Sociologists should engage with behavior genetics

Over the last thirty years, behavior geneticists have examined the influence of genetics on dozens of outcomes of interest to social scientists. A fundamental conclusion of this research is that genes have non-negligible (and often quite substantial) effects on nearly every behavior (Turkheimer, 2000). For some outcomes, the effect of genes appears strong enough to account for the bulk of (measured) similarity between parents and their children (Rowe, 1994; Harris, 1999).

These conclusions would seem to warrant a paradigm shift in the field of sociology, a discipline that has long treated biological influences as irrelevant or negligible. Yet the core curriculum of sociology still makes little mention of biological influences. Biological theories are typically given relatively little attention in commonly used textbooks, and when such theories are considered, they are often portrayed as remnants of racist crack pot theories that lack an empirical basis.

But the foundation on which behavior genetic evidence relies is no less tenable — and no more tenable — than the evidence on which sociologists rely for many of their conclusions. In other words, twin studies are no more or less scientific than regression analysis. For this reason, I think sociologists should take behavior genetics more seriously. Seriously doesn’t mean “adopt it whole,” but instead to engage with it, present it as a viable theory in textbooks, attempt rigorous critiques of its assumptions, etc.

Studies of twins reared apart

In an earlier blog post I mentioned that until very recently there were essentially three types of studies researchers use to differentiate between the effects of genes and the effects of parental socialization. I have mentioned two of those study designs – adoption studies and twin studies. There are various methodological variants on the twin study which I will not get into.

However I have not mentioned the third study design used to estimate the effects of genes. This is the so-called combination adoption twins study. This sort of study design requires that the researcher finds identical (MZ) and/or fraternal (DZ) twins who have been raised in different families since birth for at least since very early in their life. Obviously, it’s very difficult to find these kinds of people and so there are very few studies of twins raised apart.

Perhaps the most famous study of this kind is the Minnesota study of twins reared apart (aka MISTRA). Over perhaps a decade or more researchers at the University of Minnesota recruited MZ and DZ twins who reported that they had been separated at birth or a bit later. The Minnesota group recruited twins from all over the country and in some cases from outside of the United States.  Twin pairs were brought to Minnesota to answer dozens of questions gauging their beliefs, skills and personality skills. The results were stunning.

The Minnesota team reported many of their most important findings in a classic 1990 article in Science magazine. MZ twins who had been raised in separate families nonetheless were quite similar to each other in a variety of ways.  The Minnesota group also compared monozygotic twins reared apart with monozygotic twins reared together.  For example consider IQ.    Correlations between the IQs of MZ twins reared apart were quite high, albeit smaller than the IQs of MZ twins reared together.    Among the 48 pairs of MZ twins reared apart, WAIS full test scores correlated at 0.69.  Among  the 40 pairs of MZ twins reared together, WAIS full test scores correlated at 0.88.  This result provides persuasive evidence that IQ has a very large heritable component.

Limitations with studies of twins reared apart?

Theoretically, the adopted twins studies provide the most robust test one can imagine. Think about what these studies have potentially found. The idea is that you take two genetic clones, put them in different families and then 40 years later see how similar they are. If the twins are very similar then you can say with a very level of confidence that genes are making a big difference. On the other hand if the twins aren’t that similar you can say that genetic influence is weak. Of course the reality is somewhere in between but the upshot of the Minnesota study of twins reared apart as well as other similar studies such as those in Sweden is that genetic influence is substantial.

This is not to say that there are not limitations or at least questions about these sorts of studies. For example, some researchers questioned whether the twins in the Minnesota study had been raised in households that were entirely distinct from one another. You could have for example two twins whose parents had died adopted by different members of the same extended family.  Such twins adopted in different nuclear families but the same extended family may have remained in close contact for much of their lives.  In defense of the Minnesota study, researchers did collect data on the amount of time that the twins spent together and the age at which they had been separated. Generally the researchers found that age of separation and the amount of time that the twins had spent together made little or no difference on similarity in how they turned out.

The Minnesota study of twins reared apart may also suffer from selection bias.  When doing a study of twins reared apart, you don’t have the luxury of randomly choosing your sample from your population of interest. Instead when you’re looking for such a rare population as identical twins separated at birth or in childhood, you’re going to take anybody you can get. The problem with this is that the twins who end up volunteering for your study may in fact be twins who tend to be more similar to each other. Twins who are distant or dislike each other may be less likely to volunteer for such a study.   For this reason, correlations between the twins in the study on any given trait may be artificially inflated relative to what you would find if you had a truly representative sample. Nevertheless the findings of Minnesota study cannot be easily dismissed. Even if we allow for the fact that some of the twins in the Minnesota sample did in fact spend a good deal of time together prior to the study, the conclusion and the results are still striking.

