Kendler and colleagues examine behavioral traits in relation to a wealth of environmental factors over the life course as well as to the relatedness of the individuals (Kendler and Prescott 2006). In Kendler et al. (2002), for example, data on over 1,900 twins are used to fit the incidence of major depression to an additive model that incorporates many environmental factors and a “genetic risk” factor. This last factor is derived from the incidence of major depression in the co-twin and parents, with adjustments made for the degree of relatedness of the twins (monozygotic versus dizygotic). The model accounts for 52% of the variance in the trait and provides a picture of development that is rich and plausible (see figure below).
[C]onsider the results of the study of multiple twins while thinking about your own family. Consider a scenario in which the nonidentical twins are much less similar than identical twins, which means that sharing fewer genes makes a big difference (skipping here the technicalities of getting the number—heritability—that quantifies that result…). If this were the case, for whatever trait we are thinking about, e.g., IQ test score, we might say: “There is nothing I could do as a parent to change the outcome for my offspring. I am not to blame for the outcome other than having passed on my genes.” If that conclusion seems justified, we might then reason that the same is true for every other family, and thus society as a whole should not try to change what it is doing because it will not make a difference.
Complications Continue reading
Critics of twin studies as a means of linking human behavior to genes have a history of making vulnerable arguments and of missing or even reinforcing more fundamental problems. This is evident in Brian Palmer’s (2011) Slate article, “Double inanity.”
Consider a 2005 study by Rice University’s John Alford and his colleagues claiming to show that 43 percent of the variation in political ideology in the U.S. could be attributed to genetics… Duke political scientist… Evan Charney and Harvard geneticists Jon Beckwith and Corey Morris examined the flaws in the Alford study—and showed why all the other twin studies on heritability can’t possibly show what they purport to show… Twin studies rest on two fundamental assumptions: 1) Monozygotic twins are genetically identical, and 2) the world treats monozygotic and dizygotic twins equivalently (the so-called “equal environments assumption”). The first is demonstrably and absolutely untrue, while the second has never been proven.
The first more fundamental problem (let us call this MFP #1) is that no logical or empirical connection exists between heritability on one hand, which the so-called “quantitative genetics” estimates based on similarities of observable traits (such as height, IQ test scores,..), and, on the other hand, transmission of genes from parents to offspring. Palmer as well as Alford assume that the two ideas are connected.* The Alford study reports on measures of heritability but is titled “Are Political Orientations Genetically Transmitted?” Palmer leads off by presenting a figure of 50% heritability (from a 2008 study) as “Genes determine 50 percent of the likelihood that you will vote.” To counter MFP #1 critics need to take every opportunity to clarify—or at least point out—the confusingly multiple meanings of the term genetic, which include: runs in a family, percentage of genes shared (relatedness), associated with similarities of traits in genetically-defined varieties (heritability), predictability of trait changes in specific populations and locations (heritability), transmitted through germline (heritable), inborn, innate, difficult to change by environmental or social factors, etc.
(*In case the reader thinks they are connected because the definition of heritability is the proportion of variance of the trait associated with genetic differences, it is necessary to expand that last misleading shorthand to “associated with differences in the genotypic values,” that is, in the mean value of the trait for a genotype when measured across all environments in which it is raised. This “genotype” is simply a synonym for variety; it does not refer to the pairs of alleles in the Mendelian sense.)
Now consider the flawed fundamental assumptions cited by Palmer. The first is that monozygotic (MZ) twins are genetically identical. This has been central to human twin studies, but such studies can be recalibrated to account for the actual proportion of shared genes in MZ and dizygotic (DZ) twins. In any case, most of the ways that MZ twins are less than identical in their genes also influence the actual proportion of shared genes in DZ twins. While departure from previous figures for genes shared in MZ and DZ twins increases the uncertainty (“error”) of heritability estimates, this is not a fatal blow to the method of data analysis involved in twin studies.
