Table 2.
Performance comparison of alternative kriging estimators: mean errors and mean absolute errors of prediction.
| Estimators | Lung cancer | Cervix cancer | ||
| MEAN ERROR | Average | % best result | Average | % best result |
| Point kriging of raw rates | -0.013 | 41 | 0.060 | 11 |
| Point kriging of global EBS | 0.008 | 11 | 0.040 | 15 |
| Point kriging of local EBS | 0.042 | 8 | 0.044 | 23 |
| ATP Poisson kriging | -0.036 | 40 | -0.001 | 51 |
| ATP Poisson kriging (true γR(h)) | -0.032 | -0.001 | ||
| MEAN ABSOLUTE ERROR | ||||
| Point kriging of raw rates | 2.647 | 8 | 0.406 | 0 |
| Point kriging of global EBS | 2.694 | 3 | 0.406 | 1 |
| Point kriging of local EBS | 2.776 | 0 | 0.418 | 0 |
| ATP Poisson kriging | 2.471 | 89 | 0.317 | 99 |
| ATP Poisson kriging (true γR(h)) | 2.452 | 0.313 | ||
Results obtained on average over 100 realizations generated for Regions 1 and 2. Poisson kriging was conducted with the semivariogram model derived through deconvolution or inferred directly from the simulated grid risk values (true point-support model γR(h)). Bold numbers refer to best performances outside the ideal case where the true semivariogram of risk is known. The second column gives the percentage of realizations where the particular method (except ATP kriging with true γR(h)) yields the smallest prediction error.