Heritability and the Equal Environments Assumption: Evidence from Multiple Samples of Misclassified Twins
Total Page:16
File Type:pdf, Size:1020Kb
Behav Genet (2013) 43:415–426 DOI 10.1007/s10519-013-9602-1 ORIGINAL RESEARCH Heritability and the Equal Environments Assumption: Evidence from Multiple Samples of Misclassified Twins Dalton Conley • Emily Rauscher • Christopher Dawes • Patrik K. E. Magnusson • Mark L. Siegal Received: 1 August 2012 / Accepted: 16 July 2013 / Published online: 1 August 2013 Ó Springer Science+Business Media New York 2013 Abstract Classically derived estimates of heritability Adolescent Health (Add Health). Such twins were reared from twin models have been plagued by the possibility of under one ‘‘environmental regime of similarity’’ while genetic-environmental covariance. Survey questions that genetically belonging to another group, reversing the typ- attempt to measure directly the extent to which more ical GE-covariance and allowing bounded estimates of genetically similar kin (such as monozygotic twins) also heritability for a range of outcomes. In addition, we share more similar environmental conditions represent examine twins who were initially misclassified by survey poor attempts to gauge a complex underlying phenomenon assignment—a stricter standard—in three datasets: Add of GE-covariance. The present study exploits a natural Health, the Minnesota Twin Family Study and the Child experiment to address this issue: Self-misperception of and Adolescent Twin Study in Sweden. Results are similar twin zygosity in the National Longitudinal Survey of across approaches and datasets and largely support the validity of the equal environments assumption. Keywords Equal environments Á Twin Edited by Chandra Reynolds. misclassification Á Heritability Á ACE model Electronic supplementary material The online version of this article (doi:10.1007/s10519-013-9602-1) contains supplementary material, which is available to authorized users. Introduction D. Conley (&) Department of Sociology, New York University & NBER, Research has claimed to measure the heritabilities of a 6 Washington Square North Room 20, New York, wide variety of traits and behaviors, from height (Visscher NY 10003, USA et al. 2006) to autism (Liu et al. 2010) and even food e-mail: [email protected] preferences (Breen et al. 2006). Many estimates, however, E. Rauscher are based on twin pair analysis and therefore reliant on Department of Sociology, University of Kansas, Fraser Hall strong assumptions about the relative environmental simi- Room 716, 1415 Jayhawk Blvd, Lawrence, KS 66045, USA larity of identical (monozygotic, MZ) and fraternal (dizy- C. Dawes gotic, DZ) twins (the equal environments assumption that Wilf Family Department of Politics, New York University, identical and fraternal twins experience the same degree of 269 Mercer Street 3rd Floor, New York, NY 10003, USA environmental difference and/or influence each other’s outcomes to the same extent) (see, e.g., Plomin et al. 2001). P. K. E. Magnusson Department of Medical Epidemiology and Biostatistics, If society treats identical twins more similarly than fra- Karolinska Institutet, Stockholm, Sweden ternal twins, for example, the resulting unequal twin environments could cause traditional twin analyses to M. L. Siegal overestimate heritability. Department of Biology, Center for Genomics and Systems Biology, New York University, 12 Waverly Place, Although there are other approaches to estimating her- New York, NY 10003, USA itability in humans, twin comparisons are by far the most 123 416 Behav Genet (2013) 43:415–426 common approach and taken to be the least problematic shown that depending on the GE covariance assumed, because, being of a cohort together, both types of twins estimates of heritability can be driven wildly up or down. share uterine environments, experience societal events at While alternative heritability approaches are emerging, the same time and deal with family transitions also at the such as those that use sibling identity by descent (IBD) to same point in their development. In the most naı¨ve estimate phenotypic similarity (Visscher et al. 2006)or approach, narrow-sense (additive) genetic heritability (h2) those that use genetic covariance among non-related indi- is calculated as two times the difference between the intra- viduals (\2.5 % genetic relatedness) (Davies et al. 2011; class correlations of identical and fraternal twins. Narrow- Yang et al. 2010), questions about the EEA—and the sense heritability is often estimated using an ACE model, viability of traditional twin-based heritability estimates where A stands for additive genetic heritability, C for more broadly—remain. In the present study, we offer an common environment and E for unique environment approach to deal with GE covariance that relies on