Skip to main content
. Author manuscript; available in PMC: 2018 Mar 16.
Published in final edited form as: Biometrics. 2017 Jun 21;74(1):8–17. doi: 10.1111/biom.12736

Table 2.

Ratio of MSE (saturated model MSE / alternative method MSE) and Monte Carlo (MC) bias and standard errors for each scenario with sample size n = 500 over 1,000 MC datasets. Scenarios 1–9 have 10 covariates and Scenario 10 has varying covariate set sizes (p). In the simulations with correlated covariates, ρ(Vi, Vj) = 0.6 for i ≠ j ≤ 5 and ρ(Vi, Vj) = 0 for i ≠ j > 5.

GLiDeR Backward selection MADR Adaptive lasso
Scenario MSE
Ratio
MC
Bias
MC
SD
MSE
Ratio
MC
Bias
MC
SD
MSE
Ratio
MC
Bias
MC
SD
MSE
Ratio
MC
Bias
MC
SD
Independent covariates

1 1.10 0.00 0.09 1.02 0.00 0.09 1.11 0.00 0.09 1.12 0.00 0.09
2 1.09 0.00 0.09 1.02 0.00 0.10 1.12 0.00 0.09 1.13 0.00 0.09
3 2.91 0.00 0.09 1.04 0.00 0.15 3.06 0.00 0.09 3.20 0.00 0.09
4 1.01 0.00 0.10 1.01 0.00 0.10 1.02 0.00 0.10 0.79 0.05 0.10
5 1.67 −0.04 0.63 1.05 −0.04 0.80 1.65 −0.03 0.64 1.64 −0.04 0.64
6 18.35 0.00 0.31 0.95 −0.03 1.36 18.53 0.00 0.31 16.30 0.00 0.33
7 1.14 −0.05 0.84 1.04 −0.06 0.88 1.12 −0.05 0.85 1.16 −0.05 0.83
8 1.26 0.01 0.12 1.04 0.01 0.13 1.25 0.02 0.12 1.35 0.01 0.11
9 1.05 0.00 0.11 1.04 0.00 0.11 1.06 0.00 0.11 1.06 0.00 0.11

Correlated covariates

1 1.16 0.00 0.09 1.01 0.00 0.10 1.20 0.00 0.09 1.20 0.00 0.09
2 1.09 0.00 0.10 1.02 0.00 0.10 1.09 0.00 0.10 1.14 0.02 0.10
3 5.55 0.00 0.13 0.94 −0.02 0.32 5.29 −0.01 0.14 7.49 0.02 0.11
4 1.05 0.01 0.11 1.04 0.01 0.11 1.08 0.01 0.11 0.25 0.20 0.11
5 3.30 0.39 1.99 1.02 0.05 3.64 2.68 0.45 2.20 2.14 0.99 2.15
6 305.39 −0.02 0.92 1.61 −0.10 12.69 74.41 0.02 1.86 199.97 0.05 1.14
7 1.27 0.06 0.86 1.09 0.05 0.93 1.15 0.07 0.91 1.96 0.05 0.69
8 1.28 0.01 0.15 1.12 0.01 0.16 1.29 0.02 0.15 1.29 0.02 0.15
9 1.05 0.00 0.11 1.06 0.00 0.11 1.06 0.00 0.11 1.06 0.00 0.11

Scenario 10

p = 100 1.60 0.00 0.26 * * * * * * 0.51 −0.42 0.20
p = 500 13.65 −0.06 0.32 * * * * * * 6.22 −0.42 0.20
p = 1000 13.76 −0.14 0.29 * * * * * * 7.43 −0.39 0.20

Bold indicates significant difference (5% significance level) between MSEs (testing equality) from the saturated method (full model) vs. the alternative method using the paired t-test.