Table 3.
Simulation results to evaluate the proposed penalized likelihood method with Lasso penalty when p > n, where the candidate regulators were simulated as continuous variables. five-fold CV (Lasso1)), 5-fold two-stage CV (Lasso2) and the BIC (Lasso3) are used to select the tuning parameter λ. As a comparison, results from single variable analysis based on the likelihood ratio test using p-value of 0.05 (LR1) and p-value of 0.05/p (LR2) are also presented. For each procedure, the column labeled with C (or IC) represents the average number of correctly (or incorrectly) identified variables and their SEs, and E represents the percentage of the simulations that identify the three relevant variables exactly
| p= 200 | p = 500 | |||||
|---|---|---|---|---|---|---|
| Method | C | IC | E | C | IC | E |
| (β0, β1, β6, β12) = (−0.5, −1.0, 1.5, 1.0) | ||||||
| Lasso1 | 2.72(0.66) | 9.41(4.80) | 0 | 2.27(1.10) | 9.28(7.45) | 0 |
| Lasso2 | 2.63(0.67) | 4.19(3.45) | 4 | 1.80(1.14) | 2.85(3.37) | 2 |
| Lasso3 | 2.74(0.55) | 4.05(3.08) | 5 | 2.50(0.82) | 4.92(3.75) | 4 |
| LR1 | 2.66(0.51) | 34.85(11.15) | 0 | 2.66(0.51) | 84.71(26.04) | 0 |
| LR2 | 1.75(0.73) | 2.17(2.25) | 1 | 1.57(0.73) | 3.01(3.29) | 1 |