An automated stopping rule for MCMC convergence assessment
Tóm tắt
In this paper, we propose a methodology essentially based on the Central Limit Theorem for Markov chains to monitor convergence of MCMC algorithms using actual outputs. Our methods are grounded on the fact that normality is a testable implication of sufficient mixing. The first control tool tests the normality hypothesis for normalized averages of functions of the Markov chain over independent parallel chains started from a dispersed distribution. A second connected tool is based on graphical monitoring of the stabilization of the variance after n iterations near the limiting variance. Both methods work without knowledge on the sampler driving the chain, and the normality diagnostic leads to automated stopping rules. These stopping rules are implemented in a software toolbox whose performances are illustrated through simulations for finite and continuous state chains reflecting some typical situations and a full scale application. Comparisons are made with the binary control method of Raftery and Lewis.
Tài liệu tham khảo
Brooks, S., and Roberts, G. (1995), “Diagnosing Convergence of Markov Chain Monte Carlo Algorithms”, Tech. report 95-12, Stat. Lab., U. of Cambridge.
Capéraà, P., and Van Cutsem, B. (1988), Méthodes et Modèles en Statistique non paramétrique, Dunod.
Chauveau, D. and Diebolt, J. (1998), An Automated Stopping Rule for MCMC Convergence assessment Research Report RR-3566, INRIA Rhône-Alpes.
Chung, K.L. (1967), Markov Chains with Stationary Transition Probabilities, Springer-Verlag, Berlin.
Diaconis, P., Graham, R. and Morrison, J. (1990), “Asymptotic analysis of random walks on a hypercube with many dimensions,” Random structures and algorithms, 1, 52–72.
Diebolt, J. and Robert, C.P. (1994), “Estimation of Finite Mixture Distributions by Bayesian Sampling,” Journal of the Royal Statistical Society, B56, 363–375.
Gelfand, A.E., and Smith, A.F.M. (1990), “Sampling Based Approaches to Calculating Marginal Densities,” Journal of the American Statistical Association, 86, 398–409.
Gelman, A. and Rubin, D. B. (1992), “Inference from Iterative Simulation Using Multiple Sequences,” Statistical Science, 7, no. 4, 457–511.
Guihenneuc-Jouyaux, C. and Robert, C.P. (1998), “Finite Markov chain convergence results and MCMC convergence assessment”, Journal of the American Statistical Association, to appear.
Kemeny, J.G. and Snell, J.L. (1960), Finite Markov Chains, Springer-Verlag, New York.
Raftery, A.E., and Lewis, S. (1992), “How many iterations in the Gibbs Sampler?,” in Bayesian Statistics, J.O. Berger, J.M. Bernardo, A.P. Dawid and A.F.M. Smith (eds.), 4, 763–773, Oxford University Press, Oxford.
Raftery, A.E., and Lewis, S. (1996), “Implementing MCMC,” in Markov Chain Monte Carlo in Practice, W.R. Gilks, S.T. Richardson and D.J. Spiegelhalter (eds.), pp. 115–130, Chapman and Hall, London.
Robert, C.P. (1996), Méthodes de Monte Carlo par Chaînes de Markov, Economica, Paris.
Shapiro, S.S., and Wilk, M.B. (1965) “An analysis of variance test for normality”, Biometrika 52, 591–611.
Seoh, M. and Hallin, M.(1997), “When does Edgeworth beat Berry and Esséen?,” Journal of Statistical Planning and Inference, (to appear).
Tierney, L. (1994), “Markov Chains for Exploring Posterior Distributions (with discussion),” Annals of Statistics, 22, 1701–1762.