. Author manuscript; available in PMC: 2019 Aug 20.

Published in final edited form as: J Comput Graph Stat. 2018 Aug 20;27(4):763–772. doi: 10.1080/10618600.2018.1474115

Table 1. Simulation - Continuous covariates.

The average proportion of datasets (±SE) in which the five true biomarkers are ranked among the top five according to three measures of variable importance, namely (1) original RSF algorithm (p_RSF ); and (2) variable importance from the modified RSF algorithm (p₁). 1 − S_J₊₁, φ₁, φ₀ denote the cumulative incidence in the reference group, sensitivity and specificity, respectively.

1 − S_j₊₁	φ₁	φ₀	No missing data		Missing all data following first positive self-report

			p_RSF	p₁	p_RSF	p₁
0.10	1.00	1.00	0.736(±0.0441)	0.812(±0.0391)	0.736(±0.0441)	0.786(±0.0410)
	0.75	1.00	0.716(±0.0451)	0.748(±0.0434)	0.716(±0.0451)	0.768(±0.0422)
	0.61	0.995	0.636(±0.0481)	0.684(±0.0465)	0.636(±0.0481)	0.712(±0.0453)
	1.00	0.90	0.518(±0.0500)	0.760(±0.0427)	0.518(±0.0500)	0.586(±0.0493)

0.30	1.00	1.00	0.792(±0.0406)	0.828(±0.0377)	0.792(±0.0406)	0.850(±0.0357)
	0.75	1.00	0.764(±0.0425)	0.794(±0.0404)	0.764(±0.0425)	0.794(±0.0404)
	0.61	0.995	0.686(±0.0464)	0.708(±0.0455)	0.686(±0.0464)	0.736(±0.0441)
	1.00	0.90	0.620(±0.0485)	0.814(±0.0389)	0.620(±0.0485)	0.670(±0.0470)