More IQ studies: the quagmire thickens
And the fun part was that both articles contradicted my previous position, only in opposite directions.
My previous position, and scientific consensus a few years ago, was that variation in IQ is 50% genetic and 50% "non-shared environmental". Everyone knows what genetics are, but "non-shared environment" is the result of studies where they compare identical twins, fraternal twins, siblings, and unrelated people raised either together or apart. What they find is that approximately 0% of variation in IQ (or anything else!) is due to your parents and family and house ("shared environment") - this is the take-home point of The Nurture Assumption. The remaining 50% of variance isn't genetic or due to household environment, so without really having any idea what it is we call it "non-shared environment". People (especially me) assumed that this was things like what teachers you had at school, what friends you made, that study abroad experience in France your summer of sophomore year, et cetera.
So in the process of demolishing that, let's start with Shalizi. His article was 100% statistics, something I'm terrible at, so I had trouble following, but I gather his main point is that current estimates of the heritability of IQ, mostly produced from twin studies, show several signs of being overestimates. He gives a few reasons:
1. "Identical twins raised apart" may be raised "apart" in the same town, or even go to the same school. This means that of course they would be more similar than two randomly selected members of the population even for reasons totally other than genetics.
2. There are interactions and correlations between genes and the environment. For example, social class is an environmental factor, but we might expect people in certain social classes to have different frequencies of certain genes. And a gene that causes some small effect (like makes you slightly better at math) could then have an environmental effect out of all proportion (and so you get selected for the Math Gifted program and are taught math much more intensively).
3. Statistics things I'm not confident I understand well enough to write about. Apparently "heteroskedasticity" is a word now? Whatever they are, Cosma seems to demolish most heredity studies pretty thoroughly, and he gives a strong impression of being the smartest person in the world.
So he concludes that because of these points we can't even be sure that any of IQ is heritable. This might seem like an overreaction, but his numbers are sound (he points out, for example, that correcting twin studies for location can drop the IQ correlation from .8 to about .2) So who knows, maybe genetics is 0 and it's all just nonshared environment?
Except then comes the Turkheimer study, which shows quite conclusively that none of it is the usual type of non-shared environment we might expect.
Seriously. They look at all the different plausible sources of non-shared environment: birth order, parent behavior, friends, teachers, so on, so forth, and they all come out so close to zero as to make no difference. They claim that some of it might be gene-environment interactions, which I take to mean something like one child having a gene that makes her respond well to strict parenting, and another child having a gene that makes her respond badly to strict parenting, and then strict parenting appears to have no effect even though it actually caused two opposing effects. To me this seems like a stretch.
So none of it is heredity, none of it is shared environment, and none of it is non-shared environment. Maybe there are invisible gremlins who take IQ points out of a little bag and put them into people's heads? Or what?
Maybe it's not quite that bad. Both Shalizi's research and research from elsewhere on the Internet provides some tools we can use to build back some of heredity and nonshared environment (shared environment still seems pretty much screwed).
First, Shalizi notes that identical twins raised apart were still in the womb at the same time. If you read Biodeterminists' Guide, you know that's a big deal. He cites a study saying it could explain up to 20% of variance. That variance would look genetic on first glance, since the twins were raised apart and so supposedly not exposed to the same environmental factors, but it could actually be a part of non-shared (if you're not a twin) or shared (if you are) environment.
Second, a lot of studies might be confused because heritability of IQ is different in different conditions. Shalizi points to numbers saying it can be as low as 0 for poor people and as high as 0.8 for rich people. This makes sense. Suppose (this is my own interpretation, not Dr. Shalizi's) that everyone had a natural potential determined by genetics, but that various harmful factors like the ones in the Biodeterminist's Guide could knock off points. In that case, if you're poor and exposed to many such factors, the number of environmental factors you were exposed to would be the main source of variance, and if you're rich and sheltered from such factors, your genetic potential would be more important.
Likewise (this was in a link of Shalizi's rather than the paper itself) heritability of IQ is different at different ages. That is, to the weak degree you can even measure IQ in a three year old it's not very heritable, but if you measure the person again at 30 it will correspond with parents' scores better. That makes sense; at 3 you would expect. It might be possible to pull out actual heredity numbers in specific cases that don't fail statistically as bad as some of the others.
Third, other sources suggest that school might not be as unimportant a factor as the behavioral geneticists seem to think. There are a few papers showing 4 or 5 point lasting IQ gains for people who get very intensive early interventions (usually putting the children of mentally disabled people into a special school - and no, this isn't just regression to the mean, they had a control group). Shalizi even posts a claim from a book called Genetic Diversity and Human Equality, which claimed "an experiment on ghetto children whose mothers had IQs of below 70. Some of these children received special care and training, while others were a control group. Four years after the training period the IQs of the former averaged 127 and those of the latter 90, a spectacular difference of 37 points." I don't have the book (it's page 14-15 if any of you do) but if Shalizi believes that I might have to revoke his "apparently the smartest person in the world" status.
Fourth, Molenaar et al find us our invisible gremlins. They say that chaotic developmental factors may have butterfly-effect type downstream actions on large-scale features, like intelligence. So two million steps later, the direction one neuron started growing in the womb might randomly grant or cost you several IQ points and change your personality. This could explain some of the rest of non-shared environment.
And even though Cosma is very impressive and I'm reluctant to critique an essay I don't fully understand, I can't agree that we know nothing. For example, studies directly comparing identical vs. fraternal twins in the same household should escape many (though not all) of the problems he finds with twin adoption studies, but these reinforce the conventional wisdom. Genetic factors become more apparent with age and status, which seems more consistent with them being real effects than signs of a poorly run experiment (it also significantly raises Cosma's "best estimate" of .34 for certain cases that we probably care about disproportionately). Because shared environment is so minimal, it's hard to argue that it's a big confounder. And other tests for other cognitive abilities don't show the same amount of heredity in twin study, which means either IQ is uniquely confounding or highly heritable.
Overall I respect his statistical purist point of view that our current statistics aren't good enough to be very confident in the heredity of IQ, while also respecting everyone else in the field who says their best guess is around 0.5.
Right now if I had to guess, I would say within group variance in IQ is due 30 - 50% to genetics, 20% to in utero environment, and the remaining 30 - 50% some combination of chaotic factors and schooling/social development. If I wanted to hedge my bets maybe I'd add that the schooling/social development factor is higher during childhood, and the genetic factor is higher during adulthood.
Between groups is a totally different kettle of fish, and I'm much more skeptical of the idea that there are significant genetic between-group differences than I was yesterday, although I'm sure once West Hunter makes their next good point I'll switch right back. But that's a different post.
And for either kind of variance if you told me it was actually the gremlins I wouldn't be a bit surprised.
(I also read Cosma's other essay on g. It made sense and was convincing, but I'm less interested in it from an "oh my, what real neurological concept does this correspond to" perspective and more interested in the fact that IQ tests are clearly measuring something and we can use that to find things out about this mysterious idea of "intelligence" even if it turns out to be something else entirely.)