In the past fifteen years computational statistics has been enriched by a powerful, somewhat abstract method of generating variates from a target probability distribution that is based on Markov chains whose stationary distribution is the probability distribution of interest. This class of methods, popularly referred to as Markov chain Monte Carlo methods, or simply MCMC methods, have been influential in the modern practice of Bayesian statistics where these methods are used to summarize the posterior distributions that arise in the context of the Bayesian prior-posterior analysis (Tanner and Wong, 1987; Gelfand and Smith, 1990; Smith and Roberts, 1993; Tierney, 1994; Besaget al., 1995; Chib and Greenberg, 1995, 1996; Gilks et al., 1996; Tanner, 1996; Gammerman, 1997; Robert and Casella, 1999; Carlin and Louis, 2000; Chen et al., 2000; Chib, 2001; Congdon, 2001; Liu, 2001; Robert, 2001; Gelman at al, 2003). MCMC methods have proved useful in practically all aspects of Bayesian inference, for example, in the context of prediction problems and in the computation of quantities, such as the marginal likelihood, that are used for comparing competing Bayesian models.
