Monday 27 January 2014

Genius in the genes

A recent article “Is genius in the genes” Times Educational Supplement, 24 January 2014 by Prof Rose raises three arguments against the view that genetics has a significant effect on intelligence:

1 That although contemporary studies of the genome can account for 40-50% of the variance in intelligence in very large “samples of discovery”, those particular findings account for no more than 3% of the variance when tested on other large samples. That is, they do not replicate well.

2 That one particular paper has shown that heritability of intelligence is high in wealthy families (60%) and low in poor families.

3 That much of the effect on intelligence is not due to genetics but to epi-genetics.

In return, I would like to make three observations:

A) That intelligence is heritable has been shown for over six decades, simply by comparing the high concordance in ability of identical twins with the lower concordance of ability in fraternal twins. Genetic similarity is the most plausible cause of this intellectual similarity. Other techniques, including comparing the intelligence of distantly related persons with those of entirely unrelated persons sustain the view that relatedness in genetics leads to similarity in intellectual power. Geneticists have shown that there is an effect across a wide range of consanguinity, but they are still searching for the next step, in which they show exactly how this comes about. Heritability estimates measure the extent of the genetic effect, genomic studies try to identify genetic causes. “Environmentality” estimates still stand, despite the lack of detailed knowledge as to how family and social effects may influence intelligence (perhaps the number of words spoken by mother to child, and the number of books in the house). Similarly, heritability estimates still stand.

B) Turkheimer et al (2003) Psychol Sci. 2003 Nov;14(6):623-8. Socioeconomic status modifies heritability of IQ in young children.

This paper has been quoted at me many times. No problem with that, but one has to read the paper carefully to see if it merits the particular emphasis which is being placed on it. Here is the abstract:

Scores on the Wechsler Intelligence Scale for Children were analyzed in a sample of 7-year-old twins from the National Collaborative Perinatal Project. A substantial proportion of the twins were raised in families living near or below the poverty level. Biometric analyses were conducted using models allowing for components attributable to the additive effects of genotype, shared environment, and nonshared environment to interact with socioeconomic status (SES) measured as a continuous variable. Results demonstrate that the proportions of IQ variance attributable to genes and environment vary nonlinearly with SES. The models suggest that in impoverished families, 60% of the variance in IQ is accounted for by the shared environment, and the contribution of g enes is close to zero; in affluent families, the result is almost exactly the reverse.

Turkheimer et al. begin by noting a difference between correlational studies of inheritance of intelligence (which suggest strong genetic effects) and mean group difference studies of intelligence which suggest an environmental effect of training programs and adoption. They suggest that this may be due to non-linear effects, namely that bad environments have a bigger effect on intelligence, and that once environments are adequate, no further environmental boost takes place. In his very useful summary of the field Earl Hunt (2010) “Human Intelligence” suggests that the best approach is to always give means and correlations, because both are informative. I would go further back into statistical history, and like A.E. Maxwell (1972) “Basic statistics for medical and social science students” ask for a simple plot of the raw data so that we can judge for ourselves what they look like. Statistics are summaries, after all.

The study was conducted on 319 twin pairs of whom 114 were monozygotic and 205 were dizygotic. That is an extraordinarily high number of identical twins. Monozygotic births are usually 3 per 1000, dizygotic births 33 per thousand, a tenfold difference. Something has boosted the number of monzygotics in the sample.

The twins were classified as 43% White, 54% Black, and 3% “other.” The sample included a high proportion of impoverished families. The median number of years of education of the head of household was between 10 and 11 years. The median occupation was “service worker”; 25% of the household heads received occupational ratings of “laborer” or lower, including 14% with no occupation. The median family income was between $6,000 and $7,000 annually, equivalent to $22,100 in 1997 dollars, the most recent year for which an equivalent scale was available. Twenty-five percent of the families had incomes below the 1973 poverty level for a family of four (U.S. Census Bureau, 2002). Opposite-sex pairs were combined with the same-sex pairs in all analyses.

This is a study which should have shown the average IQ for the above classifications. In that way we would have been given points of reference to compare this paper with others, and to help us understand the later statistical analysis.

In fact, if you look back at the Sandra Scarr adoption studies, her preliminary results at 7 years suggested enormous adoptive family effects but the follow up at age 17 showed reduced effects, and stronger correlations with the abilities of the genetic parents. Usually, all studies show a bigger effect of family life on 7 year olds than 17 year olds, which itself argues against the idea family life is a major crucible of intellect.

So, this is a study which in my view should have given more details about means, standard deviations and correlations, which would have made it easier to link up with previous published work. It was conducted at age 7 which would have maximized the effects of family life. The sample size is relatively small by modern standards of genetic research, though is pretty healthy by the less demanding standards of psychological research. There is an extraordinarily high number of monozygotic births. There is no racial breakdown of results, which is a great pity, because it would add good data on child development for different racial groups in America. If, as seems likely, the Black kids were mostly in the poor category with bad family environments (remember that SES bundles together genetic and environmental variables) and the White kids were in the adequate to good environments, then the finding would be tantamount to saying that at age 7 the effects were thus: Black children were largely influenced by their poor environments, White children largely influenced by their genetics (and their at least adequate environments). If the data have been re-analysed or extended in subsequent publications then the picture might become clearer.

