DIAGNOSTIC PLOTS FOR IMPROVED PARAMETERIZATION IN BAYESIAN-INFERENCE

A major problem with the Bayesian analysis of statistical models is th at the computation of posterior and predictive summaries typically req uires the numerical evaluation of complicated multivariate integrals. The accuracy and efficiency of these computations are usually sensitiv e to the choice of parameterization for the statistical model. This pa per describes a graphical method which can be used both for assessing a proposed parameterization and choosing a new improved parameterizati on.