In the most extreme version of this scenario, the 103 observed replications in East Asians at P<0.05 that find the same risk allele that had been previously discovered in European studies would be the result of type I error with a P = 0.05 threshold. In that case, the 103 positive replications would be just the 2.5% ( = 5% type I error×50% probability of the same risk allele) of a large pool of 4,120 replication attempts in East Asians (95% C.I. = 3,418–4,959, assuming a Poisson distribution). In other words, 4,017 ( = 4,120−103) associations failing to find the same risk allele at P<0.05 would have remained unreported. Given the huge amount of unpublished GWAS that this scenario would imply, we discard a big impact of publication bias in our analysis.
