This paper, by Phil Rushton and the late psychometrician Douglas Jackson, reports on the basis of SAT data that males average more g than do females. So apparently Rushton is joining Richard Lynn, Paul Irwing, and Helmuth Nyborg in promoting a male advantage.

I have reservations about this paper. The authors use principal components on item data, but in my own experience the first unrotated principal component is readily contamined by group factors. On the SAT-M especially, this may lead to a spurious male advantage because of the male edge on the spatial/quantitative factor that the SAT-M measures above and beyond g. But the authors find a comparable sex difference on the g that they extract from the SAT-V as well, so perhaps we don't have to worry too much about this. It also seems to me that principal components is too far from the state of the art for the analysis of item data. PC models the probability of a correct response to a given item as a linear function of the latent trait, which is often a poor approximation. To get an intuitive grasp for why this is so, realize that for a really high-g subject PC might give a probability of passing a given item greater than one, or for a really low-g subject a negative probability. The state of the art for addressing the question of sex differences with these data, it strikes me, is multidimensional item response theory (which forces the probability of a correct answer to asymptote to one at the high end and to chance level at the low end). On the other hand, it is hard to see why the model imprecision should tend to favor males in the robust and consistent manner found in this paper.

Caveat: I am not a psychometrician, so my worries may well be baseless. Presumably, the referees who gave a thumbs-up to this paper know more than I do.

Another potential objection is that SAT takers are a selected sample of the population. If males have greater variance in g, they will overpopulate the left half of the bell curve as much as they do the right half. The point dovetails well with this paper by Rosalind Arden and Robert Plomin replicating the finding that boys do indeed show greater variance in mental ability (something denied by Elizabeth Spelke). Interestingly, they also report that in their sample boys creep ahead in the mean as well after lagging for most of the preteen years.

Finally, the current issue of Behavior Genetics is a special devoted to linkage QTL studies. Three of the reports are of mental ability. Encouragingly, an American study found a signal (albeit not significant by conservative standards) for a QTL linked to Full Scale IQ on chromosome 6 that is very close to a signal from a combined Dutch and Australian sample that was reported by Nick Martin's team last year in American Journal of Human Genetics. In my book, this result should count as a replication. Yukon ho! There are some other converging findings as well, although the overall picture is still somewhat murky.

Update: JP of The Contingency Table has alerted us to a new paper finding an association between dysbindin-1 and a (probable) measure of g. You can find it in the GNXP Forum. The finding has flaws, as JP points out at his blog and in the comments, but I think it is worth getting at least mildly excited about because dysbindin-1 is right at a suggestive QTL signal found in two different studies. Props to JP.