tradi- (essentially an error term). However, more recently, much tional twin methods. Exploiting variability in whether or more complex structural models have been offered to not the twins accurately perceived their own zygosity, we account for various complications such as the fact that—as putatively reverse the direction of social environmental a result of assortative mating1 at the parental level—fra- similarity (and confounding) that is typically present in ternal twins may share more than 50 % of their genes. twin studies relying on the EEA. In other words, we take Likewise, non-linear interactions between alleles—such as advantage of twins who believe they are fraternal when dominance—have been modeled in attempts to get at they are in fact identical (or vice versa) and assess whether broad-sense heritability (H2) (see Purcell 2002 for a review the degree of twin similarity differs from twins who of these models and simulation exercises and Purcell and accurately perceive their zygosity. Thus, we are able to Sham 2002 for an empirical example). Perhaps most replicate the standard ACE model—the workhorse of importantly, the ‘‘equal environments’’ assumption (EEA) behavioral genetics—and interrogate a key assumption of has been relaxed. The naı¨ve calculation mentioned above is the paradigm. based on the EEA. That is, it assumes that the covariance If heritability estimates and twin similarity are similar between environment and genetics is zero. Put another regardless of perceived zygosity, results would support the way, the simple estimation of heritability requires the EEA and lend credence to traditional twin ACE heritability rather heroic assumption that identical twins experience the strategies. If, on the other hand, results indicate a strong same degree of similarity in environment (including reci- relationship between perceived zygosity and heritability— procal effects on each other) as do (same sex) fraternal with lower heritability based on genetic zygosity—it would twins.2 cast doubt on traditional twin heritability strategies and The newer models include an estimate of the degree to suggest that many traits might be more socially malleable which environmental similarity varies with genetic like- than previous research based on such strategies would ness. However, these are just that: estimates—often based suggest. on questions about whether or not respondents were ‘‘dressed alike’’ growing up, whether they were viewed as similarly as ‘‘two peas in a pod’’ and so on (see, e.g., Misclassified twin research Lichtenstein et al. 1992; Rodgers et al. 1999; Rowe and Teachman 2001; Guo and Stearns 2002). Such questions We are not the first researchers to pursue this ‘‘misclassi- are likely to capture only some of the ways that environ- fication strategy’’ to interrogate heritability estimates. mental similarity differs across identical and fraternal twin Goodman and Stevenson (1989) used this methodology to pairs, which is troubling since Goldberger (1979) has disentangle genetic and environmental effects in a sample of 13-year-old British twins and estimated that hyperac- 1 Assortative mating is the non-random selection of mates in a tivity and attentiveness are about half heritable. They population. For example, brunettes may be more likely to pair with assigned ‘‘true’’ zygosity based on ‘‘physical similarity, the other brunettes (positive assortative mating) or non-brunettes (neg- number of choria and placentae, and the hospital doctors ative assortative mating). 2 ascription of zygosity and the parental opinion’’; when Technically, if their genetic similarity in appearance, for instance, these sources disagreed, fingerprints were analyzed and is causing the twins to be confused and/or treated more similarly, then that is an effect of genes and thus should unproblematically be part of blood group was gathered in a few cases (Goodman and the overall ‘‘genetic’’ effect (Jencks 1980). However, this logic flies in Stevenson 1989). Xian et al. (2000), Scarr and Carter- the face of common sense understandings of what we mean by genetic Saltzman (1979), and Kendler et al. (1993) found evidence effects and makes the estimates less externally valid to the rest of the to support the EEA for other behavioral traits based on a non-twin population. Moreover, bias is introduced by any increased cross-sibling interaction that leads to increased similarity in variety of twin data. Kendler et al. (1993) examined major phenotypes. depression, generalized anxiety disorder, phobia, bulimia 123 Behav Genet (2013) 43:415–426 417 and alcoholism using female twins from the Virginia Twin (ADHD); depression; cumulative high school grade point Registry. Xian et al. (2000) examined alcohol and drug average (GPA); and birth weight. Each of these phenotypes dependence, nicotine dependence, major depression, and has