Bayesian Analysis Hypothesis Testing and Model Selection推荐.pdfVIP

6 Hypothesi s Testin g an d Mode l Selection For Bayesians, model selection and model criticism are extremely important inference problems . Sometimes these ten d t o become much more complicated than estimation problems . In thi s chapter , some of these issues will be dis- cussed in detail . However, all models and hypotheses considered here are low- dimensional because high-dimensional models need a different approach . Th e Bayesian solutions will be compared and contrasted with th e corresponding procedures of classical statistic s whenever appropriate . Some of th e discussion in thi s chapter is technical and it will not be used in th e rest of th e book . Those sections that are very technical (or otherwise can be omitted at first reading) are indicated appropriately . These include Sections 6.3.4, 6.4, 6.5, and 6.7. In Sections 6.2 and 6.3, we compare frequentist and Bayesian approaches t o hypothesis testing . We do th e same in an asymptotic framework in Section 6.4. Recently developed methodologies such as th e Bayesian P-value and some non-subjective Bayes factors are discussed in Sections 6.5 and 6.7. 6.1 Preliminarie s First, let us recall some notation from Chapter 2 and also let us introduce some specific notation for th e discussion that follows. Suppose X having density f{x\6) is observed, with 6 being an unknown el- ement of th e parameter space 0. Suppose that we are interested in comparing two models MQ and Mi , which are given by MQ : X ha s density f{x\6)