To evaluate Hypothesis 1, that positive reinforcement is more strongly associated with alcohol consumption than is negative reinforcement among individuals without AD, we fit a generalized estimating equation (GEE) model with logarithm link and Poisson distribution with frequency of alcohol consumption as the outcome variable and positive and negative reinforcement as predictors (N =1824, 53.1% female). Individuals diagnosed with AD at the time of assessment were excluded from this set of analyses. To test Hypothesis 2, that negative reinforcement is more strongly associated with AD than positive reinforcement, we fit a GEE model with logit link and binomial distribution that predicted AD by positive and negative reinforcement. All participants were included in this analysis (N = 2556, 51.6% female). For both models, linear and quadratic effects of age were included as covariates to take into account the non-linear trajectory of alcohol use between age 18 and 30, as previously described (Dick et al., 2014; Li et al., 2017). Age was centered at 21, where alcohol consumption started to stabilize (Dick et al., 2014; Li et al., 2017). In addition to
