We then conducted several preliminary analyses designed to identify potential mediators for further testing. To begin, we examined overall racial/ethnic differences in candidate mediators (using t-tests and chi squares) as well as associations between candidate mediators and alcohol outcomes, controlling for heavy drinking (using negative binomial regressions). Candidate mediators included the disadvantage-related factors and cultural/demographic factors identified in Fig. 1. Next, we used a series of multivariate regressions to individually explore roles for each potential mediator in explaining disparities. Specifically, negative binominal models were fitted to predict the count of dependence criteria and, separately, consequences. Raw models estimated effects for race/ethnicity (Black and Latino vs. White), heavy drinking score, and their interaction. Subsequently, each potential mediator was (separately) added to the model, along with the interaction between that mediator and heavy drinking: we thus accounted not only for the mediator’s main effect, but also for the extent to which it modified relationships between heavy drinking and alcohol outcomes. We then tested for significant differences in the coefficient representing the race/ethnicity effect on the intercept in models with and without the
