Figure 2.
Power for rejecting a test of H0 : Ψ(P0) = 0 versus H1 : Ψ(P0) > 0 at the 0.025 level at different sample sizes. Horizontal axis on a log scale. All methods have power increasing to 1. The stabilized one-step has lower power than the non-oracle CV-TMLE, though its power is typically more comparable to the stabilized one-step relative than the oracle CV-TMLE which knew the optimal subgroup from the outset.