson 1968), where attitude measurement is simplified by presenting each item in a one- or two-word format. Respondents are instructed to answer with the first reaction that comes to mind: “agree,” “uncertain,” or “disagree.” Data were collected by mail, with mail and telephone followup of nonrespondents when needed. Approximately two years later, the same attitude items were included in a follow-up questionnaire mailed to twins aged 50+ years, providing measures of attitude stability for 1,019 men and 2,912 women. In the remainder of this article, we ply the two-wave extended family portions of these data in order to better explore the nature and transmission of political attitudes.
Over the last 30 years, a variety of methods in psychology, psychiatry, and genetics have been used to verify that MZ and DZ pairs are not unequally influenced by different environments for a wide array of behavioral traits (for a review, see Medland and Hatemi 2009). These methods include comparing the twin trait similarity for blood-determined zygosity and for family-perceived zygosity among those twins for whom genetic zygosity is misperceived by family members (blood-determined zygosity is consistently found to be the better predictor—see Matheny, Wilson, and Dolan 1976; Plomin, Willerman, and Loehlin 1976; Scarr and Carter-Saltzman 1979); observing twin treatment by family members and others to examine differences in behaviors toward the different twin types (Lytton 1977); measuring specific environmental indicators for each twin and modeling differences in environment for the trait of interest while controlling for actual zygosity (Kendler et al. 1987; Heath, Jardine, and Martin 1989); extending the CTD by partitioning the shared environment into the overall common environment, C residual , which is completely correlated for all twin pairs, and that which is influenced by the perceived zygosity, C specific , (Hettema, Neale, and Kendler 1995; Kendler et al. 1993; Xian et al. 2000); and utilizing actual genetic similarity, known as identity by descent (IBD), rather than assuming that DZ twins or full siblings share on average 50% of their segregating genes. Regarding this last method, Visscher et al. (2006) obtained exact measures of genetic sharing of sibling pairs, and excluded MZ twins, thus removing any equal environmental concerns, and found that the heritability estimate for height was very similar to that derived from traditional CTD analyses. Perhaps of most relevance to questions about the EEA is recent work by Hatemi, Funk et al. (2009). Utilizing a longitudinal panel study of adolescent twins (aged 8–18) to assess political attitudes every two years, they found that there was no difference in MZ/DZ twin pair similarity throughout adolescence but that twin pair differences in political attitudes emerged later, when twins had departed from the parental nest. Thus, in order for it to be believed that a violation of the EEA is responsible for the heritability estimates previously reported, it would be necessary to argue that a special MZ twin environment for political attitudes exists in adolescence but remains dormant until adulthood, when it is triggered by some unidentified mechanism that then shapes adult preferences.
In light of these findings, a substantial amount of evidence runs against the existence of a special twin environment for political beliefs. Still, since doubts continue, we adopt an alternative strategy here. Directly testing for potential differences between MZ and DZ pair environments and for the method by which these differences might influence the trait for each zygosity type requires specific common environmental measures not typically available. Such direct tests include analysis of MZ twins reared apart and adoption studies, but these approaches have their own problems (see Medland and Hatemi 2009). Our approach here is to include data on nontwin siblings, thereby allowing the model to partition variance separately for siblings generally and for twin siblings specifically. If the more similar treatment of MZ twins were indeed influencing relevant (i.e., political) trait values, then the more similar treatment of DZ twins relative to nontwin full siblings should also affect that trait. The degree of genetic similarity of DZ twins and full siblings is the same, so after correcting for fixed effects (e.g., age), differences between twins and nontwin siblings provide an indirect estimate of twin-specific environmental effects. In simple terms, while we cannot identify specific EEA violations, we can identify the total amount of variance attributable to twinspecific environmental effects.
The addition of parents and nontwin siblings to the analysis also allows for a more extended exploration of the extent to which some part of the genetic variation is nonadditive. Nonadditive genetic influences arise from interactions either within a gene (known as dominance) or between genes (known as epistasis; Neale et al. 2003). Typically, twin models focus on the additive estimate because the combined effect of all genes can be estimated with more confidence than models which partition out nonadditive influences. This limitation is important when diagnostics suggest nonadditive effects are present. Diagnostics for detection of nonadditive influences are most often performed by comparing the MZ and DZ twin correlations. If the MZ pair correlation is significantly more than twice as large as the DZ correlations, nonadditive influences are likely to be important (Neale et al. 2003). The correlations presented in Table 1 show that for only three of the 28 items (for females) and seven of the 28 items (for males)—a total of only 10 out of a possible 56—is there any suggestion of nonadditive effects. For these 10, the DZ correlations are only slightly below half of the MZ correlations; thus preliminary analyses give little cause to suspect significant nonadditive effects.
