We have relied on permutation testing to estimate significance thresholds, as this preserves the correlation structure in the sample. Similar results from the NBS and 58BC samples suggest that sampling variation in the correlation structure is negligible. Permutation tests are time consuming, and a convenient alternative is to estimate an effective number of tests from the genotype correlation matrix. Our results from fitting Beta distributions indicate that such an effective number can be found, at least at the current density, but we found that Patterson's moment estimator was an order of magnitude too low for correcting the minimum P–value. It is an open problem to directly estimate the appropriate correction for minimum P from genotype data.
