The multiple-group SEM had close to good fit to the data, χ2(715) = 1188, CFI = 0.966, TLI = 0.957, RMSEA = 0.019 (90% CI = 0.017–0.021). Table 2 contains covariate-adjusted path coefficients for the SEM model. As noted by MacKinnon (2008), the relation between the independent variable and mediator, as well as the relation between the mediator and outcome, together are the necessary relationships (and sufficient if their product is statistically significant) in mediation analysis. There was a statistically significant and negative association between PAS and the mediator ISS, which was the first necessary condition to establish mediation through ISS, in both labeled and labeled individuals. In addition, ISS was significantly associated with both past-year AUD and past-year internalizing disorder, which was the second necessary condition to establish mediation through ISS, for both labeled and unlabeled individuals. In contrast, the associations between PAS and SNI were not statistically significant, which precluded mediation through SNI in both labeled and unlabeled individuals. The association between SNI and past-year AUD was statistically significant in unlabeled individuals only, and SNI was not associated with past-year internalizing disorder.