The equal environments assumption

Over the last 30 or 40 years, there has been a considerable amount of research — perhaps two dozen studies — examining whether environmental similarity biases the results of twin studies. The vast majority of this research finds little to no evidence that twin studies are biased in this regard.  Most twin studies have cited this research in support of what is known as the equal environment assumption or EEA for short.   The equal environment assumption is perhaps a misnomer because it doesn’t mean that we assume that MZ twins share environments that are as similar as DZ twins. Instead, the equal environments assumption simply states that environmental similarity between twins does not have much of an impact on trait similarity.

There are however some gaps in the research investigating the EEA and that’s where my dissertation and forthcoming book came in. In my dissertation, I attempted to test the EEA in a more comprehensive fashion then had been done previously. Using a nationally representative sample of twins, I tested the impact of a wider array of environmental similarity measures. I also looked at the impact of all of these environmental similarity measures on genetic estimates for a wide variety of outcomes that social scientists are interested in.  So what did I find? When controlling for a wide variety of measures of environmental similarity I found that the bias affecting twins studies is perhaps larger than many behavior geneticists like to believe but also smaller than many critics of behavior genetics like to believe.

Let me try to explain what I did in my dissertation in layman’s terms. What I essentially did was to find MZ twins and DZ twins whose parents and peers treated them with the same level of similarity at least in so far as could be measured.  Consider a concrete example. Most MZ twins are in good contact with each other. Meanwhile DZ twins are less likely to remain in good contact. Part of what I did in my dissertation was to find the DZ twins who do remain in close contact like many of the MZ twins and then to compare that subset of the DZ group with the MZ group. The idea here again is to eliminate any effects of environmental similarity. In some cases when I controlled for environmental similarity, the apparent effects of genes declined dramatically. But in many cases the effects of genes were robust to the inclusion of controls for environmental similarity. This means that when I compared the DZ twins who spend a lot of time together and had very similar experiences with the MZ twins, I still found the same genetic effects as would have been found in the standard twin study that didn’t control for environmental similarity.

One thing that we can conclude is that whatever bias introduced by environmental similarity is not large enough to invalidate the central conclusion of twin studies. The overall conclusion reached by behavior geneticists from twin studies is that nearly every trait you can imagine is heritable. The general idea is that the effects of genes pervade our life and explain a lot of the differences between us. My dissertation added a caveat to that conclusion: perhaps genetic effects are not quite as large as some of the twin studies lead us to believe.  Nonetheless, genetic effects are pervasive and often substantial.

 

Potential limitations / problems with adoption & twin studies

I have explained how you can estimate the effects of genes on traits that vary with data from adopted kids or with data from MZ and DZ twins.  Both of these study designs have limitations, which, depending on your point of view, either have little impact on the results, or render the results meaningless.

I’ll briefly mention several potential problems with each study design before introducing a third study design hat potentially avoids all of these problems.

Potential problems with adoption studies

• Data from the biological parents of adopted children are hard to come by.  An adoption study is still doable provided one has data from at least one adoptive parent as well as from the adopted kid herself.  The researcher will have to assume that the assignment of children to parents occurred essentially at random.  This assumption is sometimes tenable, sometimes not.
• Adoptive parents are sometimes biologically related to their adopted children in some way (e.g. an aunt or uncle adopts his/her niece when the niece’s parent dies or becomes unable to care for the child.)  In this case, adopted kids may behave more similarly to their adoptive parents simply because they share some of their genetic make-up.
• Adoptive parents may treat their adopted kids differently than they treat their biological kids.  Evolutionary psychology would predict this, although there’s evidence against it (http://asanet.ba0.biz/images/press/docs/pdf/Feb07ASRAdoption.pdf).  If parents invest less in their adopted kids than they invest in their biological kids, then we might expect any effect of parental socialization on adopted kids to be muted to some degree.

Potential problems with the classic twin study

Since MZ twins are genetic clones, they tend to:

• Experience very similar treatment from everyone who stands to influence them
• Spend a great deal of time together
• Experience a powerful bond, perhaps unlike the bonds experienced by any other pair of human beings

Meanwhile, DZ twins share no more genes than non-twin full siblings, so they will tend to experience relatively less similar treatment, spend less time together, and experience less powerful bonds than MZ’s.

