In that piece, Ta-Nehisi writes of his having mostly avoided the race-intelligence pit o’ fail, because he does not to his own satisfaction know enough about the technical arguments there. As John says, that doesn’t prevent Ta-Nehisi from accurately diagnosing the moral bankruptcy of Sullivan’s thinking.
But his framing of the post did drive me back to one of the best web sources of insight on issues of method and meaning in statistical reasoning, the invaluable Cosma Shalizi, proprietor of the Three Toed Sloth blog. Given that Ta-Nehisi provided an implicit shout out for such sources, I thought I’d dig back into Cosma’s work to provide some context.
Back in 2007, when William Saletan decided to trumpet the “work” of notorious race/IQ “scientist” J. Philippe Rushton, in seeming ignorance of Rushton’s long and well documented record as an associate and aider-and-abetter of overtly racist segregationists,* Cosma wrote a series of long and very carefully reasoned posts explaining what’s goes terribly wrong with most writing about IQ and anything — not just race. For an overview of the problems with concept of “g” — a general intelligence trait — see this honking elephant of a post. I can recommend two much shorter and nicely wicked dialogues on the question of intelligence and plasticity — or, really, on the ease with which really bad thinking slips into such questions.
The point of all this, worked through in some detail in the third of his four posts on the subject, is that studies of intelligence that assert “truths” about connections between some conception of the heritability of IQ and race do so from a position of what may most kindly be assumed to be simple incompetence — though in some cases a presumption of malice seems justified.
As Cosma himself warned his readers, any honest confrontation with the methodological hell that is IQ research leads one into a quagmire. I think I’ve pointed you towards close to 20,000 words (and figures) of his attempt to navigate the mire. So, if you want to skip the some thousands of words towards which I’ve pointed you, here are his conclusions in nice, compact form:
- The most common formulae used to estimate heritability are wrong, either for trivial mathematical reasons (such as the upward bias in the difference between monozygotic and dizygotic twins’ correlations), or for substantive ones (the covariance of monozygotic twins raised apart neglects shared environments other than the family, such as maternal and community effects).
- The best estimate I can find puts the narrow heritability of IQ at around 0.34 and the broad heritability at 0.48.
- Even this estimate neglected heteroskedasticity, gene-environment interactions, gene-environment covariance, the existence of shared environment beyond the family, and the possibility that the samples being used are not representative of the broader population.
- Now that people are finally beginning to model gene-environment interactions, even in very crude ways, they find it matters a lot. Recall that Turkheimer et al. found a heritability which rose monotonically with socioeconomic status, starting around zero at low status and going up to around 0.8 at high status. Even this is probably an over-estimate, since it neglected maternal effects and other shared non-familial environment, correlations between variance components, etc. Under such circumstances, talking about “the” heritability of IQ is nonsense. Actual geneticists have been saying as much since Dobzhansky at least.
- Applying the usual heritability estimators to traits which are shaped at least in part by cultural transmission, a.k.a. traditions, is very apt to confuse tradition with genetics. The usual twin studies do not solve this problem. Studies which could don’t seem to have been done.
- Heritability is completely irrelevant to malleability or plasticity; every possible combination of high and low heritability, and high and low malleability, is not only logically possible but also observed.
- Randomized experiments, natural experiments and the Flynn Effect all show what competent regressions also suggest, namely that IQ is, indeed, responsive to purely environmental interventions.
In other words: Sullivan is not just wrong; he is not only trafficking in just the all-dressed-up-for-the-21st-century version of the slave-holder’s self justification — as Ta-Nehisi so clearly demonstrated; he is not playing the honest broker, merely seeking research to settle vexing questions; he is writing in what can at this point only be chosen ignorance of what those with actual understanding of the methods and pitfalls of social science have been trying to tell him for lo-these-many years.
I know that I sometimes praise Sullivan for his moral acuities: his strong stand against torture, his long advocacy of civil equality in the GBLT arena, his belated and partial recognition of the flaws of the Republican party. But when you stack up his role in creating myths about health care in the Clinton years, and his championing of this kind of stuff, I don’t see how you can count his career as a net positive for the society.
*My only Rushton story: I was in the room at a mid-’80s AAAS meeting when Rushton was just breaking on the scene with his attention grabbing assertion that intelligence was inversely correlated with penis size, which was in turn specifically correlated with particular races. It was a day that made me proud to be a science writer, as the great SF Chronicle¹ science editor David Perlman got up in that session and just roasted Rushton. None of this, “some say, others disagree” transcription-monkey reporting. Just a thing of beauty and a great lesson in the need to build and pay attention to one’s bullship filter. That is, of course, exactly what Saletan failed to do.
¹Yes. That paper, which, like lots of Bay Area folk, we regularly referred to as the SF Comical, did/does have some truly first class writers and reporters, among whom Perlman is one of the best.
Image: Adriaen Pietersz.van de Venne, Fools Have the Most Fun, 1661