Table 2.

Comparison of the power and type I errors for a simulated microarray dataset

	Normalisation method
	Scale		BABAR		Quantile		print-tip loess
No. of genes	Power	Error	Power	Error	Power	Error	Power	Error
118	0.66	0.022	0.62	0.030	0.51	0.032	0.48	0.033
220	0.73	0.034	0.58	0.053	0.57	0.051	0.51	0.056
407	0.77	0.055	0.62	0.088	0.61	0.089	0.56	0.091
756	0.82	0.088	0.67	0.150	0.65	0.158	0.63	0.163
1402	0.87	0.144	0.76	0.260	0.74	0.300	0.71	0.314

The power and proportion of type I errors for a dataset consisting of 6 simulated microarrays [23] were calculated via the samr R package [25]. Analysis followed one of four methods of microarray data normalisation for different numbers of differentially expressed genes. For BABAR, analyses were run after background correction (subtraction) with 'within arrays' block-by-block centering of the medians, 'between arrays' cyclic loess with averaging of the log₂-ratios, and final median centering of the data. For the other methods, analyses were run after the data were background corrected (normexp) with a 'within arrays' method for data centering (printtiploess), followed by the appropriate 'between arrays' method (none, quantile or scale normalisation).