In contrast, we observed little risk from either parental separation or parental alcoholism on timing of substance involvement for African American offspring. In adjusted Cox models, somewhat surprising protective effects were found. However, results from propensity score analyses suggest very cautious interpretation of survival-analytic findings, regardless of covariate control. Intact and separated African American families were poorly matched on risks upstream of parental separation, including parental alcoholism, and no African American families at highest risk of separation remained intact; thus, we have little confidence in comparisons of African American offspring outcomes examined as a function of parental separation. Such findings also call into question use of statistical adjustment for race/ethnicity in research comparing intact and separated families. Because inferences can be made only about a very restricted range of the propensity score distribution for African Americans, if data from European and African ancestry families were pooled and race/ethnicity adjusted statistically, we would in effect be making predictions for African Americans in regions with zero data-points, i.e., intact African American families where parents are at high probability of separation.
