Table 5.

Simulation study: average power and type I error rate of the 95% credible interval of the treatment effect, based on 500 simulated data sets.

Scenario	1	2	3	4	5	6
Time trend	No	Yes	Yes	No	No	No
Heterogeneity	Small	Small	Small	None	Moderate	Large
Power
Current data	0.690	0.692	0.690	0.662	0.668	0.670
Pooled data	0.842	0.946	0.728	0.876	0.734	0.650
Pocock’s method	0.842	0.930	0.708	0.854	0.750	0.660
Power prior with α = 0.5	0.836	0.926	0.706	0.846	0.742	0.646
MPP	0.820	0.906	0.698	0.840	0.734	0.668
MAP approach	0.756	0.794	0.710	0.786	0.714	0.682
Robust MAP approach	0.744	0.772	0.686	0.738	0.700	0.682
Method of Murray et al.	0.826	0.924	0.722	0.858	0.728	0.672
Test-then-pool method	0.716	0.726	0.682	0.796	0.628	0.660
Type I error rate
Current data	0.042	0.040	0.040	0.046	0.040	0.048
Pooled data	0.096	0.162	0.064	0.052	0.254	0.542
Pocock’s method	0.068	0.118	0.038	0.040	0.152	0.428
Power prior with α = 0.5	0.068	0.118	0.046	0.036	0.156	0.428
MPP	0.052	0.084	0.040	0.038	0.076	0.102
MAP approach	0.030	0.046	0.040	0.030	0.040	0.048
Robust MAP approach	0.030	0.040	0.040	0.034	0.036	0.048
Method of Murray et al.	0.086	0.138	0.062	0.044	0.144	0.172
Test-then-pool method	0.046	0.062	0.042	0.040	0.042	0.050

Note: The width of each side of the 95% binomial proportion confidence interval (not shown in the table) is approximately 2% to 3% for the type I error rate and 4% for the power.