A more developed assessment of the presence of nonadditive effects is made possible by the inclusion in our data set of nontwin family members. Additive genetic effects typically produce trait correlations that are at similar levels for DZ twins, nontwin siblings, and parentoffspring pairs, and that, for all three of these relationship categories, average at least half the size of the MZ twin pair correlations. When nonadditive genetic effects dominate, the MZ twin correlations will remain robust, but all three of the other family pairs will exhibit much reduced similarity. This distinction gives rise to the readily apparent family history of traits that exhibit “narrow sense” heritability (i.e., heritability that “runs in the family” and that characterizes simple additive genetic effects) in contrast to traits that exhibit only “broad sense” heritability (i.e., traits that show little clear clustering in families despite the fact that they may have equally strong, if more complex, genetic underpinnings).
As it turns out, extended twin-family studies of personality provide clear evidence of nonadditive genetic effects (Keller et al. 2005), but our Table 1 provides no evidence of this pattern for political temperaments. At least for the Wilson-Patterson items examined here, trait correlations are very similar across same-sex DZ twins, nontwin siblings, and parent-offspring pairs, and for all three of these relationship categories, average correlations are approximately half the size of the MZ twin pair correlations. Political temperament as measured here appears to exhibit narrow sense heritability, in clear contrast to the broad sense heritability that characterizes personality traits.
An answer can be found by looking at the interspouse correlations for the mate pairs in the VA30K study. This survey was completed by the spouses of 4,387 twins as well as by 773 mate pairs with twins as offspring—a total of 5,160 spousal pairs, making it ideal for inspecting interspousal correlations. Table 2 consists primarily of the 28 Wilson-Patterson Inventory items and also, at the top, an overall additive index of “liberal/conservative” responses to these 28 items, but for purposes of comparison we also include results from four nonpolitical variables contained in the data set: extraversion and neuroticism (as measured by items in the Eysenck Personality Quotient), plus height and weight.
Correlations for political attitudes far outstrip those for physical and personality traits. Extraverts are as likely to marry introverts as other extraverts, and the interspousal correlation for neuroticism is not much larger. The correlations for height and weight of spouses are positive and statistically significant but small, suggesting that taller and heavier individuals do indeed have spouses who tend to be tall and heavy but that this pattern is often violated. In direct contrast, attitudes on political and social items are quite likely to be shared by mate pairs. The correlation for the overall index of attitudes is extremely high (.647) and inspection of the individual items indicates why. Though the correlations for some of the less salient items, such as military drill, modern art, federal housing, and censorship, are modest, most others are substantial and, as was the case in Table 1, the correlations for hot-button issues such as school prayer, abortion, gay rights, and living together are very high.
...Of course, some of this interspouse similarity could be the result of assimilation over the course of a relationship or to social homogamy (the tendency of people to mate with those around them). However, Martin et al. (1986) find that the correlation between mates is due primarily to assortation and not to convergence. With regard to most attitudes, spouses do not become more similar with the passage of the years.
Measurement error is always a concern but especially with survey items for which respondents frequently provide answers that do not reflect their true feelings (see Converse 1964; Zaller 1992). When respondents change their answers to the same item, suspicion grows that researchers are picking up noise or error. Error of this sort may create a particular problem for variance components modeling because standard methodological procedures push the error term into estimates for the unshared (unique) environment, thus inflating the apparent importance of idiosyncratic environmental events at the expense of estimates of the importance of both the shared environment and additive genetic influences.