The second fundamental assumption cited by Palmer, the equal environment assumption, emerges from discussions that, in line with MFP #1, take heritability as a measure of the influence of genes and the percentage not included in heritability as a measure of the influence of environmental factors. Disputing the equal environment assumption by examining how well twins share specific environmental factors runs the risk of reinforcing MFP #1.
Let me note three more fundamental problems in twin studies. The easy slide from the “genetics” of heritability estimation in twin studies to the genetics of identifiable gene sequences is made possible because partitioning of trait variation into components uses models of hypothetical, idealized genes with simple Mendelian inheritance and direct contributions to the trait. The analysis of trait similarity among relatives can, however, be undertaken without using models of hypothetical genes (Taylor 2012). Recognizing this possibility makes it easier to dispense with the logically and empirical unjustified assumption that, all other things being equal, similarity in traits for relatives is proportional to the fraction shared by the relatives of all the genes that vary in the population. After all, as can be shown using plausible models of the contributions of multiple genes to a trait, all other things being equal, ratios of DZ similarity to MZ similarity that are not .5 and vary considerably around their average (Taylor 2012). MFP #2 then, which affects all twin studies to date, is that heritability estimates based on this assumption are unreliable.
Fundamental problem #3 is that, in analyses of human data, ‘‘genotype–environment interaction’’ variance is discounted or subsumed in an inflated estimate of “heritability.” To explain this problem, the meaning of genotype-environment interaction in quantitative genetics, including twin studies, first needs clarification. To aid this clarification, the following note uses variety and location where others might use genotype and environment.
In everyday terms, a high degree of variety–location interaction simply means that the responses of the observed varieties across the range of the observed locations do not parallel one another. That is, one variety may be highest for the trait in one location, but another variety may be highest in another location-or, at least, the difference between any two varieties may change location to location. This classical [quantitative genetics] sense of genotype–environment interaction is distinct from the use of the same term (or, synonymously, ‘‘gene-environment interaction’’) for situations in which the ‘‘genotype’’ denotes a value of a measured genetic factor, the ‘‘environment’’ denotes a value of a measured environmental factor, and an interaction means that the quantitative relation between the trait and one of the factors varies according to the measured value of the other factor (Taylor 2012).
It may be that genotype–environment interaction variance turns out to be a small component of the variance for most human traits. However, to show this, not assume this, requires data sets that are rare (e.g., for the same population and location data are needed on MZ raised together, MZ raised in independent families, and unrelated sibs raised together). The implications of MFP #3 are significant:
If it is not assumed that variety–location interaction variance can be discounted in human studies, two common claims become open to question: (a) The effect of family members growing up in the same location (family) is of small importance; (b) The trend for heritability estimates to increase over people’s lifetimes is evidence that ‘‘genetic’’ differences come to eclipse ‘‘environmental’’ differences…. The first claim requires showing not only that the location variance is a small component of the total variance, but also that the variety–location-interaction variance is small. The second claim also requires showing that the variety–location-interaction variance is negligible; otherwise it could equally well be that the interaction component increases over time (Taylor 2012).
The final MFP is a corollary of MFP #1:
the [genetic and environmental] factors underlying the development of observed traits may be heterogeneous, that is, they do have to be the same from one set of relatives to the next, or from one family (location) to the next. It could be that pairs of alleles at a number of loci, say, AAbbccDDee, subject to a sequence of environmental factors, say, FghiJ, are associated, all other things being equal, with the same outcome for the trait as are alleles aabbCCDDEE subject to a sequence of environmental factors FgHiJ (Taylor 2010). If underlying factors can be heterogeneous, the use of heritability as a basis for judging a trait to be a good candidate for molecular research… becomes unreliable (Taylor 2010). (Similarly for research that builds on the other fractions of the variance).
Even if it were to be shown that genetic differences among MZ twins did not qualitatively alter heritability estimates and that the equal environment assumption held in some cases, the four MFP’s render human heritability estimates unreliable quantitatively and as a basis for research into the genetics (in the sense of actual genes) of human behavioral traits.
