Figure 3A shows the full results of this comparison. The curves are difficult to distinguish in this plot, so Figure 3B and 3C magnifies either end of the range to highlight the salient features. The grid lines in all three panels are shown at the same vertices to help convey the degree of magnification. The results can also be summarized by the best-guess error rate for each method (x-intercept on the plots): BEAGLE (default), 6.33%; BEAGLE (50 iterations), 6.24%; fastPHASE (K = 20), 6.07%; fastPHASE (K = 30), 5.92%; IMPUTE v1, 5.42%; IMPUTE v2 (k = 40), 5.23%; IMPUTE v2 (k = 80), 5.16%; MACH, 5.46%. Figure 3 shows that IMPUTE v1 (blue) achieved error rates that were consistently, if only slightly, lower than those of MACH (cyan) across the range of calling thresholds, and that both methods yielded lower error rates than fastPHASE (black) and BEAGLE (green). The IMPUTE v2 run with k = 40 (solid red line) attained similar accuracy to IMPUTE v1 at stringent calling thresholds (Figure 3B), although IMPUTE v2 gained a slight advantage at more
