Skip to main content

Table 8 Performance comparison of alternative estimators: variance of risk estimates. Results obtained on average over 100 realizations generated under two different population size scenarios and 3 types of risk map (1 = observed, 2 = smooth, 3 = random). Poisson kriging was conducted with the semivariogram estimated from the underlying risk values (true γR(h)) or the simulated mortality rates. Bold numbers refer to best performances (i.e. the variance of risk estimates is the closest to the true risk variance reported in the first row) outside the ideal case where the true semivariogram of risk is known.

From: Geostatistical analysis of disease data: estimation of cancer mortality risk from empirical frequencies using Poisson kriging

Estimators WF population BF population
BREAST CANCER Scenario 1 Scenario 2 Scenario 3 Scenario 1 Scenario 2 Scenario 3
True risk values 7.891 5.982 7.891 7.891 5.982 7.891
Population-weighted average 5.635 4.560 0.346 8.538 6.881 3.035
Global Empirical Bayes 3.628 1.722 5.178 0.290 0.145 1.026
Local Empirical Bayes 6.571 4.941 5.257 9.364 8.576 5.339
Poisson kriging (true γR(h)) 6.792 5.176 5.561 8.156 6.781 4.161
Poisson kriging 6.922 5.093 5.498 8.332 7.044 4.143
CERVIX CANCER       
True risk values 0.946 0.596 0.946 0.946 0.596 0.946
Population-weighted average 0.485 0.498 0.055 0.909 0.918 0.422
Global Empirical Bayes 0.474 0.213 0.593 0.132 0.050 0.162
Local Empirical Bayes 0.727 0.538 0.625 3.197 2.059 2.746
Poisson kriging (true γR(h)) 0.793 0.599 0.671 0.941 0.910 0.550
Poisson kriging 0.804 0.596 0.607 1.269 1.137 0.690