In the restricted dataset, we used IMPUTE v1.0, IMPUTE v2.0, BEAGLE v3.0.2, fastPHASE v1.3.2, and PLINK [15] v1.03 to impute each chip's masked genotypes from the other chip's study sample genotypes and the reference panels. IMPUTE v2, BEAGLE, fastPHASE and PLINK used the 918 diploid individuals and the 120 HapMap CEU haplotypes as an expanded reference panel for imputation, while IMPUTE v1 was provided with only the HapMap reference panel. We ran IMPUTE v1, BEAGLE, and PLINK on their default imputation settings, and we also performed separate BEAGLE runs with 50 iterations (rather than the default 10). We ran fastPHASE with 20 and 30 clusters (K = 20 and K = 30, in separate runs), 15 starts of the EM algorithm to estimate model parameters, and 35 iterations per EM start. As in Scenario A, we first fit the fastPHASE clustering model to the reference data, then instructed the software to impute each of the 459 study individuals independently, conditional on the fitted model. Finally, we set IMPUTE v2 to use 40 and 80 conditioning states (k = 40 and
