The January 2005 issue of American Psychologist was devoted to the subject of race. As Steve pointed out at the time, most of the issue was complete fluff, with the exception of the paper by David C. Rowe. Rowe had died 2 years earlier, and wrote the paper during the convalescence prior to his death.

The article makes several points, which might have been better presented as a series of article. We can forgive Rowe for not taking the time. The final point is an outline of a research project that could "convincingly demonstrate a genetic origin of racial [IQ] differences." Rowe first suggests the use of multilocus DNA tests to estimate individual ancestry among African Americans. The measures of individual ancestry would then by compared with IQ scores and other measures to control for social/cultural confounding (Rowe suggests skin color). Moving beyond bulk comparisons of individual ancestry, Rowe suggests that MALD (mapping by admixture linkage disequilibrium) be used to associate ancestry at individual regions of the genome to IQ. MALD is the technique that Rowe suggests would "convincingly demonstrate a genetic origin of racial differences." I agree that this would work, but I don't know if any research finding could "convince" people.

R.C. Cooper thinks this is a terrible idea. In fact, Cooper seems to say that everything about Rowe's suggestion is a bad idea, including the utility of MALD. This appears to be a version of the Cavalli-Sforza squid ink effect. Cooper has subsequently published studies using MALD to examine hypertension among African Americans. If executed, Rowe's proposal would taint Cooper's research platform with charges of racism. His kitchen-sink refutation of Rowe's proposal makes sense in this light.

Related: Thompson and Gray: Neuroscience, genes, and IQ

Update by Darth Quixote: Another important aspect of Rowe's last publication is its description of an earlier study applying structural equation modeling (SEM) to black-white differences in mental test scores. See the notes here for a description of the measurement equations whose parameters are estimated by SEM. Think about the intercept in multiple regression; it should be easy to see that the measurement equations can also incorporate mean differences between groups.

The tests were the Mathematics, Reading Comprehension, and Reading Recognition scales of the Peabody Individual Achievement Test. The sample consisted of white and black full-sib and half-sib pairs in the NLSY. The observed test scores were modeled as indicators of latent genetic and environmental factors. In the best-fitting model, blacks averaged lower than whites on both the genetic and environmental factors; in fact, the black deficit in the environmental factor was greater than the deficit in the genetic factor. (It is widely accepted that the shared environment can affect observed test scores in children.) However, the tests were more sensitive to the genetic factor than the environmental factor, resulting in genetic differences between blacks and whites accounting for 74 percent of the black-white gap in Reading Recognition, 66 percent of the gap in Reading Comprehension, and 36 percent of the Mathematics. The conclusion tentatively supported by this study is clear: if the causes of the black-white gap in mental test scores are a subset of the same causes that differentiate individuals within either race, then genetic differences between the populations account for a nontrivial portion of the gap.

But this kind of methodology represents the past, although Rowe predicted that it would be useful for some years to come. MALD, and other genomic methods, represent the way forward.