Figure 17.

Scatter of realizations in the multivariate space of performance criteria (cervix cancer). For each risk and population scenario, a principal component analysis is conducted on the correlation matrix of the seven performance criteria, and individual realizations are projected in the space of the first two rotated components. The estimators are: population-weighted mean (black), global (light blue) and local (dark blue) empirical Bayes smoothers, and Poisson kriging using the true semivariogram of risk (yellow) or the one modelled from observed mortality rates (red). The position of criteria labels on the graph indicates their correlation with the two principal components (criteria with small correlations are not displayed).