We next sought to evaluate whether ‘winner's curse’ [33], [34] was deflating replication rates. Therefore, we extracted simulations in which the estimated coefficients randomly decreased and found that simulated replication remained >90% at BF>5 and near 100% at higher BF even when the effect size declined substantially (e.g. 30% drop in regression coefficient; Figure S9). Effect sizes would need to be over-estimated by 2-fold or greater across the entire set of eQTLs with UC BF>5 to result in the observed rates of replication. Furthermore, two lines of reasoning suggest winner's curse is not a major contributor to the observed rates of non-replication. First, we note that bias resulting from winner's curse should be progressively less pronounced as the true effect size increases, which in turn will correlate with significance estimates in the discovery panel [34]. However, replication rate was essentially flat even at extremely stringent thresholds (Figure 3C). Additionally, the resampling experiments demonstrated that, in direct contrast with a winner's curse prediction, effect sizes would need to be increasingly more severely over-estimated at higher thresholds (3-fold or more) to result
