February 17, 2008

60%, 0%, whatever

In yesterday's post on "Behindology", I suggested that the Italian national pastime of dietrologia -- the "technique of the double, triple, quadruple hypothesis" that aims "to detect, behind the apparent causes, true and hidden designs" -- is nothing but Theory of Mind (ToM) gone wild. (Or perhaps it's ToM responding appropriately to Italian socio-political realities, I'm not sure.) In any case, the reason that we've discussed ToM some 30 times in other LL posts should be obvious from the definition that I gave a few years ago ("Mind-reading fatigue", 11/8/2003):

Theory of mind is a term introduced by Premack and Woodruff (1978) to refer to a set of abilities that may be uniquely human: to attribute mental states such as beliefs, knowledge and emotions to self and others; to recognize that the mental states of others many differ from one's own; to use these attributed states to explain and predict behavior; and to predict how such mental states would be affected by hypothetical actions.

Humans may not be the only animals with ToM abilities -- it's surprisingly hard to tell -- but I think everyone agrees that we're a lot better at this than chimps and gorillas. So it's not surprising that there's been a lot of interest, over the past couple of decades, in the idea that ToM abilities might be an evolved cognitive specialization, a sort of "mental module". It's been suggested that this ability is neurologically localized (the paracingulate cortex has been mentioned), and that at least some parts of the autism spectrum might be related to ToM deficits.

As a result, there are many reasons to be interested in "twin studies" that are designed to tease apart the genetic and environmental influences on ToM abilities. And if such studies are set up to distinguish ToM abilities from general verbal abilities, so much the better.

Unfortunately, as we've mentioned a number of times recently, such studies are quite difficult to interpret. And in the course of looking for something else, I recently stumbled over a really suprising example of these problems.

I'm going to leave the detailed analysis for another post. But today, I want to set the stage by quoting the quantitative conclusions of two studies with the same first author, published six years apart, which used the same experimental design (ToM and verbal IQ tests on monozygotic  vs. dizogotic twins) and the same statistical method (analysis of variance), but came up with radically different estimates of the genetic contribution to individual differences in ToM abilities.

Claire Hughes & Alexandra Cutting, "Nature, nurture and individual differences in early understanding of mind" (Psychological Science, 10, 429– 432, 1999) summarize their results like this:

The model attributes the variance in understanding mind to substantial genetic influence [60%], negligible shared environmental influence [7%], and moderate nonshared environmental influence [33%]. Of particular interest are genetic influences on theory of mind and verbal IQ. Two thirds of the genetic variance in theory of mind was independent from genetic variance in verbal IQ (.632/(.452 + .632)). In other words, only about a third of the genes influencing theory of mind also influence verbal IQ.

Smells like an evolved mental module, right?

But now read the statistical summary from Claire Hughes, S.R. Jaffe, F. Happe, A. Taylor, A. Caspi & T. Moffit, "Origins of Individual Differences in Theory of Mind: From Nature to Nurture", Child Development 76(2): 356-370, 2005.

As Table 1 shows, MZ and DZ correlations for ToM were identical (r=.53), suggesting substantial shared environmental influence but negligible genetic influence on individual differences in ToM. Table 2 summarizes the goodness-of-fit statistics and parameter estimates from the quantitative genetic modeling of these data. The proportion of variance accounted for by the latent genetic and environmental factors can be calculated by squaring each of the parameter estimates. For example, genetic influences accounted for 7% of the variance in children's ToM (i.e., .26 × .26). The strongest influences on individual differences in ToM were shared and nonshared environmental factors, which accounted for 48% and 45% of the variance, respectively.

Because genetic influences on ToM were nonsignificant in the full univariate model, we tested the fit of a more parsimonious model in which these genetic effects were hypothesized to be zero. The fit of the reduced model was not significantly different from the fit of the full model, χ2diff(1)=.81, ns. Thus, genetic factors do not account for significant variation in 60-month-old children's ToM.

And the summary from the same paper of the results of fitting a bivariate model:

In this large sample of 1,116 pairs of 60-month-old twins, 44% of the variation in ToM scores was accounted for by ToM-specific nonshared environmental influences, 20% by ToM-specific shared environmental influences, 21% by common shared environmental influences on ToM and verbal ability, and 15% by common genetic influences.

[Note that "shared" here means "familial"; and "common" means "influencing both verbal ability and ToM to the same degree".]

So quantitative estimates for the proportion of variance due to genetic influences (whether common to ToM and verbal ability or not) range from 0% to 60%, with stops along the way at 7% and 15%. (Another study, which I've spared you, has other intermediate values as well.)

What's going on? The main difference in the two papers is the subject pool: the 1999 paper involved 119 pairs of 4-year-old twins, apparently selected without any special demographic criteria; while the 2005 paper was based on 1,116 pairs of 5-year-old twins, selected to be representative of the SES demographics of the UK as a whole.

Why should this have made such an enormous difference? Stay tuned to find out.

[ Cosma Shalizi writes:

Actually, substantial genetic variance is exactly what I would _not_ expect from an _evolved_ mental module. Imagine a trait under directional selection, say positive (to map on to the ToM case). Since high ToM scorers have higher fitness, and high ToM scores are at least partly genetic, the next generation is disproportionately descended from those with high-ToM-score genes, and their genetic variance in ToM is reduced. Iterate until remaining differences in ToM no longer have substantial fitness effects, and you are left with a very small amount of genetic variance in ToM. I suppose its possible that the environmental variance could be shrinking as well, so as to keep the heritability constant or even increase, but the magnitude of the genetic variance should be small either way.

Alternately, one could conjecture that ToM is not an evolved but an _evolving_ mental module, _recently_ brought under selection. But this gets us into Julian Jaynes territory...

Certainly an ability that's never been under selective pressure would be expected to show quite a bit of genetic variation. So the fact that a DZ/MZ twin study shows "substantial genetic influence" on individual variation in performance is a lousy argument for past adaptation, since a trait that selection has never operated on ought to be quite variable.

And as Cosma notes, genetic influence on a variable ability is also consistent with evolution in progress. It seems plausible that ToM abilities are still under active selection, since there's quite a bit of individual variation, and clearly some effects on survival and reproductive success. And we don't need to get into Julian Jaynes territory to think that ToM evolution is still underway -- this process might have been going on for several hundred thousand (or even several million) years. (Jaynes believed that consciousness originated less than 3,000 years ago, when the invention of writing fragmented the "bicameral mind", the natural human condition in which language had evolved to communicate racial memories from the right hemisphere to the left hemisphere of the cerebral cortex, which our ancestors interpreted as advice in the voices of the gods. Really.)

But finally, for many evolved traits, the relevant fitness landscape is complex, so that the target of adaptation will not be a genetically uniform population. In the case under discussion, perhaps there are trade-offs between ToM abilities and susceptibility to paranoia, or self-esteem problems, or hierarchical position, or whatever, so that the equilibrium situation retains a lot of genetic variation.

So all this means that "substantial genetic variation" in individual abilities of a given sort is consistent with pretty much any evolutionary status -- no selection, selection far from equilibrium, selection at equilibrium, past selection no longer active, and so on.

In this case, we start with an experimental and statistical mystery: how can one study can find 60% of ToM variance accounted for by genetic factors, while another study (with essentially the same design, the same test instruments, and the same statistical modeling, but just a different sample) yields an estimate of 7%?

But whether the answer is 60% or 7% or (somehow) both (e.g. because of radical differences in the subject pools), Cosma is right, we need a lot more information before we can come to any conclusions about the evolutionary interpretation. ]

Posted by Mark Liberman at February 17, 2008 09:46 AM