A set of follow-up analyses were conducted to determine whether there were interactions among the predictors. Recognizing the large number of potential interactions that might be tested, and to avoid Type I error with this modest sample size, an a priori decision was made to focus only on the significant covariate effects from the full model. A total of six 2-way interactions were created from the significant covariates of FH of AUDs, offspring gender, conduct problems, and prior THC use. One interaction yielded a significant likelihood ratio test (LRT) when a model with the cross-product and constituent main effects was compared to a model excluding the cross-product: Conduct problems by prior THC use (χ2 = 4.92, df=1, p=.03; b=.39, p=.006). Thus, allowing for this interaction effect improved model fit and there was evidence to reject the linear additivity assumption for these covariates.
