Chi-square tests assessed bivariate associations with recovery. Multiple logistic regression models estimated the odds of AR (excluding individuals with NR) and the odds of NR (excluding those with AR). Because of the large number of candidate correlates, those with bivariate p-values >.500 were excluded from the initial models. Each model was manually reduced to retain only those correlates with p-values <.10 or whose removal would change the referent category for nominal variables. Thus, if any category of a multicategorical variable had a p-value <.10, all categories were retained in the reduced models. Covariates with marginal levels of significance (.05 - <.10) were retained for comparisons with prior analyses but are not considered as definitive indicators of associations in their own right. Past-year average daily ethanol intake (log ounces) and number of AUD symptoms were categorized for presenting bivariate associations but treated as continuous in the multivariate models. Statistical analyses employed SUDAAN (Research Triangle Institute, 2008), a software package that uses Taylor-series linearization to yield variance estimates that account for complex, multi-stage sample designs.
