Table 4.
Comparison of imputation methods for β = (−0.5, 0.1) under the Cox model with heavy interval censoring.
| EXACT | Imputation methods |
|||||||
|---|---|---|---|---|---|---|---|---|
| RIGHT | MID | MEAN | MEDIAN | MODE | RAND | MI | ||
| Mean of β̂1 | −0.5036 (−0.5021) | −0.4832 (−0.4897) | −0.4938 (−0.4764) | −0.4829 (−0.4919) | −0.4813 (−0.4920) | −0.4823 (−0.4838) | −0.4792 (−0.4845) | −0.4786 (−0.4849) |
| SD(β̂1) | 0.1569 (0.1606) | 0.1578 (0.1589) | 0.1600 (0.1614) | 0.1569 (0.1596) | 0.1561 (0.1597) | 0.1590 (0.1610) | 0.1585 (0.1619) | 0.1554 (0.1568) |
| Mean ASE(β̂1) | 0.1554 (0.1553) | 0.1552 (0.1552) | 0.1553 (0.1553) | 0.1551 (0.1553) | 0.1551 (0.1553) | 0.1552 (0.1553) | 0.1551 (0.1553) | 0.1585 (0.1605) |
| MSE | 0.0246 (0.0258) | 0.0252 (0.0253) | 0.0256 (0.0266) | 0.0249 (0.0255) | 0.0247 (0.0256) | 0.0256 (0.0262) | 0.0255 (0.0264) | 0.0246 (0.0248) |
| Coverage of 95% CI (ASE) | 0.949 | 0.943 | 0.946 | 0.952 | 0.953 | 0.940 | 0.948 | 0.957 |
| Coverage of 95% EBCI | – | 0.938 | 0.944 | 0.948 | 0.949 | 0.955 | 0.953 | 0.942 |
| Coverage of 95% CI (ASE) | (0.942) | (0.947) | (0.944) | (0.945) | (0.945) | (0.949) | (0.935) | (0.956) |
| Coverage of 95% EBCI | (−) | (0.947) | (0.943) | (0.945) | (0.947) | (0.949) | (0.949) | (0.940) |
| Power (ASE) | 0.904 | 0.881 | 0.882 | 0.876 | 0.878 | 0.882 | 0.864 | 0.861 |
| Power (EBCI) | – | 0.890 | 0.887 | 0.884 | 0.880 | 0.875 | 0.877 | 0.885 |
| Power (ASE) | (0.912) | (0.903) | (0.870) | (0.901) | (0.903) | (0.890) | (0.887) | (0.883) |
| Power (EBCI) | (−) | (0.912) | (0.872) | (0.908) | (0.906) | (0.898) | (0.899) | (0.909) |
| Size (ASE) | 0.044 | 0.048 | 0.052 | 0.052 | 0.052 | 0.056 | 0.048 | 0.040 |
| Size (EBCI) | – | 0.059 | 0.060 | 0.057 | 0.055 | 0.052 | 0.054 | 0.059 |
| Size (ASE) | (0.060) | (0.062) | (0.053) | (0.064) | (0.064) | (0.055) | (0.061) | (0.042) |
| Size (EBCI) | (−) | (0.065) | (0.053) | (0.062) | (0.066) | (0.051) | (0.055) | (0.060) |
Note: Numbers not in parenthesis are from GA setting; numbers in parenthesis are from GB setting. Sample size n = 200, 1000 simulated data sets, and 1000 bootstraps per simulated dataset.