PLINK (Purcell et al., 2007) was used to conduct all autosomal association analyses using logistic regression models. Genotype was coded log-additively (AA, Aa/aA, aa) and covariates representing sex, age at interview, whether participants were drawn from COGA or FSCD (COGEND serving as the reference group) and the first two principal components to account for ethnic variation in allele frequencies, were also included. As our primary analysis, we examined SNPs in SNCA, which has been previously studied in relation to alcohol craving, as well as genes in the dopamine pathway (DRD1, DRD2, DRD3, DRD4, SLC6A3). For the analysis of these genes of a priori significance, we set a liberal threshold of p < 0.05. Next, we conducted a genomewide search – as these analyses are exploratory, statistical significance in the GWAS was set at p < 5 × 10−8. For both association analyses, primary analyses were conducted to identify SNPs associated with alcohol craving. For SNPs that were implicated in the primary analysis of craving, secondary analyses were performed examining other comparisons. These included an analysis of only non-craving individuals, contrasting
