Thursday, June 29, 2006
Rikurzhen has pointed us to this paper in press at Intelligence by the Minnesota psychometrician Wendy Johnson and her colleagues. This is actually the latest in a series of technical papers analyzing the mental test scores of the twins, spouses, and siblings (biological and adoptive) who took part in the Minnesota Study of Twins Reared Apart (see here, here, here, and here).
I think the results of the first paper in the series are worth briefly highlighting. The Minnesota group administered three different IQ batteries to their sample. It turns out that the respective g factors extracted from all batteries are almost perfectly correlated. This is strong evidence that the g factor extracted from any large and diverse battery of mental tests is an estimate of a one true g. A continuing theme of these papers is an attack on the Cattell-Horn-Carroll model of the structure of mental abilities. Rejecting the prominent distinction in this model between fluid and crystallized g, they favor a model first put forth by Sir Cyril Burt and Philip Vernon which posits a verbal-educational and spatial-mechnanical split in the stratum below g. I will try to sidestep this issue in my selective summary of their most recent paper. I do not think the background of this minor controversy is necessary to understand the topic of this post, which is sex differences. Johnson and her colleagues extracted four dimensions of variation from their data: (1) a unipolar domain-general mental ability, that is, our old friend g; (2) a bipolar verbal-image rotation dimension; (3) a bipolar focus-diffusion dimension; and (4) a unipolar memory ability. By "bipolar" is meant a trade-off; for example, after controlling for g, individuals who tend to do well on verbal tests tend to do poorly on image rotation. "Focus-diffusion" refers to a trade-off between the ability to concentrate attention on a particular task and the ability to maintain holistic sensitivity to an overall gestalt. Males showed a small and insignificant advantage on g amounting to about 2 IQ points. Males tended toward the image rotation end of the second dimension, females toward the verbal end. Males tended toward the focus end of the third dimension, females toward the diffusion end. Females showed a significant advantage in memory. I admit that I'm having trouble seeing how this factor structure is consistent with their overall model. For this reason I think the results are best conveyed by the sex differences in residual test scores after controlling for g. The following are some representative standardized sex differences in the residuals of the regressions on age and first principal component scores (an estimate of g factor scores) with positive values indicating a male advantage: Spelling: -0.657 Arithmetic: 0.528 Information: 0.393 Word Fluency: -0.636 Associative Memory: -0.415 Coding: -0.826 Picture Completion: 0.807 Picture Arrangement: 0.396 Mechanical Ability: 1.431 (!) Block Design: 0.478 Cubes: 0.752 Mental Rotation: 1.039 The WAIS tests (Arithmetic, Information, Coding, Picture Completion, Picture Arrangement, Block Design) are described in an earlier post. The lower-order factors below g showed significant heritability. Effects of shared environment, on the other hand, were dropped without significant deterioriation of model fit. This is yet another item of evidence that the environmental causal agents distinguishing the sexes in mental abilities, if any, do not resemble those that are said to distinguish one family from another. |