Repeated measures offer the difference between “reliable variance” and a measure at “one point in time.” To take one example, if spousal concordance exists for a “political” phenotype (see previous section), it might be expected that concordance is due to long-term political similarity, rather than any error-prone single assessment. Repeated measures offer one approach to estimating and, thus, controlling for such short-term fluctuations, thereby making it possible to correct estimates (Eaves 1973). Not accounting for this error may affect conclusions concerning the relative importance of the primary shapers of attitudes, thereby leading to erroneous interpretations. The VA30K data set provides a solution to this problem as well. In addition to including thousands of nontwin respondents, portions of the instrument were administered again, approximately two years later (note the contrast with typical procedures that repeat items just weeks, days, or even minutes apart), to nearly 4,000 of the initial respondents. These two separate soundings make it possible to correct for response instability, thereby affording more accurate estimates of the relative influence of additive genetic, as well as shared and unshared environmental influences...In the last two columns of this table, the testretest coefficients are presented, first for males and then for females. As can be seen, these numbers are quite low on salient items such as school prayer, abortion, the death penalty, and gay rights, but the “measurement error” is much higher precisely for those less salient responses for which sentiments could reasonably be expected to vary from one time to the next: property taxes, federal housing, military drill, pacifism, and censorship. Measuring and accounting for these differential levels of test-retest correlation greatly improves the accuracy of the estimates produced by the extended twin family analysis we are about to undertake.
The model presented in Figure 2 allows for (1) additive genetic influences for males and females (h m and h f ) on the latent constructs that represent opinions on each of the individual attitudes (Johannsen 1911); (2) environmental effects not shared by twins or siblings, (unique environment) e m and e f ; (3) environmental effects shared by male and female siblings and DZ twins but not transmitted from parents (common environment) c m and c f ; (4) additional environmental similarity between twins (MZ and DZ) because twin environments often correlate more highly than siblings, t m and t f ; (5) direct social transmission (“vertical cultural inheritance”; Cavalli-Sforza and Feldman 1981) from mothers and fathers to their sons and daughters (u m , u f , v m, and v f ); and (6) phenotypic assortment between spouses m (correlation between mates—“assortative mating”). In addition, the nuclear family model contains two parameters corresponding to the correlations between the genotypes and phenotypes of both parents individually (r gm and r gf ). Under the assumption that the model parameters are stable over generations, these can be expressed as functions of the other parameters of intergenerational transmission. Assuming that genetic effects are additive, the paths from parental to offspring genetic influence are fixed at 0.5 (Jencks et al. 1972; Morton 1974). The path for nontwin siblings is obtained by allowing the effects contributing to the twin-specific environment (T) to be uncorrelated in siblings. A recent study by Hatemi, Medland, and Eaves (2009) found significant quantitative (and in some cases qualitative) sex difference in the variance components analyses for political attitudes. Thus, in the model used here males and females are not equated, but rather are estimated independently within the same model. Due to the already complex nature of the model, we do not correct for genetic and environmental influences that may vary with age.
The central finding of the table is that heritability estimates for political and social attitudes persist even when extended family data rather than twinonly data are used, when maximum-likelihood estimates rather than simple polychoric correlation transformations are employed, when mate assortation is acknowledged, and when repeated soundings are included for reliability. A quick scan down the two columns reporting additive genetic influences (one for males and one for females) indicates heritability consistently in the .3 to .7 range. The individual attitudes showing the largest additive genetic influences appear to be those directed toward school prayer and X-rated movies, with heritability being responsible for roughly two-thirds of the variation in these particular attitudes (for males and for females). “Living together” is also strongly heritable but illustrates the fact that sometimes additive genetic forces are quite different for males and females. The additive genetic term is .40 for males but .84 for females. Attitudes toward gay rights and immigration are also among the items showing the highest degrees of heritability, so it would appear the issues widely perceived to be hot-button social issues are the very issues that tend to be strongly heritable, just as Tesser (1993) predicted.
In addition, the design employed here is able to provide distinct estimates of vertical cultural inheritance. As can be seen, for both males and females, cultural “inheritance” is minimal, never over .11 and usually under .05. Consistent with earlier findings, party identification appears to be an exception. Here we find that vertical transmission influences are similar to those of the shared environment, but still less than the unique environment and cultural effects that come from siblings, especially co-twins.
