depends on the size of the main effects and the proportion of individuals experiencing SLE. Based on our conservative power calculations, presented in Table II Supplementary, our largest sample (SelfRepDep) has substantial power to detect a G × E. However, our sample constructed to most closely match Caspi et al. [2003] has quite low power. These power calculations are the minimum we expect since we necessarily used binary variables for depression and SLE to compute power, and the power will be greater for the ordinal depression and continuous SLE scales used throughout this article. Second, despite our improved genotyping and subclassing by rs25531, we could not distinguish those with genotypes LgLg from those with genotype LgLa. However since we estimate that LgLg comprise only 0.9% of the total sample (or 12% of those genotyped as LgLa) we do not believe that being able to disentangle these genotypes would change our results. Third, self-reported depression, whilst assessed at the same time as SLE, assesses “recent” experiences so does not represent those of the 12-month duration of SLE reporting.
