Bayesian Analysis Statistical Preliminaries推荐.pdfVIP

Statistical Preliminarie s We review briefly some of th e background that is common to both classical statistics and Bayesian analysis . More details are available in Casella and Berger (1990), Lehmann and Casella (1998), and Bickel and Doksum (2001) . The reader interested in Bayesian analysis can go directly t o Chapter 2 after reading Section 1.1. 1.1 Commo n Model s A statistician , who ha s been given some dat a for analysis, begins by providing a probabilistic model of th e way his dat a have been generated . Usually th e data can be treate d as generated by random sampling or some other random mechanism . Once a model is chosen, th e dat a are treate d as a random vec- tor X = (Xi,X2,.. . jXn) Th e probability distribution of X is specified by f{x\6) which stand s for a joint density (or a probability mass function), and 0 is an unknown constant or a vector of unknown constant s called a parameter . The parameter 0 may be th e unknown mean and variance of a population from which X is a random sample , e.g., th e mean life of an electric bulb or the probability of doing something , vide Example s 1.1, 1.2, and 1.3 below. Often th e dat a X are collected t o learn about 0, i.e., th e modeling precedes collection of data . Th e set of possible values of ^ , called th e parameter space, is denoted by 0 , which is usually a p-dimensional Euclidean space IZ^ or some subset of it , p being a positive integer . Our usual notation for dat a vector and parameter vector are X and 6, res