Nevertheless, at the time of writing the WTCCC reported successful replication of 10 associations from 11 attempts. Assuming that the failed replication was a false positive, this does not constitute success by classical measures: a family–wise type–1 error occurred, and the false discovery proportion was nearly 10%. Only the Bayesian interpretation that posterior odds for association are 10:1 seems satisfactory. Therefore, the fact that the study is considered a success throws light on how multiplicity is interpreted in practice. Family–wise error is too conservative for rejecting individual hypotheses, and the FDR can be higher than traditional type–1 error rates. Posterior odds of 10:1 seem acceptable, which justifies the use of P<0.05 for hypothesis testing provided that the prior odds are close to evens. Such a prior is reasonable for genomewide association scans as long as the whole genome is considered rather than limited subsets of markers.
