Monday, April 21
Grant Miller: Managerial Incentives in Public Service Delivery
Kaufman, J.S., Arline T. Geronimus, and Sherman James. 2007. "Faulty interpretation of observed racial disparity in recurrent preterm birth." American Journal of Obstetrics and Gynecology, 197(3): 327-327.
Kistka et al1 reported that black women in Missouri had a higher risk of preterm delivery and recurrent preterm delivery than white women and that these differences persisted after adjustment for some variables available from birth certificates. Aside from the statistical error of not accounting for clustering by the mother in the variance estimation, the analysis is unremarkable. What is problematic is the authors’ interpretation. The adjusted odds ratio for black race represents the contrast conditioned on measured predictors and demonstrates a persistent excess risk for black women. This excess risk is caused by factors that were not measured, but the authors vigorously asserted that these unmeasured factors are most likely genetic.
This interpretation is fallacious because adjusting for some small set of crudely measured social or environmental variables does not imply that other variables from this general class are balanced across the racial groups.2 The argument is especially suspicious here because the modeled covariates were only those available from birth certificates. For example, social status was assessed only by dichotomized education and public assistance program participation. Arguing that conditioning on these few factors makes black and white women indistinguishable with respect to other important social exposures is not credible.
The authors based their assertions about the importance of genetic mechanisms primarily on results for recurrence. These results suggest only that the etiologically important unmeasured factors are stable over time, however, not that they are genetic. The authors’ argument is reminiscent of earlier epidemiologic myths about ethnic differences, such as the assertion a century ago that tuberculosis was a genetic disease linked to the trait of red hair on the basis of its higher incidence among the Irish. The authors also ignored a substantial literature that contraindicates their proposed explanation. For example, multiple studies have found that the risk of adverse pregnancy outcome for black immigrants to the United States is comparable with that of native-born whites, not to native-born blacks. and 
Imagine how preposterous it would be if, after having ruled out a few murder suspects named Bob, Fred, and Karl, a police detective were to declare triumphantly that the murderer must therefore have been a woman. Science would be better served if we made well-reasoned inferences about the effects of things we actually measured, rather than unfounded speculations about things that we did not measure.