Taylor, P. J. (2010) “Three puzzles and eight gaps: What heritability studies and critical commentaries have not paid enough attention to,” Biology & Philosophy, 25:1-31.
Taylor, P. J. (2012) “A gene-free formulation of classical quantitative genetics used to examine results and interpretations under three standard assumptions,” Acta Biotheoretica, 60(4): 357-378.
Since publishing on the laws of behavioral genetics (Turkheimer 2000, 2004), Turkheimer, his colleague Robert Emery, and their students have analyzed the similarity of offspring of monozygotic twins to clarify the relationship between parental traits, especially divorce, and behavior of their offspring. Turkheimer (2008, 4) describes the logic of their analyses:
[I]f a genetic propensity to be aggressive makes parents more likely to get divorced, and those same genes when passed to the children make them more likely to be aggressive on the playground, then one will observe an association between divorce and playground aggressiveness that will not really be a causal consequence of divorce… But in identical twin parents… none of the differences between the children can arise from differences in the genes of their twin parent, so if the children do differ, we can (almost…) rule out a genetic explanation of the association.
Suppose poor families are more likely to be divorced than well-off families, and children raised in poor families are more likely to be delinquent. [We could] observe an association between divorce and delinquency that doesn’t have any causal relationship to divorce. But twin parents share their family history of poverty, so if the children of the divorced twin are more likely to be delinquent than the children of the non-divorced twin, the parental poverty isn’t a plausible alternative explanation…
Turkheimer follows up the parenthetical “almost” with caveats concerning, for example, the contribution of the other non-twin parent. Not included among his caveats, however, are the following points:
- a. shortcomings in estimation of human heritability (Taylor 2012; also previous post), presuming that heritability is what is meant by “genetic propensity”;
- b. unreliability of the heuristic that all other things being equal, similarity in traits for relatives is proportional to the fraction shared by the relatives of all the genes that vary in the population (Taylor 2012; also previous post);
- c. heritability (“genetic propensity”) as a measure of similarity in the trait does not translate in any direct way to hypotheses that invoke underlying genes or genetic factors (Taylor 2010); and
- d. asymmetry in conceptualization of genetic and environmental (or social) factors—the latter are measurable; the former unknown and potentially heterogeneous (Taylor 2010; also previous posts).
Turkheimer (2008, 5) reports “a rich variability of outcomes”—some indicating a genetic association, some an association with the social factor, and some ruling out hypothesized associations. We do not know, however, how much this variability of outcomes is generated by the unreliable heuristic (issue b). Yet suppose that, for the purposes of discussion, we had results accompanied by an analysis of sensitivity to variation away from the unreliable heuristic value or, even, that the heuristic were dispensed with because we had data on the necessary classes of relatives (Taylor 2012). Could we then interpret similarity or dissimilarity of offspring of monozygotic twins as a “reflection of the environmental and genetic developmental processes that underlie complex human behavior” (Turkheimer 2008, 5)? My answer: Not readily. Whereas Turkheimer’s phrases “genetic propensity” and “genetic explanation” suggest an equivalence or a direct translation between measures of similarity in the trait and hypotheses that invoke genetic factors underlying the trait, this is not so (issue c and “Gap 2” in Taylor 2010). “Genetic explanation,” moreover, suggests a symmetry with environmental or social explanation, but this is not so either. The asymmetry can be seen by examining the first scenario that Turkheimer uses to describe the logic of the analyses and an alternative, then considering what interventions to modify the developmental processes follow from the analyses.
The first scenario is the one in which there are no grounds for a “genetic explanation” of the association between divorce and the outcome, aggression of child. This scenario is summarized in table 1. (Notice that the non-measured, unknown genetic factor or composite of factors contributing to the similarity of the first pair of twins for the trait—here, aggression—is shown as distinct from the genetic factor contributing to the similarity of the second pair of twins given that the analysis cannot show the factors to be the same or similar.) To reduce the incidence of aggression seen in children of divorced parents, it would be plausible under this scenario to focus on policies to reduce divorce without reference to genetic factors or genetic relatedness.
