Identification in Bayesian disease mapping and spatial regression
*Ying C. MacNab, University of British Columbia 

Keywords: Bayesian disease mapping and spatial regression; identification, univariate disease mapping, multivariate disease mapping

Bayesian hierarchical model formulation, with related methods for statistical inference, is at the core of the Bayesian disease mapping methodology. Many partially identified disease mapping models, such as the BYM convolution model for mapping of single disease or health outcome as well as the shared component models and some of the multivariate CAR models for mapping of two or more diseases and health outcomes, have been presented in disease mapping applications and implemented within the Bayesian hierarchical model framework where known priors or informative hyper-priors are specified. In this presentation I discuss the issue of identification in Bayesian disease mapping and spatial regression. Via Monte Carlo simulations, I show that the potential implication of partial identification in Bayesian disease mapping and spatial regression is biased posterior risk prediction - biases in the estimated posterior risk means and/or posterior risk standard derivations.