P Green and A. O'Hagan
University of Bristol and University of Nottingham
Publication details: Nottingham University Statistics Research Report 98-01. 1998.
Carlin and Chib proposed a Markov chain Monte Carlo method for exploring alternative models. We show that, contrary to a claim by Carlin and Chib, it is not necessary to sample from the pseudo-priors which are a feature of their approach. Although this may improve the efficiency of the Carlin and Chib method, examination of one of their examples suggests that the reversible jump method of Green is considerably better. For this example, we also calculate the fractional Bayes factor, which avoids the specification of artificial proper priors, which are needed by both the Carlin and Chib method and reversible jump.