al., 1997) were not included in the logistic regression models. While we recognize that omitting these predictors is a limitation of this study, the focus of this paper was determining whether the risks for substance problems in young adulthood were common or specific across these substances. Our approach is mirrored in an expanding literature of biometrical studies (e.g., Goldman and Bergen, 1998; Kendler et al., 2007, 2008; Rhee et al., 2006; True et al., 1999; Young et al., 2006) that provide evidence of a common genetic and/or environmental factor contributing to the comorbidity among these substances in both adolescence and adulthood. Fourth, dependence measures do not impose the clustering criterion of the DSM-IV or distinguish between dependence with or without physiological symptoms. Unlike previous adolescent epidemiological studies, we used a threshold of three symptoms of dependence, to be consistent with the threshold for an adult diagnosis and to allow for comparison to adult studies. We chose not to impose clustering because the prevalence of disorders was only slightly changed across substances when clustering was required. Lastly, several of the confidence intervals in Table 5 are substantial. We believe this is a result of the low prevalence rates of SUDs in