Distinguishing genetic influence from family socialization

When we see children behaving like their parents, we often assume that the children learned the behavior from their parents.  For example, if we see a boy with violent behavior, and we find out his father is also violent in many circumstances, we might assume that the boy learned violence from his father.

But correlations between the behavior of parent and child may reflect common genetic predispositions, not parental socialization.  How can we differentiate the influence of genes from the influence of socialization?

Until fairly recently, there were basically three ways researchers could separate the effects of genes from the effects of parental socialization: adoption studies, twin studies, and (rarely) a combination adoption twin studies.   Below, I review adoption studies and twin studies.  I’ll get to the adopted twin studies a bit later.

In the most thorough adoption studies, researchers managed to collect data from kids who were adopted soon after birth, as well as data from the kids’ biological and adoptive parents.  One could estimate the importance of genes by looking at whether adopted kids behaved more like their biological parents or more like their adoptive parents.  One could estimate the importance of family socialization by looking at the similarity of adopted kids with their siblings who were biologically unrelated to them.   All the variation in outcomes unaccounted for by genes and family socialization was attributed to extra-familial influences.

In a twin study, the researcher gathers data from pairs of twins.  He ascertains whether the twins were born from one egg or two eggs.  Twins born of one egg (monozygotic twins) are genetic clones.  Twins born of two eggs (dizygotic twins) share half of their genes on average, although the exact amount shared varies from pair to pair.

In other words, dizygotic (DZ) twins share roughly half as much of their DNA as monozygotic (MZ) twins.   Researchers exploit this fact in order to  estimate how much variation in a trait is accounted for by variation in genes.  According to the basic model underlying the so-called classic twin study, the variance in the trait accounted for by genes is equal to double the difference between the correlation between MZ twins and the correlation between DZ twins.  Take IQ, for example.  IQ scores of MZ cotwins correlate at around 0.8.  DZ twins’ IQs correlate at around 0.4. Based on these numbers, we could say that genes account for 80% of variation in IQ.  (0.8-0.4) × 2 = 0.8.

Disentangling nature from nurture

Take some trait on which people vary.  Personality characteristics, intelligence, educational attainment, political attitudes, etc.  Now consider all of the variation in that particular trait within a nation such as the United States.  Some of that variation may be due to variation in family upbringing.  (A man who has little confidence in himself attributes his insecurities to his critical mother.) Some of the variation is due to differences in non-familial environment.  (That same man’s insecurity might be an outgrowth of persistent putdowns by peers in primary school.)  And finally, some of the variation in that trait across the population will stem from variation in genetic propensities. (That unfortunate man carried genes that, however directly or indirectly, engendered a lack of confidence.

Can we divvy up the variation amongst these three sources of influence — family environment, non- family environment, and genes?

Yes, as it turns out, we (sort of) can.  Behavior geneticists have been doing this for a very long time.  According to the traditional behavior genetics model, variation in a trait is like a pie which has to be divided into three pieces.  One piece of the pie represents variation due to genetic influence.   Another piece of the pie represents the influence of family socialization.  This is called “shared environment.”  The third piece of the pie – the piece left over after you cut out the other two – includes environmental influences from outside the family.  This piece of the pie is called “non-shared environment.”

Now, the question is: how do researchers divide up the pie?  What percent goes for genes (BG’s call this piece A)?  What percent of the pie goes for shared environment (BG’s call this piece C), and what piece goes for non-shared environment, which BG’s call E?

It turns out you need some special kinds of data to decide how to slice the pie.  Let’s start with the easy part: separating family effects – both genes (A) and shared environment (C) from non-family effects (E).  Researchers can estimate the overall impact of family by looking at data on siblings.  Looking at the variation within and between siblings for, say, educational attainment, in a large, representative dataset will give you a good sense of how much family matters overall to someone’s educational attainment in this society.  The more similarity we observe between siblings in educational attainment, the more power we can attribute to families in influencing this trait.  The more that siblings are different in educational attainment, the less important that family influence would appear to be.

So data on “ordinary” (non-adopted, biological) siblings allows us to separate the impact of A & C on the one hand, from the effects of E on the other hand.  But such data does not allow us to distinguish between the effects of genes and the effects of family (mainly parental) socialization.  We often see (uncanny) similarities between parents and their children, but we can’t discern the origin of this similarity.  Say Johnny and his mother are both outgoing.  We don’t know how much the similarity between Johnny and his mom is due to their shared genes or due to the fact that Johnny was raised by his mother.

In the next post, I’ll talk about how BG’s have typically separated A from C.