ABSTRACT
The conclusions of a Bayesian analysis are conditional on the appropriateness of an assumed probability model, so we need to be satisfied that our assumptions are a reasonable approximation to reality, even though we do not generally believe any model is actually “true.” Many aspects of an assumed model might be questioned: observations that don’t fit, the distributional assumptions, qualitative structure, link functions, which covariates to include, and so on. We can distinguish three elements that can be applied in an iterative man-
ner (O’Hagan, 2003).
