Because the differences between the point estimates for problem use in the AA and EA models were so large, we conducted an additional analysis aimed at determining whether they were likely to be reflective of a true difference that was not detected due to low power resulting from a relatively small number of African Americans in the full sample. In a series of univariate models using the full sample, the estimate for C in the EA subsample was multiplied by a constant ranging from .5–50 in increments of .5 (totaling 100 models), thus setting the relative value of C in the EA subsample to between .5 and 50 times that in the AA subsample. (A more detailed description of this approach by Duncan et al. (unpublished) is available upon request.) The best-fitting model was reached when the constant was set to 18, and it remained at that value for all subsequent models. Results indicate that the most likely model of problem use is one in which the shared environmental component is 18 or more times greater in European Americans than African Americans.
