Our framework was motivated by the idea that larger reference datasets should make imputation faster and simpler, rather than slower and more complicated. We believe that our work represents a first step toward bridging current imputation practice with the paradigm suggested by Kong et al. (2008), in which large population samples eliminate the need for complex models and reference panel selection, and investigators do not have to balance efficiency and accuracy. Detailed population models and reference panel weighting schemes may provide modest accuracy improvements in the short term, but we expect that the power gains from such developments will seldom justify the added computational costs.
