Bayesian Analysis Bayesian Inference and Decision Theory推荐.pdfVIP

Bayesian Inferenc e an d Decision Theor y This chapter is an introduction t o basic concepts and implementation of Bayesian analysis . We begin with subjective probability as distinct from clas- sical or objective probability of an uncertain event based on th e long run relative frequency of its occurrence . Subjective probability, along with utility or loss function, leads t o Bayesian inference and decision theory, e.g., estima- tion, testing , prediction , etc . Elicitation of subjective probability is relatively easy when th e observa- tions are exchangeable . We discuss exchangeability, its role in Bayesian anal- ysis, and its importance for science as a whole . In most cases in practice , quantification of subjective belief or judgment is not easily available . It is then common t o choose from among conventional priors on th e basis of some relatively simple subjective judgment s about th e problem and th e conventional probability model for th e data . Such priors are called objective or noninformative . These priors have been criticized for vari- ous reasons . For example , they depend on th e form of th e likelihood function and usually are improper , i.e., th e tota l probability of th e parameter space is infinity. Here in Chapter 2, we discuss how they are applied ; some answers t o th e criticisms are given in Chapter 5. In Section 2.3 of thi s chapter , ther e is a brief discussion of th e many advantages of being a Bayesian . 2. 1 Subjective an d Frequentist Probability Pr