Multiple imputation by chained equations [67] was used to impute missing values 20 times for age (two individuals), education (ten), marital status (seven), cigarettes per day (CPD) (seven), and Fagerström Test for Nicotine Dependence [68] (FTND) score (forty-two). Regression analyses were performed on each imputed data set and the results were combined with adjustment to the variance of regression parameters to reflect the additional variance attributable to the imputations [69]. Regression analyses were run for all SNPs using an additive model (and for rs2072661, with the dominant model [35,46,53,54]), and with adjustment for the other chr15q25.1 SNPs [30], when appropriate. Regression analysis was conducted with data from all 26 arms (except for rs2072661, where we excluded the two arms from the RCT that discovered the SNP association) and included variables for the SNPs, demographics [age (age and age squared), education (presence or absence of college degree), gender, marital status (married or other)], dependence measures [FTND and CPD (coded as in the FTND)], interactions with demographic variables (CPD x age, CPD x gender and FTND x gender), the first ten principal
