Friday, August 20, 2010

When to blame your parents, and for what

Studies linking some aspect of parental behaviour with some trait in their offspring are depressingly common in the sociological literature. Though these studies typically only report a correlation between parental behaviour and whatever the trait is in the offspring, the implication, and often the explicit conclusion, is that one causes the other. These kinds of stories get huge play in the popular press (and in the blogosphere), where the conclusion of a causative relationship is rarely challenged. For example, the finding that children who grow up with more books in the house are more successful academically is taken as evidence that simply having books around makes kids smarter.

This kind of thinking illustrates a common and fundamental flaw in interpreting sociological or epidemiological findings – correlation does not imply causation. Red hair and freckles are highly correlated but one does not cause the other. Both are caused by something else (a mutation in a gene controlling pigmentation). It seems a simple enough distinction but it is astonishing how pervasive this mistake is, even among academics supposedly trained in statistical methodology.

In the case of books, the conclusion that having them around is the causative factor on academic success is simply not warranted by the findings. The data from this kind of study design do not pertain to that question. The books could simply be an indicator of the real cause (like freckles). It seems quite possible that the underlying link is between the IQ of the parents (or some other cognitive trait predicting both academic success and bookishness – curiosity, open-mindedness, interest in more abstract topics) and that of their children. (It is well established that such traits are quite heritable).

I am not claiming that that actually is the explanation – just that it is a highly plausible one that must be considered. In fact, the study design does not permit this conclusion to be drawn either, and that illustrates one of the major problems in dissecting the possible effects of nature and nurture. It is hugely difficult to separate confounding genetic effects on behaviour of both parents and offspring from the effects of the behaviours themselves. Adoption studies – especially of identical twins reared apart – do provide one way to dissociate genetic effects from those of the family environment. These have consistently found large effects of shared genes and very little effect of family environment on a wide range of behavioural traits.

A far more tricky task is to dissociate the effects of parental behaviour prior to birth on the future behaviour of their offspring – adoption studies obviously cannot accomplish that. However, researchers in Cardiff, led by Anita Thapar, have come up with a clever and powerful new study design which does the trick. They have made use of the growing frequency of in vitro fertilisation to examine the effects of smoking during pregnancy. It is well known that smoking during pregnancy is associated with low birth weight and a number of other health issues. It is also associated with higher rates of antisocial behaviour in the offspring. Do these correlations really reflect the effects of smoking itself or could smoking be an indicator of a distinct underlying cause?

The IVF study design, which looked at records of 779 children, allowed these factors to be dissociated by splitting the mothers into two groups – those who were biologically related to their offspring and those who had used donor eggs and thus were unrelated to their offspring. These two groups were then examined for a correlation between the smoking behaviour of the mother during pregnancy and the birth weight and a measure of antisocial behaviour of their offspring. The findings were remarkably clear – smoking was associated with lower birth weight regardless of genetic relatedness. This effect is congruent with results from experimental animal studies on the effects of nicotine, cigarette smoke or carbon monoxide on birth weight and there are a variety of biological mechanisms postulated to explain the effect. So, all the evidence is consistent with this being a genuine effect of prenatal smoking per se.

But a very different picture was observed with respect to antisocial behaviour. High rates of antisocial behaviour were observed only in those mothers who smoked during pregnancy and who were related to their offspring. So, prenatal smoking itself does not seem to influence antisocial behaviour – it is more likely an indicator of some underlying genetic effect on behaviour of both the mother and the offspring. (See here for more on this).

So, smoking while pregnant is bad, mkay, for lots of reasons, but it will not make your child antisocial. And I would never argue against having books around, but articles proclaiming “Want smart kids? Here’s what to do” are uncritically promulgating an unfounded conclusion (also known as “talking shite”).

Evans, M., Kelley, J., Sikora, J., & Treiman, D. (2010). Family scholarly culture and educational success: Books and schooling in 27 nations Research in Social Stratification and Mobility, 28 (2), 171-197 DOI: 10.1016/j.rssm.2010.01.002

Rice, F., Harold, G., Boivin, J., Hay, D., van den Bree, M., & Thapar, A. (2009). Disentangling prenatal and inherited influences in humans with an experimental design Proceedings of the National Academy of Sciences, 106 (7), 2464-2467 DOI: 10.1073/pnas.0808798106


  1. That's really interesting, thanks. I hadn't thought of IVF as providing such research opportunities! Now that adoption is so much less common than it was 30 years ago, it's getting harder to do adoption studies, but IVF studies may be even better...

  2. "Adoption studies – especially of identical twins reared apart – do provide one way to dissociate genetic effects from those of the family environment. These have consistently found large effects of shared genes and very little effect of family environment on a wide range of behavioural traits."
    Careful here. This article:
    explains nicely why we should be extremely wary of heritability figures based on twin studies. As the author explains, traits like "wearing earrings" have very high heritability but this does not mean that this behaviour is a genetic effect.

  3. The article you cite makes a very valid criticism of the inference that because a trait is heritable within one population, and heritable in another population, that the difference in mean of a trait between those two populations must also be explained by genetic differences. That argument does not hold - it's completely spurious in fact, as the article nicely illustrates.

    The bit about earrings is also completely spurious though. Well-designed and appropriately interpreted twin studies, including adoption studies, are very well suited to distinguish the contributions of genetic and environmental variance to phenotypic variance (within the population studied, at that particular time). (Controlling for sex effects is an obvious and commonly employed method to get around that particular possible confounding factor).