Marginal Likelihood From the Metropolis–Hastings Output

摘要: This article provides a framework for estimating the marginal likelihood purpose of Bayesian model comparisons. The approach extends and completes method presented in Chib (1995) by overcoming problems associated with presence intractable full conditional densities. proposed is developed context MCMC chains produced Metropolis–Hastings algorithm, whose building blocks are used both sampling estimation, thus economizing on prerun tuning effort programming. Experiments involving logit binary data, hierarchical random effects clustered Gaussian Poisson regression count multivariate probit correlated to illustrate performance implementation method. These examples demonstrate that practical widely applicable.