Notes: In any study offspring will come from several twins, but the number shown to illustrate the logic of the scenario. Black/white cells indicate presence/absence of the factor. Diagonal hatching downwards denotes offspring of any one of an identical twin pair are not identical in their genetic factors because there is a second parent and, for the same reason, offspring of parents that include the different identical twins are not identical in their genetic factors. Diagonal hatching upwards denotes twin parents are not identical in the measured factor because heritability is less than 1.
Now imagine a new scenario, shown in Table 2, in which the analysis does not rule out a “genetic explanation” of the association between divorce and the outcome, aggression of child.
Table 2. Analysis of offspring of monozygotic twins: Behavioral outcome does not rule out a “genetic explanation” of the association between divorce and the outcome, aggression of child.
Notes: In any study offspring will come from several twins and randomly chosen parents, but the number shown are sufficient to illustrate the logic of the scenario. Black/white cells indicate presence/absence of the factor. Diagonal hatching downwards denotes offspring are not identical in their genetic factors to the twin or randomly chosen parent because there is a second parent and, for the same reason, offspring of parents that include one or the other of an identical twin pair are not identical in their genetic factors. Diagonal hatching upwards denotes a) twin parents are not identical in the measured factor because heritability is less than 1; and b) similarly, for non-twin parents, the hypothetical genetic factor is not expressed fully as the measured environmental factor.
What is learned about possible interventions to modify the “environmental and genetic developmental processes that underlie complex human behavior”? Clearly, we should not focus on policies to directly reduce divorce. But, beyond that? We could seek to identify the unknown genetic factors associated with aggression of a parent, but nothing in the analysis rules out these factors differing from twin pair to twin pair and among non-twinned parents. In light of that possibility, we could restrict our focus to close relatives (Taylor 2010). For identical twins who become parents, once aggression is seen in children of one twin, we could advise the other twin to be more attentive to the issue of aggression. That is, we could seek ways to help the parent reduce their aggression insofar as it affects the children (starting perhaps before the parents have offspring). And we could seek ways to help the children reduce their aggression. However, we could also advise such attention to aggression independently of knowledge about genetic relatedness and of hypotheses about underlying genetic factors. There is no justification for thinking that, because (unknown) genetic factors have an influence, environmental interventions in the developmental processes are unlikely to succeed or a diversion of resources from measures more likely to be fruitful. In sum, in analysis of the similarity of offspring of monozygotic twins, the environmental factors are measurable and point to interventions, but the genetic factors are unknown, potentially heterogeneous, and informative only for advising close relatives even in the thought experiment where issues a and b have been overcome. (Of course, behavioral outcomes and environmental factors may also be heterogeneous [Gatze-Kopp et al. 2012], but this does not help us interpret the outcomes of the analyses of Turkheimer and colleagues.)
Gatzke-Kopp, L. M., M. T. Greenberg, et al. (2012). “Aggression as an equifinal outcome of distinct neurocognitive and neuroaffective processes.” Development and Psychopathology 24(Special Issue 03): 985-1002.
Taylor, P. J. (2010). “Three puzzles and eight gaps: What heritability studies and critical commentaries have not paid enough attention to.” Biology & Philosophy 25(1): 1-31.
Taylor, P. J. (2012). “A gene-free formulation of classical quantitative genetics used to examine results and interpretations under three standard assumptions.” Acta Biotheoretica 60(4): 357-378.
Turkheimer, E. (2000). “Three laws of behavior genetics and what they mean.” Current Directions in Psychological Science 9(5): 160-164.
Turkheimer, E. (2004). Spinach and Ice Cream: Why Social Science Is So Difficult. Behavior genetics principles: Perspectives in development, personality, and psychopathology. L. DiLalla. Washington, DC, American Psychological Association: 161-189.
