Regression procedures were used to investigate the association of the each of the four allelic risk factors with each alcohol phenotype. Analyses were adjusted for sex, age, and FSU status, as drinking behavior differs by these subgroups in Israel (Hasin et al., 2002a; Hasin et al., 2002b; Spivak et al., 2007; Shmulewitz et al., 2012). Logistic regression was used for binary phenotypes and Poisson regression (with overdispersion) was used for the count phenotypes, as they showed skewed distributions. The count phenotypes were also modeled using normal, Poisson (without overdispersion), negative binomial, and zero-inflated distributions; data best fit the overdispersed Poisson distribution based on the largest likelihood and smallest goodness-of-it indices (Akaike's Information Criterion and the Bayesian Information Criterion). For binary phenotypes, results are reported as odds ratios (ORs), indicating the increase in the odds of the phenotype given the absence of the protective allele. For count phenotypes, results are reported as risk ratios (RRs), indicating the relative increase in the mean phenotype value given the absence of the protective allele (Hasin et al., 2002a). To exclude association due to population
