Introduction to Bayesian Analysis - University of （介绍贝叶斯分析-大学）.pdf

Introduction to Bayesian Analysis c Lecture Notes for EEB 596z, B. Walsh 2002 As opposed to the point estimators (means, variances) used by classical statis- tics, Bayesian statistics is concerned with generating the posterior distribution of the unknown parameters given both the data and some prior density for these parameters. As such, Bayesian statistics provides a much more complete picture of the uncertainty in the estimation of the unknown parameters, especially after the confounding effects of nuisance parameters are removed. Our treatment here is intentionally quite brief and we refer the reader to Lee (1997) and Draper (2000) for a complete introduction to Bayesian analysis, and the introductory chapters of Tanner (1996) for a more condensed treatment. While very deep (and very subtle) differences in philosophy separate hard-core Bayesians from hard-core frequentists (Efron 1986, Glymour 1981), our treatment here of Bayesian methods is motivated simply by their use as a powerful statistical tool. BAYES’ THEOREM The foundation of Bayesian statistics is Bayes’ theorem. Suppose we observe a random variable and wish to make inferences about another random variable , where is drawn from some distribution . From the definition of conditional probability, Again from the definition of conditional probability, we can express the joint probability by conditioning on to give Putting these together gives Bayes’ theorem: With possible outc