Table 2.

Summary of estimator performance over 5,000 simulations. The rows denote target parameter, the study design, and the estimator: unadjusted, TMLE with logistic regression, and TMLE with SuperLearner (“TMLE+SL”).

Target & Design	Estimator	Bias	σ	MSE	rMSE	Power	Coverage
			Sample Size n = 30
PATE & not matched	Unadj	2.3E-4	2.2E-2	4.8E-4	1.00	0.27	0.95
	TMLE	6.8E-4	1.9E-2	3.6E-4	0.75	0.36	0.94
	TMLE+SL	2.9E-4	1.6E-2	2.6E-4	0.55	0.48	0.93
SATE & not matched	Unadj	3.1E-4	2.0E-2	4.2E-4	0.88	0.27	0.96
	TMLE	7.5E-4	1.7E-2	3.0E-4	0.63	0.39	0.95
	TMLE+SL	3.7E-4	1.4E-2	2.0E-4	0.42	0.52	0.95
SATE & matched	Unadj	5.4E-5	1.5E-2	2.2E-4	0.46	0.37	0.98
	TMLE	3.7E-4	1.4E-2	2.1E-4	0.43	0.44	0.97
	TMLE+SL	1.3E-4	1.1E-2	1.3E-4	0.27	0.58	0.97
			Sample Size n = 50
PATE & not matched	Unadj	−1.3E-4	1.7E-2	3.0E-4	1.00	0.41	0.94
	TMLE	1.1E-4	1.5E-2	2.2E-4	0.75	0.53	0.94
	TMLE+SL	−3.1E-6	1.2E-2	1.6E-4	0.53	0.68	0.94
SATE & not matched	Unadj	4.8E-5	1.6E-2	2.5E-4	0.86	0.41	0.96
	TMLE	2.9E-4	1.3E-2	1.8E-4	0.60	0.55	0.96
	TMLE+SL	1.8E-4	1.1E-2	1.1E-4	0.38	0.70	0.97
SATE & matched	Unadj	−1.8E-4	1.1E-2	1.1E-4	0.38	0.59	0.98
	TMLE	−1.6E-4	1.0E-2	1.1E-4	0.36	0.66	0.97
	TMLE+SL	−5.7E-5	8.2E-3	6.7E-5	0.23	0.81	0.98