Table 2.
Simulation results of the sparse association methods in combination with the model selection criteria for m = 102 assuming three different correlation structures for the exposure and outcome variables. For each of the three correlation matrices, the proportion of times out of 200 simulations in which the true model was selected (T), the mean number of variables that were false positives (FP) and false negatives (FN) are reported. The population CCA loadings of the outcome variables for the three correlation structures are b1 = (0.58, 0.58, 0.58, 0, …, 0)T, b2 = (0.54, 0.54, −0.65, 0, …, 0)T, b3 = c(0.71, −0.71, 0, …, 0)T respectively.
| cov1 | cov2 | cov3 | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Select | method | T | FP | FN | T | FP | FN | T | FP | FN |
| Step-CCA | 0.02 | 5.49 | 0.02 | 0.04 | 4.45 | 0.00 | 0.02 | 5.17 | 0.01 | |
| BIC | SOS-CCA | 0.80 | 0.22 | 0.01 | 0.00 | 0.34 | 1.17 | 0.61 | 0.38 | 0.16 |
| PMA-CCA | 0.92 | 0.11 | 0.02 | 0.00 | 0.16 | 1.19 | 0.00 | 1.11 | 1.14 | |
| Step-CCA | 0.02 | 4.22 | 0.02 | 0.29 | 1.25 | 0.00 | 0.02 | 4.80 | 0.00 | |
| CIC | SOS-CCA | 0.76 | 0.28 | 0.00 | 0.00 | 0.70 | 1.06 | 0.56 | 0.55 | 0.12 |
| PMA-CCA | 0.86 | 0.22 | 0.02 | 0.00 | 0.78 | 1.06 | 0.00 | 4.12 | 1.04 | |
| Step-CCA | 0.32 | 1.04 | 0.60 | 0.77 | 0.70 | 0.06 | 0.61 | 2.29 | 0.22 | |
| correlation | SOS-CCA | 0.46 | 1.92 | 0.11 | 0.00 | 7.20 | 0.78 | 0.14 | 8.73 | 0.09 |
| PMA-CCA | 0.30 | 7.66 | 0.05 | 0.00 | 10.16 | 0.98 | 0.01 | 21.50 | 0.80 | |