Analyses were conducted using linear (CPD) and logistic (smoking and cannabis initiation, current use of cigarettes and cannabis, cannabis abuse/dependence) regression models in PLINK (Purcell et al., 2007), with genotype coded log-additively (0, 1, 2 copies of the minor allele). Covariates included sex, age at interview, source of study data and 2 principal components that accounted for any residual admixture in those of European-American ancestry. Only SNPs associated with a specific tobacco smoking phenotype in the original meta-analyses were tested against that phenotype (e.g., rs6265 was only tested for smoking initiation). Likewise, only SNPs associated with any tobacco smoking measure were further tested with cannabis involvement measures.
