Nurture effects in twin studies: distinguishing between “does not” versus “cannot” matter

I may very well be living in an isolated online bubble, but I get the feeling that behavioral genetics is making quite a lot of waves on social media recently. A prominent example is the excellent article “Why parenting may not matter and why most social science research is probably wrong”, by Brian Boutwell, which has been circulating since December, and seems to have sparked something of a renewed nature-nurture discussion. Twin studies have proven to produce one of the most reliable patterns in behavioral science, virtually untouched by the current replication crisis in psychology – a certain amount of variation in pretty much any behavioral trait can be attributed to genetics, while parenting probably has a quite negligible impact on many variables we are interested in as social scientists. In my own field of political science, twin studies have also finally been making a breakthrough in recent years. For example, it appears that a substantial portion of variation in political attitudes can be explained by genetic factors.

Twin studies utilize the (sort of) natural experiment of variation in amount of shared genetics between identical, or monozygotic (MZ) and fraternal, or dizygotic (DZ), twins (100% versus 50%) to tease out how much of the variation in a given variable that can be explained by the difference in shared genetics. Put simply: if identical twins are more like each other than fraternal twins on some trait we’re interested in, we can plausibly say that at least some of that trait is heritable. Typically, a type of statistical design known as variance decomposition is used – how much variance can be explained by genetics and environmental factors. In MZ vs DZ designs, the compartments of variation we can distinguish between is often labelled “additive genetic component”, “shared environment” and “non-shared environment”.

The additive genetic component describes the part of all variation that can be explained by the shared genetics of the twins. The shared environment captures the non-genetic factors that are common to both twins – that is, essentially parenting (and a few other things). Finally, the non-shared environment captures all those other causes that may vary between any given pair of twins – circles of friends, different classes in school etc, but also randomness and non-shared biological factors – random mutations, position in the womb etc (this post makes a good case that much of the non-shared environment has been over-interpreted by some to be exclusively effects of non-shared socialization).

The accumulated knowledge from these types of studies has shown beyond any reasonable doubt that genetics actually do have a substantial impact on why humans differ in behavioral traits, and effectively punctured blank slate theories that exclusively stress socialization as an explanation for why we are who we are. While there may be some methodological concerns about twin designs, such as violations of certain statistical assumptions (the equal environments assumption for example), proponents of blank slate-type theories have yet to give any solid arguments for why heritability estimates should be seriously doubted (this debate over the validity of twin studies in a criminological journal is very illuminating).

However: there is a somewhat important error that I’ve heard a few people (though not behavioral geneticists themselves) commit when discussing the implications of these results. A while ago, to give an example, I heard a discussion on the role of parenting in crime prevention, where one side argued that parents have to get more involved in their kids’ lives, lest they turn out to be criminals, and the other side argued, on the basis of the lack of parenting effects found in twin studies, that parenting doesn’t matter and that such efforts are therefore wasted. There is a fine line here between descriptive and causal claims that is important to keep in mind. Twin study designs say something about whether parenting does explain variation in crime in a given sample, but not whether it can. When extrapolating from “parenting doesn’t matter” to “parenting cannot matter”, one commits a crucial logical error: this is a misunderstanding of what variance decomposition as a method can actually accomplish. Ironically, I’ve also heard people oppose the method as such on the same grounds: what are we supposed to do with the knowledge that people can’t change in response to environmental triggers (better to turn a blind eye, supposedly)?

To see why this is a confused conclusion, it is helpful to think of environmental components as forces, and as any forces, their strength can vary. Let’s resort to an idealized thought experiment on the heritability of height. Consider an ideal case where the force of the environment is effectively zero: a number of pairs of MZ and DZ twins are isolated in a complete vacuum away from any stimulus whatsoever, and from each other (imagine a Matrix-like stasis chamber where their physical bodies are floating around in some nutritious goo). Assume further, for the sake of argument, that there is no randomness in their developmental trajectories, or that any randomness affects them both identically. It is clear that any difference that develops between them must be due to genetic influences only.

Now let’s consider another set of twins in similar situation, but they have in this case been placed inside of quite small boxes, much too small to fit their full potential body size. It is simply not possible for them to grow to the height they would have reached if they were unconstrained. In fact, they will probably both grow to the same height, namely the height that the box will permit. The force of the environment is in this case, as opposed to the former, very strong. We will observe no difference in height differentials between MZ and DZ twins, and therefore conclude that the heritability of height is effectively zero.

The converse of this logic in the real world is that while descriptively speaking, parenting (or other environmental factors) may not matter because the force of parenting is not large enough, that doesn’t mean shared environment cannot matter if the force of parenting increases. In short, heritability estimates from a given study is simply the heritability of a trait given the set of environmental forces that particular sample of twins were exposed to. If parents were simply much more authoritarian and strict, or if psychologists developed some powerful previously unthought of parenting technique, it is not inconceivable that we would in fact observe a much higher proportion of explained variance by the shared environment component. This is not to say, of course, that this is necessarily something we would want, but it underscores why we should be careful about generalizing from explained variance in twin samples to something like human nature. For that, there are other options.

An example from political science: as I mentioned initially it has been found that a substantial portion of political attitudes are heritable. The exception seems to be left-right placement in certain European countries (like Sweden). Now, if we were to apply the erroneous reasoning outlined above, we would have a hard time understanding why heritability of ideology might vary between countries, unless we think different populations are somehow genetically different in this very specific way. The real reason, however, is probably that traditional left-right placement has a much more fundamental political role in, for example, Scandinavia, which leaves more room for environmental forces. As the authors of this study conclude: “The phrase ‘Left-Right’ in European countries takes on a different context than the terms liberal or conservative, or those ideologies formed from political attitude positions. Indeed, the phrase ‘Left-Right’ appears highly subject to local and cultural definitions and may indicate group identification more than ideological position. For example, in Sweden the term ‘Left-Right’ is strongly-related to party identification and voting behavior. The finding that the environment accounts for nearly all the variation on measures which use the phrase “Left-Right” is similar to explorations of party identification in the US during the 1980’s and 1990’s, which find that ideology is heritable, but party identification is not.”

I’m not trying to convey that I think parenting is actually potentially important – for all I know it is nothing but a thin polish on top of a solid gene-machine. However, methodologically speaking, variance decomposition cannot tell us whether that is the case. If we really wanted to see whether parenting can have an effect, we shouldn’t rely on observational approaches, but would have to conduct experiments where the force of parenting (or some other environmental factor, if we’re not interested in parenting) was randomized between subjects. Thus, if we actually care about addressing certain social ills, we still have plenty of room, and high heritability estimates shouldn’t necessarily be feared. In fact, they can even provide valuable information about where more powerful socialization is needed.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s