November 10, 2004

Linkfest, focusing in the end on sex

Looking randomly around the lingua-blogosphere, I find a troubling case of possible overnegation, a memorial to the subjunctive, Igor's Theory of Negativity, a comparison between making post and getting tenure, a couple of posts on gamers' English, hormonal palmistry of language aptitudes, and a dicussion of what quibbling about the stars means to a (well, one) linguist.

What (little) I know about the stuff on sex hormones, finger lengths and cognitive profiles is from the work of Doreen Kimura (her home page is here). The serious part is the putative relation between sex hormones and cognitive skills; a good summary of her perspective is "Sex Hormones Influence Human Cognitive Pattern", Neuroendocrinology Letters 2002; 23(Suppl. 4):67-77. The finger length business is not entirely a curiosity, since to the extent that it really reflects prenatal sex hormone levels, it serves as a somatic marker that can interestingly be correlated with all sorts of things, including adult sexual orientation as well as various test results, life choices and so on.

The jargon for the difference is 2D:4D ratio ("second digit to fourth digit ratio"). It's easy to measure -- though there is surely a real possibility for observer bias to play a role in the measurements -- and the fact that collecting data is so simple strikes me as both an opportunity and a danger. It's easy to do little studies of this that and the other, and for people to do self-evaluations from which they (or their schoolmates) may draw invidious conclusions. In the (few) cases where I've looked at the data in detail, I find results like those in Q. Rahman and G.D. Wilson, "Sexual orientation and the 2nd to 4th finger length ratio", Psychoneuroendocrinology 28:3, April 2003, 228-303, where the differences were highly significant in the statistical analysis:

For right-hand ratios, there was a significant effect of sexual orientation (F=24.237, df=1, 239 P=0.000); homosexuals having lower right-hand 2D:4D ratios than heterosexuals. There were no significant effects of gender (F=0.115, df=1, 239, P=0.735), no significant interaction (F=1.684, df=1, 239, P=0.196) and no significant effects of the covariates (all Ps>0.10). Overall, the difference between homosexuals and heterosexuals constituted a moderate to large effect.

However, these highly significant differences in the mean value were nevertheless rather small, as a proportion of the means, and also relative to the variance:

2D:4D ratio
Standard deviation
Heterosexual males
Heterosexual females
Homosexual males
Homosexual females

In other words, the mean values for homosexuals and heterosexuals differed in their sample of males by 1 part in 100, and in their sample of females by 3 parts in 100; while the standard deviations of the measurements within each subgroup were 2 to 3 parts in 100. It bothers me that the newspapers (and even science magazines) who report this kind of stuff never try to explain this aspect of the results, which could easily be gotten across with histograms, or with scatter plots when the dependent variable is something like programming skill. The excuse for not doing this, I've been told, is that readers would be confused by the details; but an equally strong reason, I suspect, is that such explanations would undermine the apparently spectacular results ("you can tell someone's sexual orientation from their finger lengths!") with a small dose of reality ("no you can't, not with any accuracy in individual cases; the experiment showed only that you can distinguish a set of 60 homosexuals from a set of 60 heterosexuals on the basis of this measurement").

One could (and should) take the same care in presenting the results about demonstrated sex differences in cognitive skills, which are often of a similar nature.

Though it has nothing at all to do with sex, language, or for that matter with other aspects of cognition, I was also interested in this post on Galileo's middle finger.


Posted by Mark Liberman at November 10, 2004 11:51 AM