To assess the accuracy of genotype imputation from reference panels of diverse ancestry, we performed leave-one-out cross-validations in data from HapMap Phase 3. The HapMap 3 paper (The International HapMap Consortium 2010) includes a number of similar comparisons, which use a series of carefully controlled experiments to show how population ancestry in the reference and study data affects imputation accuracy. By contrast, our main goal is to validate a general strategy for using ancestrally diverse reference data. One way to view this distinction is that their experiments provide information about the kinds of reference data one might want to collect to improve imputation accuracy in a given population, whereas our experiments illustrate a strategy for getting good results from whichever data are available. To the extent that we consider how reference panels of different ancestries affect imputation accuracy, the main point is to guide intuition about why the overall strategy works.
