Testing The Race of Mother Hypothesis: A Technical Review
My paper has been finally published. The review thread can be accessed here. I already discussed the technical issues in the paper but in this post I am going to extend a bit more. Still, you may want to read my post on Human Varieties first.
Starting with the CPP, the paper reported the racial groups' scores without disaggregating by marital status, unlike what Willerman et al. (1974) did. First, the sample was small. Second, "controlling" for marital status may further partial out genetically-induced traits. Controlling for SES was already a doubtful procedure to begin with. Yet, reporting cognitive scores before and after controlling for SES allows us to evaluate its effect across racial groups. For example, if parenting was strongly correlated with SES, but SES has no impact on group differences, it may suggest that parenting isn't as important as environmentalists claimed.
But because I wanted a complete analysis, I reported the aggregated and disaggregated scores by marital status for every groups. The spreadsheet reports everything you need to know. The fact that the IQ gap did not reduce among unmarried women, within the Black-White mating, is hard to interpret. Because I used listwise deletion, those unmarried women are living with a husband (the spreadsheet displays the same combined Ns). It cannot be argued that the unstable home environment of the single mother produces such IQ deficits. A more likely explanation is calling for sampling error. In the spreadsheet, you will notice that unmarried and married Black women had children with respectively 100 and 97.8 IQ points at age 4. This doesn't make any sense from either the hereditarian or the environmental hypothesis. This point is important. Unfortunately, I realized now I forgot to mention it in the paper.
With respect to the Add Health, the usage of the wrong sampling weight variable, or lack of, greatly affects the outcome, and therefore the conclusion, especially for the Black-White verbal gap. I do not know how the result is affected if the complete sample (using the private Add Health data) is analyzed. But considering the mother's effect across data is generally weak, I am still confident about the Add Health result. Of course, it's not impossible that verbal IQ would show a different pattern. More data is needed.
The pattern of the cognitive gap across data seems to vary in accordance with both parents' education levels. One pattern which strikes the most among education levels, is the one from the CPP. It shows that Black mothers with White fathers had lower education levels than White mothers with Black fathers. But in the Add Health and the HSLS, it was the opposite. I also mentioned it in the paper, but Chiappori et al. (2016) analyzed the PSID and found that Black mothers who intermarry typically had higher education levels than White mothers who do. Considering the explanation given by these authors makes perfect sense, I came to the conclusion that the CPP was less representative than the other two. And as it was clear from the data description itself, the study generally focused on women with health complications.
I was asked to perform a meta-analysis using the Inverse Variance Method despite technical issues related to sampling weight downward-biasing the standard errors. Fortunately, I learned it was possible to compute the standard errors with Bootstrap without using sampling weights. Since those have non-integer values, Bootstrapping could cause serious bias. The spreadsheet shows an alternate meta-analytic method, using sample size as weight instead of standard errors. The result is still consistent with the one reported using the Inverse Variance Method.
Chiappori, P. A., Oreffice, S., & Quintana-Domeque, C. (2016). Black–white marital matching: race, anthropometrics, and socioeconomics. Journal of Demographic Economics, 82(4), 399–421.
Willerman, L., Naylor, A. F., & Myrianthopoulos, N. C. (1974). Intellectual development of children from interracial matings: Performance in infancy and at 4 years. Behavior Genetics, 4(1), 83–90.