Figure 10.

Impact of modeling approach and risk threshold on the proportion of counties wrongly classified as having low or high cancer mortality risk. Fifty realizations of the spatial distribution of cancer mortality rates were simulated and then analyzed using: Bayesian approach (BYM model), point Poisson kriging (PK) based on adjacent counties, and area-to-area Poisson kriging (ATA PK) based either on adjacent counties (same neighbors as BYM model) or the 32 closest counties in terms of distance between population-weighted centroids. The probability for each county to exceed a risk threshold proportional to the area-wide rate was averaged for counties that actually exceed or not that threshold. The ratio of these probabilities is a measure of discriminatory power and plotted as a function of the risk threshold (A, B). Counties with a probability larger than 0.75 are flagged as having significantly higher risk, and the resulting proportion of false positives (C, D) and negatives (E, F) are plotted as a function of the risk threshold.