Replication may also be useful in identifying a more reliable estimate of the effect size for the association. Signals selected based on statistical significance thresholds in underpowered settings are likely to have (on average) inflated effects due to the winner’s curse phenomenon. [26–30] Replication should take this into account during the sample size calculations for the replication efforts; the effect estimate from the initial study may be inflated, leading to an under-estimate of the number of subjects needed to reliably detect it. [26–29] Analytic methods are available to adjust for winner’s-curse bias, but studying the marker in additional samples (beyond those used to initially identify the marker) will help produce more unbiased estimates of the genetic effect. Accurate estimates of marker risks are important (even if the marker is only a surrogate for the as-yet unknown causal variant), as they may be used for personalized predictive purposes. [31, 32]
