Table 2.

Empirical power (and mean number of prognostic genes selected in the prediction models) from 100 simulation data sets under β₁ = β₂ = β₃ = 0.3, p = 200, n = 200, and block compound symmetry correlation structure with a block size of 20 and a correlation coefficient of ρ = 0.2. The power for CV methods are approximated by 200 permutations.

Censoring proportion	Resampling method	Select top
		Three genes	Six genes
15%	50-50 hold-out	0.63 (1.81)	0.51 (2.22)
	70-30 hold-out	0.63 (2.23)	0.59 (2.74)
	Twofold CV	0.86 (1.83)	0.81 (2.33)
	Fivefold CV	0.95 (2.50)	0.90 (2.79)
	10-fold CV	0.95 (2.61)	0.88 (2.87)
	LOOCV	0.92 (2.74)	0.62 (2.94)
30%	50-50 hold-out	0.62 (1.69)	0.54 (2.11)
	70-30 hold-out	0.63 (2.07)	0.50 (2.51)
	Twofold CV	0.71 (1.55)	0.71 (2.11)
	Fivefold CV	0.88 (2.28)	0.87 (2.67)
	10-fold CV	0.83 (2.43)	0.79 (2.78)
	LOOCV	0.46 (2.54)	0.44 (2.83)
50%	50-50 hold-out	0.38 (1.15)	0.33 (1.84)
	70-30 hold-out	0.39 (1.72)	0.35 (2.30)
	Twofold CV	0.52 (1.18)	0.47 (1.78)
	Fivefold CV	0.66 (1.84)	0.57 (2.42)
	10-fold CV	0.62 (2.00)	0.57 (2.55)
	LOOCV	0.22 (2.20)	0.21 (2.68)