. Author manuscript; available in PMC: 2015 Feb 20.

Published in final edited form as: Stat Med. 2013 Sep 9;33(4):555–568. doi: 10.1002/sim.5969

Table III.

Simulation results for a binary endpoint based on 2000 Monte Carlo data sets of size 1000. For IPW, a logistic regression for the propensity score includes all 20 X variables is fitted to the data. For ANCOVA, all 20 X variables are included in the model. For OPT, all 20 X variables are included in separate logistic outcome regression models for the two arms. SE_M is the standard error based on Monte Carlo samples; SE_A is the average of large sample standard error estimates over the Monte Carlo samples. Cov.Prob. is the coverage probability of the 95% confidence interval based on large-sample normal approximation. (Δ₀,Δ₁) are the true proportions in the control and intervention arms, respectively.

(Δ₀,Δ₁)			SE_M/SE_A (×10000)					Cov. Prob.
		UNADJ	IPW	ANCOVA	OPT	UNADJ	IPW	ANCOVA	OPT
(0.1, 0.15)	k=5	209/209	180/174	176/174	157/152	0.950	0.941	0.948	0.943
	10	208/209	174/174	170/174	152/150	0.948	0.952	0.955	0.946
	20	208/204	177/174	173/174	155/149	0.950	0.943	0.946	0.941
(0.3, 0.39)	k=5	288/300	214/220	214/220	200/205	0.965	0.959	0.966	0.962
	10	301/300	227/220	224/220	210/205	0.942	0.944	0.946	0.947
	20	296/300	223/220	220/220	209/203	0.953	0.945	0.949	0.939