Table 3.

False discovery rates of variable selection methods, for each scenario

Method	Pairwise correlation between all variables
	0.2	0.5	0.8
	5 signals among 20 variables
One-at-a-time*	0.09 (0.01)	0.58 (0.02)	0.40 (0.02)
Stepwise	<0.01 (<0.01)	<0.01 (<0.01)	0.01 (<0.01)
Two-step BVS*	<0.01 (<0.01)	<0.01 (<0.01)	<0.01 (<0.01)
	5 signals among 100 variables
One-at-a-time*	0.25 (0.02)	0.77 (0.02)	0.67 (0.02)
Stepwise	0.01 (<0.01)	0.03 (0.01)	0.08 (0.01)
Two-step BVS*	<0.01 (<0.01)	0.01 (<0.01)	0.01 (<0.01)
	5 signals among 1000 variables
One-at-a-time*	0.53 (0.03)	0.93 (0.01)	0.78 (0.02)
Stepwise	0.07 (0.01)	0.14 (0.01)	0.27 (0.01)
Two-step BVS*	0.01 (<0.01)	0.01 (<0.01)	0.06 (0.01)

Mean false discovery rate, the proportion of noise variables selected, is displayed for 200 simulations with the corresponding Monte Carlo errors in brackets.

*Selection thresholds chosen to match the sensitivity of the stepwise method in each simulation, for which a nominal P-value inclusion threshold of 0.05 was used.