In summary, though the finding makes sense, in that bad environments seem to have (deleterious) effects, and adequate to good environments neutral to slightly positive effects, I think it would be better to get more detail about the sample means and correlations, and the distribution of Black and White scores before according this study the status of having major implications for the heritability of intelligence.

C) Epi-genetics. There are certainly some epigenetic effects in mammals, but it remains to be seen how substantial they are. For example, if such effects were substantial then they would reduce the predictable effects of pure genetic transmission, and should reduce heritability estimates. Particularly, the difference between one egg and two egg twins should be very much reduced by things happening outside the genome. Nonetheless, heritability estimates remain high, particularly in adequate to good environments. Further research may improve our understanding, but it is hard to see how it could become a major source of variance.


I think that genetic studies are making all of the running at the moment. Environmental studies have suggested some environmental mechanisms which might boost intelligence, but they are very far away from finding the underlying “memes”. In some ways, the present difficulty in finding exactly how genes account for much of intelligence is reassuring: avoiding false positives is not usually achieved in psychological research. It is good to see geneticists bringing big samples to bear on the heritability of intelligence, and making sure they publish their attempted replications promptly, showing that they have yet to produce replicable results.

The search continues.


  1. The study was conducted on 319 twin pairs of whom 114 were monozygotic and 205 were dizygotic. That is an extraordinarily high number of identical twins. Monozygotic births are usually 3 per 1000, dizygotic births 33 per thousand, a tenfold difference.

    That's not right. The ratio of 2 DZ to 1 MZ pairs is quite normal.

  2. Rose's article can be read here. With respect to GxE, this study is much more pertinent than Turkheimer's which was based on data collected, IIRC, 40 or 50 years ago. Plomin et al. found no evidence of moderation of genetic influence by socioeconomic status in the comtemporary UK, although there was some evidence of greater shared environmental variance among the poor.

    There are a number of dishonest and misleading statements in Rose's piece. For example, he dismisses MZA studies by reference to Burt's apparent fraud, but is silent on the five other, independent MZA studies that produced results very similar to Burt's.

    Rose's claim that if environmental influences were equal, heritability would be 100% is wrong, too. Firstly, there's always some measurement error. Secondly and more importantly, a subtantial part of non-genetic variance for many traits is not due to environment but due to developmental noise.

    His remarks on topics like g and the shared/unshared environment distinction are baseless. He does not even try to tackle the empirical facts.

    His comments on epigenetics make little sense. He does not even define the term, which can mean just about anything. In any case, epigenetics is unlikely to bias any heritability estimates.

    Still, it seems that over the years the consistent confirmation of genetic hypotheses in behavior genetics has had some effect on Rose. His opposition to hereditarian explanations in this new article is rather half-hearted compared to what it used to be.

  3. "this is a study which in my view should have given more details about means, standard deviations and correlations": it doesn't say much for the standard of refereeing in your trade that they were allowed to get off with suppressing or omitting such important stuff.

  4. The low heritability of intelligence in poor families is most likely due to their higher levels of non-paternity

    1. Good point, but severely weakened because both identical twins will have the same father. How about fraternal twins? That depends not just on the bastardy rate but on the number of women who have simultaneous men in their lives. Is there any data, even anecdotal, on that?

  5. In addition to the above comment about Plomin's study. I wanted to make two quick points. Firstly Plomin's study, really Handscombe's study, suggests that trying to determine these effects with less than 2500 pairs is problematic. It also has a literature reviews of a roughly even number of papers that have, haven't found this effect and notably the smaller papers were more likely to find it. Secondly a more recent paper by Maciej TrzaskowskIi et al, investigating the genetic factors for SES/IQ, suggests using SES as an independent factor is more complicated than the Turkhiemer paper allows.

    1. Thanks. Yes, SES is a mixture of effort and opportunity, not a pure external measure of social allocation.

  6. This comment has been removed by a blog administrator.

  7. Is the 2004 Turkheimer data not available publicly? I can't think of any good reason why it shouldn't be--- they've had 10+ years to write papers on it themselves. The American Economic Review now requires authors to publicly post their data when their paper is published, unless there is a confidentiality problem. Other fields would do well to imitate.

  8. I might add that sending your data to anyone who requests it has long been the norm in economics, even when a journal doesn't require it, though the authors may delay it a short while to give themselves a chance at writing two or three papers on it first.

  9. This comment has been removed by the author.

  10. Publishing one's data promptly would solve a lot of problems, and help psychology proceed more quickly.Perhaps doing so should count as ten publications! Researchers need incentives, just like everyone else

  11. It would actually be pretty easy for a salary or a tenure committee to see the impact of publishing your data--- just see if other scholars are using it for their research. If they do, that can be as valuable as being a co-author---- you are a necessary condition for that paper being written. Coming up with a new technique everybody uses is more glamorous, though.