Syndicate content

Add new comment

Submitted by Anonymous on

At the risk of revealing my ignorance, could you expand on why you would need the adjustment in the case of pairwise randomization? My understanding from Prop 1 of their paper is that the problem arises when there's heterogeneous variance in treatment after conditioning on the controls. In the case of pairwise randomization (and other stratified randomization schemes with constant treatment propensities) it seems like treatment variance should be constant by construction.