Modeling of rates over a hierarchical health administrative structure

Identifying the distribution of the incidence rate of a disease over a region is a prediction problem where area-specific random effects are to be estimated. The authors consider the inclusion of such effects at different levels of a hierarchical health administrative structure. They develop inference procedures for this type of multi-level model and show that the predicted rates are approximately weighted sums of the crude rates obtained by pooling data on each level of the hierarchy. Their techniques are illustrated using infant mortality data from British Columbia.