Table 3.

Effect Size Measures and Variable Importance by Sample Size and Selection Model, Simulation A

	Mean effect size measures				Mean variable importance
	t test |Cohen’s d|			Logistic regression	CART			CART + Prune			RF
	v	z	w	McFadden’s pseudo-R2	v	z	w	v	z	w	v	z	w
N = 100
Linear	.563	.167	.170	.086	8.199	3.656	3.703	6.871	2.706	2.722	7.391	−.106	−.112
One split	.940	.175	.173	.177	15.224	3.431	3.257	13.903	1.729	1.687	23.271	.024	−.054
Two splits	.732	.470	.169	.155	11.125	8.117	2.065	10.667	7.485	1.423	20.075	11.71	−.147
Three splits	.166	.366	.508	.093	5.502	5.662	7.591	4.335	4.601	6.613	6.067	6.533	11.192
N = 250
Linear	.548	.105	.108	.070	18.810	8.242	7.85	14.412	5.141	5.007	12.175	−.055	−.037
One split	.927	.107	.105	.155	35.932	6.077	5.835	32.604	1.965	1.969	40.284	.157	−.141
Two splits	.731	.473	.105	.141	26.257	19.352	3.301	24.228	17.860	1.609	36.120	22.149	−.073
Three splits	.106	.354	.496	.077	15.179	13.58	17.005	11.488	10.093	14.539	13.759	12.736	20.681
N = 500
Linear	.546	.075	.074	.065	28.434	8.512	8.618	23.614	5.960	5.998	17.546	−.010	.052
One split	.920	.075	.078	.147	65.234	4.773	4.850	63.561	2.485	2.499	58.864	.011	−.010
Two splits	.737	.467	.074	.137	48.704	35.515	3.732	46.384	33.936	1.609	53.898	33.001	−.045
Three splits	.084	.355	.503	.074	26.227	22.838	29.152	21.728	17.958	26.093	22.843	20.421	31.863

Note. Cohen’s d indicates the mean of the absolute d values across simulated iterations. Random forest variable importance calculated using classification accuracy. CART = classification and regression trees; Prune = pruned CART analysis; RF = random forests.