Skip to main content

Table 1 Bias and 95% coverage for full and single predictor models.

From: The behaviour of random forest permutation-based variable importance measures under predictor correlation

Predictor True
Value
Member of
Correlated Group
Full Model:
Bias
Full Model:
95% Coverage
Single Model:
Bias
Single Model:
95% Coverage
x 1 5 T 9.16e-04 94.6 6.31 0.0
x 2 5 T -6.09e-04 97.2 6.31 0.0
x 3 2 T -0.0014 94.6 9.02 0.0
x 4 0 T 0.0011 96.2 10.81 0.0
x 5 -5 F 6.77e-04 96.0 -0.015 94.2
x 6 -5 F 2.75e-04 95.2 -0.017 94.2
x 7 -2 F -3.90e-04 93.8 -0.0017 95.8
x 8 0 F -5.26e-04 95.8 0.0050 93.4
x 9 0 F 1.86e-04 94.6 -0.012 94.6
x 10 0 F 2.47e-04 94.4 0.012 95.8
x 11 0 F -3.60e-04 95.2 0.0096 94.4
x 12 0 F -3.56e-06 95.8 0.0040 95.8