Structural equation modeling (SEM) was used to construct an integrative model including both genetic and non-genetic risk factors. Probit regression was fitted using the robust weighted least squares estimation. To increase statistical power, all smoking-related variables were dichotomized. To keep for multiple testing at a minimum, only variables with significant effects on smoking at either age 14 or 31 were entered into the model. Maternal smoking and smoking at 14 years were coded as smokers Vs. non-smokers (due to the relative low number of heavy/regular-smokers in these two groups); smoking at 31 years was coded as heavy-smokers vs. light+non-smokers (to allow the detection of CHRNA3-rs1051730 effect on smoking). Effect sizes of the predictors on outcome variables are expressed as un-standardized and standardized beta estimates. The beta coefficients are interpreted in z-score metrics since the probit analysis is based on the cumulative normal probability distribution. The standardized beta coefficients are used to compare the relative importance of the different predictors as they describe the change in standard deviation (SD) units of the outcome variable per each SD change in a continuous
