Univariate and multivariate Conditional AutoRegressive models for areal data
Fedele Greco (University of Bologna)
Friday 26th November, 2010 15:00-16:00 326
Areal data modelling has seen a considerable growth in last years following the increasing availability of Geographic Information Systems and spatial datasets. Among the various fields of applications, disease mapping is probably the leading one, both because mortality data are often available at area level due to confidentiality restrictions and because smoothed maps of relative risks are very informative in planning public health policies. Despite disease mapping studies have been widely performed at univariate level, contributes concerning multivariate modelling have been only recently developed. The first part of the seminar in focused on the discussion of prior specification in hierarchical Bayesian models for disease mapping. In the second part, an extension of the well-known univariate CAR model to cope with multivariate is discussed. In particular, a multivariate distribution for spatial random effects is constructed by explicit modelling of cross-correlation that is allowed to be asymmetric. The proposed methodology is implemented in a Bayesian framework by means of MCMC algorithms and is illustrated via its applications in multivariate disease mapping.