Analyses were based on unordered multinomial logistic regression models (Hosmer & Lemeshow, 2000) that compared the attempt and ideation groups to non-suicidal subjects. Odds ratios and asymptotic confidence intervals were computed using the method of profile likelihood (McCullagh & Nelder, 1989). Statistical significance is based on alpha = .05. Goodness-of-fit tests using the Hosmer-Lemeshow statistic were derived using independent logistic regression analyses comparing each attempter and ideation group respectively to non-suicidal subjects. Predictors were RA and PA along with covariates that included AUDIT score, primary substance of use (alcohol, reference), breadth of drug use, depressive symptoms, sex (female, reference), age, race (white, reference), and education (≥ 12 years, reference). Correlations among predictors were modest (r ≤.41).
