The first two points document strange behavior of the BEAGLE method: apparently, adding data – whether in the form of additional SNPs or additional individuals in the study sample – can cause BEAGLE's imputation accuracy to decrease. More specifically, it seems that increasing the proportion of missing data harms BEAGLE's inferences. This suggests an explanation for the third point above: as the reference panel grew and the study sample remained fixed, the total proportion of missing genotypes in the sample decreased, thereby generating datasets that were relatively less harmful to BEAGLE.