Table 3 also includes estimates for a composite attitude index labeled Liberalism-Conservatism (made up of all the items in the Wilson-Patterson Inventory) as well as for party identification. For the overall index of Liberalism-Conservatism, genetics accounts for approximately .34 of the variance in females and over half (.58) of the variance in males, while twin-specific environment and vertical cultural transmission (parental influence) account for less—.16 in females and just .03 in males. The shared environment is inconsequential. Turning to party identification, in previous analyses, party identification exhibited only modest or insignificant genetic influence and notable common environmental effects (Hatemi, Alford et al. 2009).
Table 4 provides the model fits for the twin-specific environment of the 56 independent tests (28 items for males and females independently). For males only, two traits (living together and busing) are significantly different from zero. For females, just four items (Democrats, nuclear power, capitalism, and party identification) have twin-specific effects that are statistically significant. Furthermore, with the exception of living together, divorce, and busing in males, the twin-specific environment can be dropped from the models without harming model fit (and, in fact, improving parsimony), and for females the twin-specific environment can be dropped from the model without affecting model fit for all variables except nuclear power and party identification. Thus, while, as noted above, the inclusion of nontwin siblings does not directly test specific MZ and DZ twin environments, only six of the 56 tests show twin-specific environmental effects that reach statistical significance. The charge raised by critics of earlier estimations—i.e., that EEA violations could in fact be responsible for most if not all of the impact that was attributed to genetic inheritance—is disconfirmed by the results reported here.
All told, previous claims that additive genetic influences account for at least 40% of the variance in political and social attitudes hold up even when more sophisticated modeling techniques are employed on data from family members other than just twin pairs. Modeling twin-specific environments (by including nontwin siblings) may diminish heritability estimates a bit, but correcting for assortative mating (by including parents) increases heritability estimates. Moreover, eliminating the variation attributable to measurement error (by including test-retest assessments of political phenotypes) ensures a more accurate measure of attitudes. Furthermore, we continue to see relatively weak contributions from nongenetic parental effects (socialization), with unique environmental effects being somewhere between genes and shared environment in importance."
* this is one of the most surprising things in all of psychology that I know of - how routinely in twin studies, when measuring anything and everything, 'shared environment' typically has tiny values
"Variance components estimates of political and social attitudes suggest a substantial level of genetic influence, but the results have been challenged because they rely on data from twins only. In this analysis, we include responses from parents and nontwin full siblings of twins, account for measurement error by using a panel design, and estimate genetic and environmental variance by maximum-likelihood structural equation modeling. By doing so, we address the central concerns of critics, including that the twin-only design offers no verification of either the equal environments or random mating assumptions. Moving beyond the twin-only design leads to the conclusion that for most political and social attitudes, genetic influences account for an even greater proportion of individual differences than reported by studies using more limited data and more elementary estimation techniques. These findings make it increasingly difficult to deny that—however indirectly—genetics plays a role in the formation of political and social attitudes.
The most extensive early studies on attitudes were Martin et al. (1986) and Eaves, Eysenck, and Martin (1989). Using large data sets drawn from twins in Australia and the United States, respectively, these scholars reported results that “undermine the na ̈ıve assumption that the resemblance of family members can be interpreted in purely social terms” (Martin et al. 1986, 4368). Even though later analyses, often with different data and in different countries (see Bouchard and McGue 2003; Eaves et al. 1999; Olson et al. 2001), produced similar results, political scientists took virtually no notice of these provocative findings, perhaps because political scientists typically assume attitudes are entirely the product of environmental forces such as parental socialization and do not take seriously the possibility that genes could be involved (for exceptions, see Merelman 1971; Peterson 1983; Segal and Spaeth 1993, 234; Zaller 1992, 23). Then in 2005, Alford, Funk, and Hibbing performed additional analyses on the same combined data set collected and employed by Eaves et al. (1999) and presented the results to the political science community. Similar to earlier results, the findings suggested a surprising degree of genetic influence for political attitudes but suggested that genetics may play less of a role in the direction of party identification.
