Table 2.

Simulation results comparing combining subclasses by weighting by the inverse variance to weighting by the proportion in subclass in the base simulation without positivity violations and with a constant treatment effect (DGM 1). Results given by number of subclasses for the non-survey and survey scenarios for 1,000 simulations. The results assume correct specification of the propensity score model. N = 2, 000 for 5 and 10 subclasses. N = 5, 000 for 30 subclasses to ensure adequate numbers of treated and control units in each subclass.

		Proportion in Subclass Weighting (PSW)				Inverse Variance Weighting (IVW)				Comparison
N subclass	Survey scenario	| Bias |	%Bias	95%CI Cov	MSE	| Bias |	%Bias	95%CI Cov	MSE	MSE % diff	Rec
5	No	0.069	3.19	76.90	0.007	0.062	2.69	82.50	0.005	−17.77	IVW
10	No	0.044	1.35	91.70	0.003	0.042	1.07	93.30	0.003	−9.16	IVW
30	No	0.024	0.37	93.70	0.001	0.024	0.28	94.00	0.001	−1.31	IVW
5	Yes	0.097	3.89	86.70	0.014	0.085	3.14	88.20	0.011	−22.98	IVW
10	Yes	0.073	1.55	93.80	0.008	0.064	1.24	94.40	0.006	−21.67	IVW
30	Yes	0.042	0.33	94.60	0.003	0.038	0.25	94.90	0.002	−20.70	IVW

(Note: MSE % diff = (MSE_IVW − MSE_PSW)/MSE_PSW; Rec=Recommendation.)