Modelling categorical covariates in Bayesian disease mapping by partition structures

We consider the problem of mapping the risk from a disease using a series o f regional counts of observed and expected cases, and information on potent ial risk factors. To analyse this problem from a Bayesian viewpoint, we pro pose a methodology which extends a spatial partition model by including cat egorical covariate information. Such an extension allows detection of clust ers in the residual variation, reflecting further, possibly unobserved, cov ariates. The methodology is implemented by means of reversible jump Markov chain Monte Carlo sampling. An application is presented in order to illustr ate and compare our proposed extensions with a purely spatial partition mod el. Here we analyse a well-known data set on lip cancer incidence in Scotla nd. Copyright (C) 2000 John Wiley & Sons, Ltd.