Skip to main content
. 2020 May 16;189(12):1559–1567. doi: 10.1093/aje/kwaa083

Figure 2.

Figure 2

Statistical power and coverage of estimators of the natural direct effect (NDE) and the natural indirect effect (NIE), by sample size, estimation method, and effect size, for data-generating mechanisms (DGMs) reflecting no Z and no A-M interaction. A) NDE for DGM 1 (small effect size); B) NIE for DGM 1 (small effect size); C) NDE for DGM 2 (small effect size); D) NIE for DGM 2 (small effect size); E) NDE for DGM 4 (large effect size); F) NIE for DGM 4 (large effect size). “Baron and Kenny” corresponds to the original version of the Baron and Kenny estimator (6). “Ixn exn to B and K” corresponds to the extension to the Baron and Kenny estimator that allows for A-M interaction (8). TMLE1 corresponds to variance estimated using the efficient influence curve, and TMLE2 represents variance calculated using the bootstrap. Ixn exn, interaction extension; TMLE, targeted maximum likelihood estimation.