The imputation yield, defined as the nominal number of SNPs imputed successfully, may represent an overestimation of gain because imputation requires high levels of linkage disequilibrium to perform well. To measure the effective gain through imputation accounting for linkage disequilibrium, we estimated the increase in coverage of HapMap variation on chromosome 22 from using the set of tag SNPs on the chip to using the combined set of tag SNPs on the chip and successfully imputed SNPs (Table III). The set of tag SNPs on the chip achieved chromosome coverage of 75% for the phase II CEU reference panel but only 55% for the phase II YRI reference panel. Imputation using BEAGLE, despite yielding several hundred to a few thousand SNPs, increased coverage by only 1–2% regardless of the phase II reference panel. Similarly, imputation using MACH increased coverage of phase II CEU variation by only 1% using the phase II CEU reference panel. In stark contrast, imputation using MACH increased coverage of phase II YRI variation by 21% (from 55% to 76%) using only the phase II YRI reference
