of drinking days follows a binomial distribution (e.g., non-drinking day=0, drinking day=1). The same assumption and modeling was applied to number of heavy drinking days to determine the relative risk. The number of drinks per drinking day is counting data, and was modeled using a GLRM with Poisson distribution using log link function (Nelder and Wedderburn, 1972, Kiernan et al., 2011) to determine the odds ratio. The independent variables included in the models were OPRM1 genotype, DAT1 genotype, and their interaction term, sex, AIM components (C1, C2) and their interaction term; sex and AIM components were included in the model to control for known sex and ancestral origin differences (covariates). The adjusted least square mean odds ratio (number of drinking days and number of heavy drinking days) or relative risk (number of drinks per drinking day) were calculated between genotypes and the interaction terms.
