The protocol excludes many important studies, in part because of their design features (e.g., case-only designs; [13]) or because they reported symptom dimensions rather than categorical diagnoses of depression (e.g., [14]). However, here we focus on sample size as this has been pivotal in the debate. Discovery science in genetics requires large samples, but hypothesis-testing science does not necessarily. The Culverhouse et al. replication project is not discovery science, it is hypothesis-testing science. In hypothesis-testing science, the consideration of sample size is secondary to more primary considerations of quality of the measures and correctness of design. This order of priorities may be particularly true of hypothesis testing using a meta-analysis approach, as the approach itself provides more than ample sample size. Many of the best-designed studies for testing the GxE hypothesis in question have samples under 300; these smaller studies are significantly more likely to be prospective-longitudinal and to utilize face-to-face interviews [15]. These smaller studies are also more likely to be able to establish temporal order between cause (stress) and effect (depression). In particular, studies of medical illness stressors
