Next, adjustments were made for potentially confounding or mediating variables. The design does not permit safe conclusions regarding the status of some of the predictors as confounders or mediators. Nonetheless, we a priori tentatively classified the covariates as confounding or mediating factors, based on their assumed temporal relation to alcohol abuse. Divorce and parental mental distress were considered likely confounders. Adolescent mental distress, seeing parents drunk, and adolescent social network were considered possible mediators, as they are likely to occur after the onset of parental alcohol abuse. To see what changes each variable caused to the model, adjustments were made stepwise, adding one variable at a time to the conjoint demography adjusted analysis.
