We performed these experiments with IMPUTE version 2.1.2 under the following settings: k = 80 (tuning parameter for phasing updates), iter = 30 (total number of MCMC iterations), burnin = 10 (number of iter to discard as burn-in), hap_spec_fam (flag to make the program choose a custom reference panel for each study haplotype), and Ne = 20000. The Ne parameter represents the effective size of the population being analyzed, and it is used to scale the recombination rates in the imputation HMM. It may seem odd that we use a single Ne value in populations that clearly have different effective sizes, but our pilot experiments showed that IMPUTE2 is largely insensitive to this parameter and that 20000 is a good universal value (data not shown). The approximations underlying the k and khap parameters are modeled on local genealogies with limited recombination, so we split chromosome 20 into nonoverlapping 5-Mb chunks for analysis, with a 250-kb buffer region on each side to prevent edge effects (this is a default setting in IMPUTE2).