Regardless, since 2005 interest in the heritability of political variables has increased. Hatemi et al. (2007) found that vote choice is heritable but that the majority of the genetic influence on vote choice appears to be accounted for not directly but indirectly through the heritability of political attitudes. Fowler, Baker, and Dawes (2008) made the important discovery that actual voter turnout (not self-reported turnout) also is influenced substantially by genes. Further studies found that strength of affiliation with a party (regardless of the particular party involved) is strongly heritable (Hatemi, Alford et al. 2009). The findings that genes appear to have only a modest effect on direction of party identification, mostly an indirect effect on vote choice, and a stronger effect on strength of group affiliation could make sense given that party identification and voting for specific candidates are time-bound phenomena whereas the tendency toward group attachment (regardless of the nature of the group) may run deeper...There is likely no direct genetic basis for whether or not to build a wall on the Mexican border, but there might be genes that indirectly shape perception of outgroups, sensitivity to external threat, and preference for ingroup cohesion (for evidence that this may be the case, see Oxley et al. 2008).
univariate twin-only analyses provide little opportunity to detect violations of either the equal environments assumption or the random mating assumptions. The third shortcoming is the problem of estimating and correcting for measurement error, a perennial concern for all empirical investigations but a particular concern in the CTD because measurement error creates an upward bias on estimates of the impact of unshared environment (Fisher 1918)—and in models that correct for mate assortation, as we do here, this concern is compounded (Eaves and Hatemi 2008). In this article, we address these shortcomings by employing a nuclear family design, which includes parents and nontwin siblings, as well as test-retest measures of each of the traits of interest. By applying improved methodological procedures to a valuable data set, we provide a more accurate estimate of the influence on political attitudes of genetics, shared environment, and unshared environment.
The data we utilize were also the basis for the Eaves et al. (1999), Alford, Funk, and Hibbing (2005), and Hatemi, Medland, and Eaves (2009) twin-only studies, a data set originally known as the “Virginia 30,000” or “VA30K” for short (for information on the structure of the sample and ascertainment procedures, see Lake et al. 2000; Maes, Neale, and Eaves 1997; Truett et al. 1994). The approximately 30,000 adult subjects (aged 18–84 years) were twins (N = 14,781), spouses (N = 4,391), parents (N = 2,360), relatives (N = 195), offspring (N = 4,800), and nontwin siblings of twins (N = 3,184). The inclusion of nontwin relatives is especially helpful in identifying the multiple sources of biological and cultural inheritance (Heath et al. 1985).
The social and political attitude measures were included in a 28-item contemporary attitude battery gathered as part of a larger “Health and Life Styles” inventory conducted in 1986. Item format was the same as the Wilson-Patterson Attitude Index (Wilson and Patterson 1968), where attitude measurement is simplified by presenting each item in a one- or two-word format. Respondents are instructed to answer with the first reaction that comes to mind: “agree,” “uncertain,” or “disagree.” Data were collected by mail, with mail and telephone followup of nonrespondents when needed. Approximately two years later, the same attitude items were included in a follow-up questionnaire mailed to twins aged 50+ years, providing measures of attitude stability for 1,019 men and 2,912 women. In the remainder of this article, we ply the two-wave extended family portions of these data in order to better explore the nature and transmission of political attitudes.
Over the last 30 years, a variety of methods in psychology, psychiatry, and genetics have been used to verify that MZ and DZ pairs are not unequally influenced by different environments for a wide array of behavioral traits (for a review, see Medland and Hatemi 2009). These methods include comparing the twin trait similarity for blood-determined zygosity and for family-perceived zygosity among those twins for whom genetic zygosity is misperceived by family members (blood-determined zygosity is consistently found to be the better predictor—see Matheny, Wilson, and Dolan 1976; Plomin, Willerman, and Loehlin 1976; Scarr and Carter-Saltzman 1979); observing twin treatment by family members and others to examine differences in behaviors toward the different twin types (Lytton 1977); measuring specific environmental indicators for each twin and modeling differences in environment for the trait of interest while controlling for actual zygosity (Kendler et al. 1987; Heath, Jardine, and Martin 1989); extending the CTD by partitioning the shared environment into the overall common environment, C residual , which is completely correlated for all twin pairs, and that which is influenced by the perceived zygosity, C specific , (Hettema, Neale, and Kendler 1995; Kendler et al. 1993; Xian et al. 2000); and utilizing actual genetic similarity, known as identity by descent (IBD), rather than assuming that DZ twins or full siblings share on average 50% of their segregating genes. Regarding this last method, Visscher et al. (2006) obtained exact measures of genetic sharing of sibling pairs, and excluded MZ twins, thus removing any equal environmental concerns, and found that the heritability estimate for height was very similar to that derived from traditional CTD analyses. Perhaps of most relevance to questions about the EEA is recent work by Hatemi, Funk et al. (2009). Utilizing a longitudinal panel study of adolescent twins (aged 8–18) to assess political attitudes every two years, they found that there was no difference in MZ/DZ twin pair similarity throughout adolescence but that twin pair differences in political attitudes emerged later, when twins had departed from the parental nest. Thus, in order for it to be believed that a violation of the EEA is responsible for the heritability estimates previously reported, it would be necessary to argue that a special MZ twin environment for political attitudes exists in adolescence but remains dormant until adulthood, when it is triggered by some unidentified mechanism that then shapes adult preferences.
