ABSTRACT
The basic idea behind Bayesian models is to specify both a data sampling density and a prior distribution of all unknowns. A joint probability distribution then expresses the correct relationships between the unknowns and the data. Any inference on a specific Bayesian model is based on the posterior distribution, i.e., the conditional probability distribution of the unknowns given the data. The results of such posterior inferences can be used for decision making, forecasting, stochastic structure exploration and many other problems. However, the quality of these solutions depends heavily on the underlying Bayesian model. This crucial issue had been recognized by researchers and practitioners, leading to extensive investigations on the subject of Bayesian model evaluation.
