We first calibrated each of the combinations of program and reference panel by determining the threshold of posterior probability required to achieve a concordance of 90% between imputed and experimentally determined genotypes for typed markers. We then filtered each imputed SNP based on a χ2 test of Hardy-Weinberg equilibrium at a significance level of 0.05. The purpose of this test was to detect genotype-specific imputation failure, and we were intentionally conservative about calling imputation successful because an imputed genotyping error rate of 10% is much higher than the estimated experimental genotyping error rate of 0.5% [The International HapMap Consortium 2003; The International HapMap Consortium, 2007]. We anticipated some level of Hardy-Weinberg disequilibrium due to admixture in the study sample. However, simulation studies have shown that the power of the Hardy-Weinberg test to detect disequilibrium due to admixture is low unless the difference in allele frequencies is large (>0.4) and the minor admixture proportion is high (>0.2) [Deng et al., 2001]. Similarly, simulation studies have shown that the power of the Hardy-Weinberg test to detect disequilibrium due to genotyping errors at
