Table 1 Comparison of model selection performance. O for over-selection, U for under-selection and Exact for exact-selection
Model | Working correlation matrix | O | U | Exact | MSE (SE) |
|---|---|---|---|---|---|
Sample size = 50 | |||||
Model 1: PGEE | AR1 | 15.5% | 82.0% | 2.5% | 0.47 (0.35) |
Model 2: PGEE | unstructured ⊗ AR1 | 19.0% | 79.0% | 2.0% | 0.45 (0.33) |
Model 3: GPGEE | AR1 | 1.0% | 5.5% | 93.5% | 0.49 (0.25) |
Model 4: GPGEE | unstructured ⊗ CS | 0.0% | 4.0% | 96.0% | 0.37 (0.18) |
Model 5: GPGEE | unstructured ⊗ AR1 | 2.0% | 3.5% | 94.5% | 0.24 (0.13) |
Sample size = 100 | |||||
Model 1: PGEE | AR1 | 11.0% | 75.0% | 14.0% | 0.25 (0.16) |
Model 2: PGEE | unstructured ⊗ AR1 | 33.5% | 53.0% | 13.5% | 0.16 (0.08) |
Model 3: GPGEE | AR1 | 1.5% | 2.0% | 96.5% | 0.18 (0.11) |
Model 4: GPGEE | unstructured ⊗ CS | 1.5% | 0.5% | 98.0% | 0.21 (0.12) |
Model 5: GPGEE | unstructured ⊗ AR1 | 1.0% | 0.5% | 98.5% | 0.13 (0.07) |