The hypothesized SEM model was developed and evaluated via AMOS (Arbuckle, 2014) software using maximum likelihood estimations. To optimize potential clinical implications of how LR and impulsivity operate, these two predictors are manifest variables rather than latent constructs. Time 2 potential mediators of the effects of LR and impulsivity and Time 3 outcomes were represented as latent variables and evaluated with confirmatory factor analyses. The resulting measurement model for latent variables was incorporated within the final full SEM after dropping components that were not significant. Two potential latent variables did not have subscales (e.g., the Drinking to Cope Scale and the number of DSM-IV AUD criteria endorsed), so 3 “parcels” were created as indicators for each variable by placing the items for that measure into three a priori groups (Little et al., 2002). Within the SEM, direct associations and indirect statistical mediation were evaluated using Mplus, Version 7 (Muthén and Muthén, 2012) based on bootstrapping with 1,000 resamples with results presented as 95% confidence intervals (mediation was present if the intervals did not cross zero). Good fit criteria for the
