Table 1.

Comparing the type I error rate for the four different methods

Data generated under hierarchical model
	Nominal P value
Test method	0.001	0.01	0.05	0.1
Bartlett	0.001032	0.01014	0.0504	0.1006
F test	0.001011	0.01003	0.05001	0.1001
DiffVar (abs)	0.0010029	0.01023	0.05094	0.1036
DiffVar (sq)	0.0007118	0.009800	0.0499	0.1018
Resampled data from kidney non-diseased samples
	Nominal P value
Test method	0.001	0.01	0.05	0.1
Bartlett	0.009	0.0257	0.07715	0.1331
F test	0.0089	0.0253	0.0763	0.1319
DiffVar (abs)	0.001	0.0104	0.0523	0.105
DiffVar (sq)	0.0004	0.00665	0.0448	0.0981

Data were generated in two ways: under a hierarchical model and by randomly selecting the non-diseased kidney samples. Median type I error rates are reported for 1000 simulations with no differentially variable or differentially methylated features. For simulations generated under the hierarchical model, the standard deviation with which the error rate is estimated ranges from approximately 0.00024 for rates near 0.001 to 0.0029 for rates near 0.1 with no notable difference between the methods. For the resampled data, the standard deviation ranges from approximately 0.0026 for rates near 0.001 to 0.026 for rates near 0.1 with DiffVar (sq) being noticeably less variable than the other methods. abs, absolute deviations; sq, squared deviations.