|
Abstract:
|
Maps of regional disease rates are useful tools for examining spatial patterns of disease. In this paper, we adopt a Bayesian approach for disease rate mapping and develop a new Markov random field prior. This prior has two hyperparameters which control the sparsity and smoothness of the estimated maps. To choose the hyperparameters, we discuss methods from both Bayesian and non-Bayesian views and propose a local resampling method for automatic selection. The maximum a posteriori (MAP) estimate is used as a summary of the posterior density function and we propose a new algorithm to calculate it efficiently. This new algorithm is more generally applicable than existing LASSO algorithms, such as LARs (see Efron et al. 2004), in terms of the ability to handle more general loss functions. This is a joint work with Professor Murray K. Clayton.
|
