We have only considered common SNPs represented on the GeneChip 500k array, and moreover only those for which the χ2–distribution is accurate. Mathematically this is not a serious problem: provided uniformly distributed P–values can be obtained, our approach applies to any variants. However, it is possible that rare variants follow a different linkage disequilibrium (LD) structure, because they are likely to be younger or under different selection pressures than common variants. Our results strictly apply only to common variation, and the implicit assumption that common variation occurs at arbitrarily fine scales may be unreasonably conservative. The similarity of our estimate to that of the International HapMap Consortium [2005] suggests that this assumption has not had a major impact on our results.
