Using linear regression, we model depressive symptoms as a function of 5-HTTLPR and stressors controlling for age, sex and respondent’s education. We include a random effect for township to adjust for the clustered sample design. The few persons with L/XL genotype (n=24) are combined with L/L genotype following previous practice [26, 27]. Models 1 and 2 test the GxE interaction for trauma and major life events, respectively. Model 3 includes both GxE interactions. Subsequently, we re-estimate these models including two- and three-way interactions among stressful exposure, 5-HTTLPR, and sex. We calculate predicted CES-D scores by genotype and stressful experience using the coefficients from Model 3 and mean values for age, percent female and education. All analyses were performed using Stata, Version 10.1 [39].
