It has been proposed that gene–environment interactions (G×E) should be conceptualized as departures from additivity of risks between genetic and environmental factors, as such departures most likely correspond to biological causal mechanisms involving both genetic and environmental factors (Rothman et al., 2008; Schwartz, 2006). To follow this recommendation, we tested G×E in a generalized linear model from the binomial family with identity link estimating risk differences for binary outcomes (Wacholder, 1986). Effects of first-order predictors and interaction terms were quantified as risk differences (RD) with 95% confidence intervals (95%CI). RD reflects the absolute increase in probability of depression for each unit of the predictor. For the test of 5-HTTLPR, RD is the absolute increase in the probability of depression with each short allele (e.g. an RD of 0.05 would mean that each short allele increases the risk of depression by 5%). For the test of childhood maltreatment, the RD is the absolute increase in the probability of depression with each grade of maltreatment (e.g. an RD of 0.1 in the Dunedin study would mean that the probability of depression increases