In light of these findings, a substantial amount of evidence runs against the existence of a special twin environment for political beliefs. Still, since doubts continue, we adopt an alternative strategy here. Directly testing for potential differences between MZ and DZ pair environments and for the method by which these differences might influence the trait for each zygosity type requires specific common environmental measures not typically available. Such direct tests include analysis of MZ twins reared apart and adoption studies, but these approaches have their own problems (see Medland and Hatemi 2009). Our approach here is to include data on nontwin siblings, thereby allowing the model to partition variance separately for siblings generally and for twin siblings specifically. If the more similar treatment of MZ twins were indeed influencing relevant (i.e., political) trait values, then the more similar treatment of DZ twins relative to nontwin full siblings should also affect that trait. The degree of genetic similarity of DZ twins and full siblings is the same, so after correcting for fixed effects (e.g., age), differences between twins and nontwin siblings provide an indirect estimate of twin-specific environmental effects. In simple terms, while we cannot identify specific EEA violations, we can identify the total amount of variance attributable to twinspecific environmental effects.
The addition of parents and nontwin siblings to the analysis also allows for a more extended exploration of the extent to which some part of the genetic variation is nonadditive. Nonadditive genetic influences arise from interactions either within a gene (known as dominance) or between genes (known as epistasis; Neale et al. 2003). Typically, twin models focus on the additive estimate because the combined effect of all genes can be estimated with more confidence than models which partition out nonadditive influences. This limitation is important when diagnostics suggest nonadditive effects are present. Diagnostics for detection of nonadditive influences are most often performed by comparing the MZ and DZ twin correlations. If the MZ pair correlation is significantly more than twice as large as the DZ correlations, nonadditive influences are likely to be important (Neale et al. 2003). The correlations presented in Table 1 show that for only three of the 28 items (for females) and seven of the 28 items (for males)—a total of only 10 out of a possible 56—is there any suggestion of nonadditive effects. For these 10, the DZ correlations are only slightly below half of the MZ correlations; thus preliminary analyses give little cause to suspect significant nonadditive effects.
A more developed assessment of the presence of nonadditive effects is made possible by the inclusion in our data set of nontwin family members. Additive genetic effects typically produce trait correlations that are at similar levels for DZ twins, nontwin siblings, and parentoffspring pairs, and that, for all three of these relationship categories, average at least half the size of the MZ twin pair correlations. When nonadditive genetic effects dominate, the MZ twin correlations will remain robust, but all three of the other family pairs will exhibit much reduced similarity. This distinction gives rise to the readily apparent family history of traits that exhibit “narrow sense” heritability (i.e., heritability that “runs in the family” and that characterizes simple additive genetic effects) in contrast to traits that exhibit only “broad sense” heritability (i.e., traits that show little clear clustering in families despite the fact that they may have equally strong, if more complex, genetic underpinnings).
As it turns out, extended twin-family studies of personality provide clear evidence of nonadditive genetic effects (Keller et al. 2005), but our Table 1 provides no evidence of this pattern for political temperaments. At least for the Wilson-Patterson items examined here, trait correlations are very similar across same-sex DZ twins, nontwin siblings, and parent-offspring pairs, and for all three of these relationship categories, average correlations are approximately half the size of the MZ twin pair correlations. Political temperament as measured here appears to exhibit narrow sense heritability, in clear contrast to the broad sense heritability that characterizes personality traits.
