Some other caveats should be mentioned. The winner’s curse in the magnitude of the effect in the discovery phase may introduce spuriously inflated heterogeneity, when the discovery data are combined with subsequent replication studies. In such two-stage approaches, between-study heterogeneity should best be estimated excluding the discovery data. Conversely, if all datasets are measured with genome-wide platforms and GWA scan meta-analysis is performed in all gene variants, this is no longer an issue. In fact, if the GWA scan meta-analysis uses random effects (see below), the emerging top hits from the GWA scan meta-analysis are likely to have, on average, deflated observed heterogeneity compared with the true heterogeneity. This is because, underestimation of the between-study heterogeneity favors a variant to come to the top of the list, since it does not get penalized by wider confidence intervals in the random effects setting.
