with future high levels of functioning despite their elevated risk for AUDs. However, those steps and the relative homogeneity of the sample regarding sex, education, and race potentially limited the generalizability of the findings. An additional caveat is that, while the population was relatively large for a long-term intensive follow up, the number of individuals developing SUDs was relatively small with resulting limited statistical power in some analyses. This contributed to the decision to combine abuse and dependence into a single category, and to consider amphetamines and cocaine as stimulants. It is also important to emphasize that while the results regarding externalizing personality test scores were consistent with the predicted direction of such values, only three questionnaires were used and those were measured at least 10 years after subjects entered the protocol. In addition, earlier life internalizing phenomena were evaluated by only baseline histories of depressions or mental health treatment. Finally, the approach to regressions used here may have optimized statistical power by minimizing standard errors and producing relatively narrow confidence intervals, although an alternate method, hierarchical logistic regression, produced similar results but has the limitation of not requiring that every item in the block add significantly to the prediction.
