we were able to utilize propensity score methodology to assess the causal inferences of these associations. The use of this methodology reduces confounding due to observed characteristics and any unobserved characteristics associated with the observed ones. However, non-causal explanations for the reported associations are still possible. Fourth, although this is one of the largest prospective population-based samples available with the degree of mental health and substance use measurement necessary for this complex assessment, some analyses are still limited by small subgroup sizes, such as those by specific type of treatment history, and diagnostic classification. As a consequence, power to examine these additional subgroups was inadequate.
