As described previously (Bierut et al. 2010), covariates included sex, age, sample origin (COGA, COGEND and FSCD), cocaine dependence and nicotine dependence. The relationship between age at interview and maxdrinks was not linear in the SAGE dataset. Therefore we avoided using age as a continuous covariate in the analysis. We coded age at interview into four dummy categorical variables (defined, using quartiles, as 34 years and younger, 35–39 years, 40–44 years, and 45 years and older) and included them as covariates. Nicotine and cocaine dependence were included as covariates in SAGE because the COGEND and FSCD datasets were ascertained on nicotine dependence and cocaine dependence respectively. Association of natural log transformed maxdrinks was examined by linear regression using PLINK (Purcell et al. 2007).
