Analyses proceeded in four steps. Preliminary analyses (steps 1 to 3) were conducted in SAS (SAS Institute, 2008), and final models (step 4) were conducted in Stata (StataCorp, 2007). First, ordinal logistic regression analyses testing the association of rs2066702 and rs1229984 with maxdrinks and AUD symptoms were conducted separately in the AA and EA subsamples to estimate main effects and determine the specificity of the effects by population (i.e., rs2066702 in African-Americans and rs1229984 in European-Americans). Second, to rule out possible confounding effects of gene–environment correlations, we conducted logistic regression analyses with each SNP to determine whether exposure to childhood adversity varies by genotype. Third, we examined sex differences in maxdrinks, AUD symptoms, and childhood adversity in the full sample to determine the appropriateness of analyzing data from female and male subjects together. Independent t-tests were conducted to assess sex differences in mean maxdrinks and AUD criteria, and a logistic regression analysis was conducted to assess distinctions by sex in risk for exposure to childhood adversity. Fourth, based on findings from preliminary analyses (reported in the Results section), ordinal regression
