On a multivariate implementation of the Gibbs sampler 英文参考文献.docVIP

D García-Cortés, Sorensen* (Received 2 August 1995; accepted 30 October 1995) Summary - It is well established that when the parameters in a model are correlated, the rate of convergence of Gibbs chains to the appropriate stationary distributions is faster and Monte-Carlo variances of features of these distributions are lower for a given chain and sampling from the respective conditional posterior distributions takes place in a multivariate rather than in a scalar fashion. This block sampling strategy often requires knowledge of the inverse of large matrices. In this note a block sampling strategy is implemented which circumvents the use of these inverses. The algorithm applies in the context of the Gaussian model and is illustrated with a small simulated data set. l’échantillonnage de Gibbs. Il est de ce type qui évite le passage par ces inverses. L’algorithme s’applique dans le contexte d’un modèle gaussien et est illustré numériquement sur un petit échantillon de données simulées. échantillonnage de Gibbs / échantillonnage en bloc / analyse bayésienne *  INTRODUCTION The Gibbs sampler is anumerical technique that has received considerable attention in statistics and animal breeding. It is particularly useful in the solution of high dimensional integrations and has therefore been applied in likelihood and Bayesian inference problems in a wide variety of models. The Gibbs sampler produces realizations from a joint posterior distribution by sampling repeatedly from the full conditional posterior distributions of the parameters of the model. In theory, drawing from the joint posterior density takes place only in the limit, as the number of drawings becomes infinite. The study of convergence to the appropriate distributions is still an active area of research, but it is well established that convergence can be slow when highly correlated scalar components are treated individually (Smith and Roberts, 1993). In such cases it is preferable to block the