Turkheimer, E. (2008). “A better way to use twins for developmental research.” LIFE Newsletter (Max Planck Institute for Human Development)(Spring): 2-5.
“What does it mean to say IQ test scores are largely genetic? Does this make you think that differences among IQ scores between races are genetic?” Starting with these two questions, I aim to get an audience of high school students thinking about the three areas that make up this sequence of posts. Comments welcome about how to revise this approach so as to provide a “conceptual starter kit for thinking about genes, race, and IQ test scores.“
A. First question, “What does it mean to say IQ test scores are largely genetic?” (asking for audience input and filling in the following sequence of answers and “buts”)
1. Genes cause IQ more than environment does. But how to separate these—everyone also needs an environment to survive and grow.
2. If your parents are above-average IQ, you are more likely to be above average. But parents also pass on environment (educational options, etc.)
3. Differences in genes from one person to the next have more influence on IQ than differences in environment. But how do you track the influence of genes without knowing which of our 30,000 genes influence IQ, and even worse for environmental factors.
4. If you and your identical twin were separated at birth and raised in difference families, you’d be more similar than any two random people raised in different families. But these situations are very rare (and not necessarily at birth and not necessarily in independent families).
5. If your sibling is adopted into a professional family but you grow up poor and you end up with same IQ. But this is not on average true—usually there’s a strong boost in IQ for the adopted child.
6. Identical twins raised in the same family are more similar on average than non-identical twins raised in same family. Remember, identical twins share all their genes, but non-identical do not, so the increased similarity corresponds to sharing more genes. This answer seems better (although not without critics). It is on this basis researchers say IQ has a “heritability” of 60-80%.
(Note: Here heritability is a technical term with a quite different meaning from the idea that something is heritable if there is a gene that gets transmitted from parent to offspring.)
Continued in the next post.
Many people say Nature vs. Nurture is an ill-framed formulation, but the challenge is to explain why in a way that accounts for the persistent popularity of that formulation.
We know, for example, that both genes and environment are involved even in the cases in which there is a single gene with a major and direct effect, such as phenylketonuria (PKU). In PKU the development of individuals having two copies of a non-functioning allele for the enzyme phenylalanine hydroxylase (PAH) is extremely impaired by the level of phenylalanine present in normal diets, but much much less impaired if a special diet is maintained. OK, but one might say that genes are primary here: Without the genetic condition, we don’t need to worry about the diet (the environment). Knowing the genetics, or at least, the biochemistry associated with the genetic condition, points to the appropriate environment to alter. In fact, all kinds of changes in upbringing of individuals with the genetic condition would have no effect. In short, nature interacts with nurture, but it’s most important to know about the genetics and biochemistry. And, if that’s the case for PKU, you might well suspect that there are many other genes, perhaps of smaller and less direct effect, for which the same primacy would hold. And, why decide in advance that certain traits, such as IQ test scores, are not amenable to genetic study? One might say all that. The response sketched in this blog and follow-up installments, however, suggests that the issue is more than the (possible) primacy of genes (as just sketched above); there is a conflation of family and population involved in the persistence of the Nature vs. Nurture formulation.
Starting within a family, it’s very easy for someone to see that children physically resemble their (birth) parents (where resemble means more than look like their parents—it is look more like their parents than any two people randomly picked from the population). In most cases parents pass onto their offspring environment as well as genes, so one might well ask how important is each? That’s hard to say when parents pass on both so you imagine identical twins raised apart from birth and ask: Does the one raised at home resemble the parents more than the one raised away? But, of course, no offspring resembles both parents well—(generally) a child is male or female and lots of characteristics (e.g., height, hips) go along with that. So, perhaps you look at resemblance between offspring and same-sex parent for physical traits and between offspring and the average of the parents for other traits? That doesn’t quite work for sexual characteristics. Conversely, the average of the parents might also capture resemblance for physical traits such as height. You simply expect there to be variation around that average, e.g., female offspring end up on the shorter side of the average and male offspring end up on the taller side.
