Table 1.

Empirical type-I error and power obtained in 1,000 simulations using the three different approaches: REX (iterative regression), PEMM (probe-specific mixture model) and threshold.

				Type-I error		Power (gains)			Power (loses)
σβ	σγ	σε	${\overline{x}}_{alt}$	REX	PEMM	REX	PEMM	thres	REX	PEMM	thres
0.2	0.2	0.05	3.7	0.073	0.031	0.815	0.745	0.000	0.819	0.751	0.000
0.2	0.2	0.08	3.6	0.071	0.027	0.689	0.546	0.000	0.681	0.541	0.000
0.2	0.4	0.05	3.5	0.081	0.044	0.923	0.891	0.024	0.930	0.908	0.029
0.2	0.4	0.08	3.5	0.079	0.038	0.870	0.811	0.028	0.863	0.814	0.057
0.2	0.6	0.05	3.6	0.081	0.052	0.965	0.955	0.131	0.956	0.956	0.161
0.2	0.6	0.08	3.6	0.086	0.044	0.922	0.889	0.140	0.922	0.900	0.184
0.2	1.0	0.05	3.6	0.081	0.057	0.980	0.983	0.360	0.977	0.972	0.411
0.2	1.0	0.08	3.6	0.085	0.052	0.955	0.954	0.396	0.947	0.939	0.449
0.2	1.5	0.05	3.6	0.081	0.060	0.982	0.978	0.536	0.981	0.983	0.582
0.2	1.5	0.08	3.6	0.084	0.056	0.981	0.986	0.623	0.980	0.975	0.630
0.2	4.0	0.05	3.5	0.078	0.064	0.995	0.993	0.815	0.993	0.994	0.840
0.2	4.0	0.08	3.5	0.079	0.060	0.996	0.993	0.848	0.993	0.995	0.863
0.4	0.2	0.05	3.7	0.079	0.032	0.822	0.729	0.000	0.827	0.737	0.001
0.4	0.2	0.08	3.7	0.070	0.027	0.690	0.522	0.000	0.697	0.537	0.000
0.4	0.4	0.05	3.7	0.081	0.046	0.918	0.897	0.027	0.926	0.905	0.044
0.4	0.4	0.08	3.7	0.078	0.038	0.878	0.807	0.035	0.871	0.825	0.069
0.4	0.6	0.05	3.6	0.080	0.052	0.958	0.949	0.139	0.952	0.942	0.191
0.4	0.6	0.08	3.5	0.083	0.044	0.926	0.903	0.150	0.913	0.881	0.206
0.4	1.0	0.05	3.7	0.080	0.057	0.981	0.982	0.353	0.973	0.977	0.425
0.4	1.0	0.08	3.5	0.081	0.050	0.949	0.953	0.392	0.950	0.934	0.424
0.4	1.5	0.05	3.6	0.083	0.063	0.989	0.985	0.553	0.987	0.989	0.613
0.4	1.5	0.08	3.6	0.080	0.056	0.976	0.973	0.568	0.966	0.967	0.603
0.4	4.0	0.05	3.6	0.085	0.066	0.997	0.996	0.856	0.995	0.995	0.855
0.4	4.0	0.08	3.6	0.086	0.062	0.994	0.991	0.841	0.989	0.990	0.864
0.6	0.2	0.05	3.6	0.075	0.031	0.828	0.740	0.000	0.824	0.741	0.001
0.6	0.2	0.08	3.6	0.071	0.027	0.701	0.529	0.000	0.690	0.520	0.002
0.6	0.4	0.05	3.7	0.082	0.047	0.932	0.916	0.032	0.914	0.902	0.063
0.6	0.4	0.08	3.6	0.077	0.037	0.864	0.815	0.041	0.874	0.835	0.070
0.6	0.6	0.05	3.7	0.082	0.051	0.964	0.949	0.142	0.955	0.944	0.174
0.6	0.6	0.08	3.6	0.078	0.044	0.924	0.885	0.177	0.918	0.887	0.206
0.6	1.0	0.05	3.6	0.081	0.058	0.977	0.970	0.388	0.972	0.974	0.435
0.6	1.0	0.08	3.6	0.080	0.050	0.946	0.935	0.380	0.960	0.945	0.451
0.6	1.5	0.05	3.6	0.087	0.062	0.982	0.981	0.556	0.982	0.985	0.590
0.6	1.5	0.08	3.6	0.080	0.056	0.972	0.972	0.597	0.975	0.973	0.622
0.6	4.0	0.05	3.5	0.080	0.064	0.995	0.995	0.837	0.994	0.993	0.855
0.6	4.0	0.08	3.6	0.082	0.064	0.988	0.991	0.850	0.990	0.990	0.857

These results are for the case of having 20% of probes as a internal control probes (needed for the REX approach) and 10% of probes as a true altered probes. The results are given for different scenarios between probe variability (σ_β), probe-test variability (σ_γ) and within-probe variability (σ_ε). The column ${\overline{x}}_{alt}$ indicates the mean number of simulated altered probes.