An answer can be found by looking at the interspouse correlations for the mate pairs in the VA30K study. This survey was completed by the spouses of 4,387 twins as well as by 773 mate pairs with twins as offspring—a total of 5,160 spousal pairs, making it ideal for inspecting interspousal correlations. Table 2 consists primarily of the 28 Wilson-Patterson Inventory items and also, at the top, an overall additive index of “liberal/conservative” responses to these 28 items, but for purposes of comparison we also include results from four nonpolitical variables contained in the data set: extraversion and neuroticism (as measured by items in the Eysenck Personality Quotient), plus height and weight.
Correlations for political attitudes far outstrip those for physical and personality traits. Extraverts are as likely to marry introverts as other extraverts, and the interspousal correlation for neuroticism is not much larger. The correlations for height and weight of spouses are positive and statistically significant but small, suggesting that taller and heavier individuals do indeed have spouses who tend to be tall and heavy but that this pattern is often violated. In direct contrast, attitudes on political and social items are quite likely to be shared by mate pairs. The correlation for the overall index of attitudes is extremely high (.647) and inspection of the individual items indicates why. Though the correlations for some of the less salient items, such as military drill, modern art, federal housing, and censorship, are modest, most others are substantial and, as was the case in Table 1, the correlations for hot-button issues such as school prayer, abortion, gay rights, and living together are very high.
...Of course, some of this interspouse similarity could be the result of assimilation over the course of a relationship or to social homogamy (the tendency of people to mate with those around them). However, Martin et al. (1986) find that the correlation between mates is due primarily to assortation and not to convergence. With regard to most attitudes, spouses do not become more similar with the passage of the years.
Measurement error is always a concern but especially with survey items for which respondents frequently provide answers that do not reflect their true feelings (see Converse 1964; Zaller 1992). When respondents change their answers to the same item, suspicion grows that researchers are picking up noise or error. Error of this sort may create a particular problem for variance components modeling because standard methodological procedures push the error term into estimates for the unshared (unique) environment, thus inflating the apparent importance of idiosyncratic environmental events at the expense of estimates of the importance of both the shared environment and additive genetic influences.
Repeated measures offer the difference between “reliable variance” and a measure at “one point in time.” To take one example, if spousal concordance exists for a “political” phenotype (see previous section), it might be expected that concordance is due to long-term political similarity, rather than any error-prone single assessment. Repeated measures offer one approach to estimating and, thus, controlling for such short-term fluctuations, thereby making it possible to correct estimates (Eaves 1973). Not accounting for this error may affect conclusions concerning the relative importance of the primary shapers of attitudes, thereby leading to erroneous interpretations. The VA30K data set provides a solution to this problem as well. In addition to including thousands of nontwin respondents, portions of the instrument were administered again, approximately two years later (note the contrast with typical procedures that repeat items just weeks, days, or even minutes apart), to nearly 4,000 of the initial respondents. These two separate soundings make it possible to correct for response instability, thereby affording more accurate estimates of the relative influence of additive genetic, as well as shared and unshared environmental influences...In the last two columns of this table, the testretest coefficients are presented, first for males and then for females. As can be seen, these numbers are quite low on salient items such as school prayer, abortion, the death penalty, and gay rights, but the “measurement error” is much higher precisely for those less salient responses for which sentiments could reasonably be expected to vary from one time to the next: property taxes, federal housing, military drill, pacifism, and censorship. Measuring and accounting for these differential levels of test-retest correlation greatly improves the accuracy of the estimates produced by the extended twin family analysis we are about to undertake.
