Scott Alexander has a post up that is getting a lot of attention, Non-Cognitive Skills For Educational Attainment Suggest Benefits Of Mental Illness GenesNon-Cognitive Skills For Educational Attainment Suggest Benefits Of Mental Illness Genes. Scott concludes:

But the big shock here is schizophrenia. As of last time I checked, the leading hypothesis was that schizophrenia genes were just really bad, evolutionary detritus that we hadn’t quite managed to weed out. And although they definitely decrease IQ, they seem to be good in other ways. Not with certainty: the correction for false discovery rate kills a lot of the effect (though this is the question I would have been most interested in before reading these results, so maybe I can ignore that?). But there’s at least a faint signal here.

I’ve actually heard from psychometricians that there is a weird signal in relation to the genetic correlation between schizophrenia and various outcomes which are not all negative. One model for schizophrenia you might have is that it’s totally bad. But, another model might be that it’s “good actually” in small dosages or in some genetic backgrounds, but bad in other cases. The argument can be generalized to many “mental illnesses.”

The issue more broadly is that the genetic architecture of mental/behavioral traits is complex, and often subject to a normal distribution. Is this simply mutation-selection balance? Or is there balancing selection? There is now evidence that markers associated with homosexual behavior boost fertility when present in individuals who are not gay. This could be a simple way that these alleles remain in the population at a given frequency (assume frequency dependence). As we understand the genetics of schizophrenia and autism, and begin to screen embryos, I think these questions will become more relevant.