We have shown that previous proposals for genomewide significance have been in the right order of magnitude. It seems clear that, in a Western population, any P–value less than say 5 × 10−8 can be regarded as convincingly significant. We rely on permutation testing to estimate significance thresholds, but these should be adjusted to reflect the genomewide multiplicity. Estimation of an effective number of tests remains an open problem but one which has potential to considerably reduce the computational burden. The next generation of genotyping chips should allow more accurate estimation of significance thresholds with application to a wider range of genomic variation.