The model presented in Figure 2 allows for (1) additive genetic influences for males and females (h m and h f ) on the latent constructs that represent opinions on each of the individual attitudes (Johannsen 1911); (2) environmental effects not shared by twins or siblings, (unique environment) e m and e f ; (3) environmental effects shared by male and female siblings and DZ twins but not transmitted from parents (common environment) c m and c f ; (4) additional environmental similarity between twins (MZ and DZ) because twin environments often correlate more highly than siblings, t m and t f ; (5) direct social transmission (“vertical cultural inheritance”; Cavalli-Sforza and Feldman 1981) from mothers and fathers to their sons and daughters (u m , u f , v m, and v f ); and (6) phenotypic assortment between spouses m (correlation between mates—“assortative mating”). In addition, the nuclear family model contains two parameters corresponding to the correlations between the genotypes and phenotypes of both parents individually (r gm and r gf ). Under the assumption that the model parameters are stable over generations, these can be expressed as functions of the other parameters of intergenerational transmission. Assuming that genetic effects are additive, the paths from parental to offspring genetic influence are fixed at 0.5 (Jencks et al. 1972; Morton 1974). The path for nontwin siblings is obtained by allowing the effects contributing to the twin-specific environment (T) to be uncorrelated in siblings. A recent study by Hatemi, Medland, and Eaves (2009) found significant quantitative (and in some cases qualitative) sex difference in the variance components analyses for political attitudes. Thus, in the model used here males and females are not equated, but rather are estimated independently within the same model. Due to the already complex nature of the model, we do not correct for genetic and environmental influences that may vary with age.
The central finding of the table is that heritability estimates for political and social attitudes persist even when extended family data rather than twinonly data are used, when maximum-likelihood estimates rather than simple polychoric correlation transformations are employed, when mate assortation is acknowledged, and when repeated soundings are included for reliability. A quick scan down the two columns reporting additive genetic influences (one for males and one for females) indicates heritability consistently in the .3 to .7 range. The individual attitudes showing the largest additive genetic influences appear to be those directed toward school prayer and X-rated movies, with heritability being responsible for roughly two-thirds of the variation in these particular attitudes (for males and for females). “Living together” is also strongly heritable but illustrates the fact that sometimes additive genetic forces are quite different for males and females. The additive genetic term is .40 for males but .84 for females. Attitudes toward gay rights and immigration are also among the items showing the highest degrees of heritability, so it would appear the issues widely perceived to be hot-button social issues are the very issues that tend to be strongly heritable, just as Tesser (1993) predicted.
In addition, the design employed here is able to provide distinct estimates of vertical cultural inheritance. As can be seen, for both males and females, cultural “inheritance” is minimal, never over .11 and usually under .05. Consistent with earlier findings, party identification appears to be an exception. Here we find that vertical transmission influences are similar to those of the shared environment, but still less than the unique environment and cultural effects that come from siblings, especially co-twins.
Table 3 also includes estimates for a composite attitude index labeled Liberalism-Conservatism (made up of all the items in the Wilson-Patterson Inventory) as well as for party identification. For the overall index of Liberalism-Conservatism, genetics accounts for approximately .34 of the variance in females and over half (.58) of the variance in males, while twin-specific environment and vertical cultural transmission (parental influence) account for less—.16 in females and just .03 in males. The shared environment is inconsequential. Turning to party identification, in previous analyses, party identification exhibited only modest or insignificant genetic influence and notable common environmental effects (Hatemi, Alford et al. 2009).
Table 4 provides the model fits for the twin-specific environment of the 56 independent tests (28 items for males and females independently). For males only, two traits (living together and busing) are significantly different from zero. For females, just four items (Democrats, nuclear power, capitalism, and party identification) have twin-specific effects that are statistically significant. Furthermore, with the exception of living together, divorce, and busing in males, the twin-specific environment can be dropped from the models without harming model fit (and, in fact, improving parsimony), and for females the twin-specific environment can be dropped from the model without affecting model fit for all variables except nuclear power and party identification. Thus, while, as noted above, the inclusion of nontwin siblings does not directly test specific MZ and DZ twin environments, only six of the 56 tests show twin-specific environmental effects that reach statistical significance. The charge raised by critics of earlier estimations—i.e., that EEA violations could in fact be responsible for most if not all of the impact that was attributed to genetic inheritance—is disconfirmed by the results reported here.
All told, previous claims that additive genetic influences account for at least 40% of the variance in political and social attitudes hold up even when more sophisticated modeling techniques are employed on data from family members other than just twin pairs. Modeling twin-specific environments (by including nontwin siblings) may diminish heritability estimates a bit, but correcting for assortative mating (by including parents) increases heritability estimates. Moreover, eliminating the variation attributable to measurement error (by including test-retest assessments of political phenotypes) ensures a more accurate measure of attitudes. Furthermore, we continue to see relatively weak contributions from nongenetic parental effects (socialization), with unique environmental effects being somewhere between genes and shared environment in importance."