Once you start talking about variation and averages, and stop expecting a clean picture of resemblance in any one family, you can shift your sense of resemblance from the family to averages over many offspring-parent pairs. Going back to the identical twins, if you imagine many such pairs of twins, on average does the one raised at home resemble the parents more than the one raised away? You could also ask, if you have many sets of same-sex non-identical twins raised together and many sets of same-sex identical twins raised together, on average do the does one identical twin resemble the other twin more than one non-identical twin resemble the other? (Shifting to same-sex twins means that we don’t have the complication of differences between an offspring and its other-sex parent. And shifting to twins raised together means you don’t have to search for the rare cases of twins separated and raised in truly independent families.) If the answer in the twin-resemblance study is yes, it seems reasonable to conclude that the identical twins are on average more similar because they share all their genes whereas the non-identical twins share fewer of their genes. However, that conclusion doesn’t say that it’s the same nature—the same genes—or the same nurture that brings about the resemblance from one pair of twins to the next. So then where are you? Stay tuned for the next installment.
The possibility of underlying heterogeneity makes heritability studies even less informative than, as prominent geneticists have noted (e.g., Rutter 2002, 4), the method of data analysis not suggesting where to look for the underlying genetic factors that contribute to heritability. Here I comment on the difficulty of translating from statistical analyses of data on traits to hypotheses about the measurable genetic or environmental factors involved in the development of the traits.
First recall what heritability means:
Consider claims that some human trait, say IQ test score at age 18, show high heritability. These claims can be derived from analysis of data from relatives. For example, the similarity of pairs of monozygotic twins (which share all their genes) can be compared with the similarity of pairs of dizygotic twins (which do not share all their genes). The more that the former quantity exceeds the latter, the higher is the trait’s heritability. Researchers and commentators often describe such comparisons as showing how much a trait is “heritable” or “genetic.” (from previous post)
Now the possibility of underlying heterogeneity is that
even if the similarity among twins or a set of close relatives is associated with similarity of (yet-to-be-identified) genetic factors, the factors may not be the same from one set of relatives to the next, or from one environment to the next (from previous post).
Imagine now that I revealed that the trait in this picture was body length and the twin pairs in this figure came from different species. You might ask why do heritability studies on pairs from different species–what could we do with the answer? But note, there is nothing in the method of estimating heritability that takes into account whether the pairs came from the same species or not. Indeed, the species could be Homo sapiens in Siberia, Pseudechis porphyriacus in Australia, and Sorghum bicolor in Asia. You would not expect that the genetic factors underlying the high heritability of body length in humans, red-bellied black snakes, and sorghum were the same. I suspect that you wouldn’t take high heritability of body length from this study as an indication that this trait is a good candidate for “molecular research to identify the specific genetic factors involved” (see another previous post). The reason—you would expect that the “even if the similarity among twins or a set of close relatives is associated with similarity of (yet-to-be-identified) genetic factors, the factors [would] not be the same from one set of relatives to the next.” But how closely related do the twin pairs have to get before you stop being concerned about the possibility of underlying heterogeneity? If the twin pairs were all from the same order would we use heritability as a guide to undertake molecular research? From the same genus? Same species? Same population? The methods of heritability studies do not tell us what degree of relatedness among the twin pairs (as against within the twin pairs) is close enough that the possibility of underlying heterogeneity disappears. Which leads us back to the question of a previous post, “What can researchers do on the basis of knowing a trait’s heritability if the genetic and environmental factors underlying the observed trait are heterogeneous?”
More in the future on the challenges of getting others to appreciate the significance of the possibility of underlying heterogeneity. For now, let me note that the fact that similarity of twin pairs is not the only way to estimate heritability is not a pertinent objection. If heritability as estimated by the other means can also be estimated by similarity of twin pairs, the conceptual point made in this post must be relevant to those methods as well.