beta
/In contrast to p, all methods show a similar and clear trend of decreasing accuracy as T increases (Figure 3D, S3). More specifically, when moving from T = 5 to T = 7, R decreases by 0.003 for HAPMIX, 0.006 for LAMP-LD, and by 0.004 for each of FLARE and RFMix. This is likely due to the increased number of ancestral recombinations for larger values of T that produce shorter local ancestry tracts. As shorter tracts have fewer SNPs than longer ones, they have lower informativeness, making LAI more challenging. Furthermore, calling local ancestry at tract boundaries is difficult, and a larger number of such boundaries may also contribute to the reduction in accuracy. Here, as with the admixture proportion, HAPMIX proves to be the most robust LAI method with respect to changes in demography.