Bayesian Computation with R Single推荐.pdfVIP

3 Single-Parameter Models 3.1 Introduction In this chapter, we introduce the use of R in summarizing the posterior distri- butions for several single-parameter models. We begin by describing Bayesian inference for a variance for a normal population and inference for a Pois- son mean when informative prior information is available. For both problems, summarization of the posterior distribution is facilitated by the use of R func- tions to compute and simulate distributions from the exponential family. In Bayesian analyses, one may have limited beliefs about a parameter and there may be several priors that provide suitable matches to these beliefs. In esti- mating a normal mean, we illustrate the use of two distinct priors in modeling beliefs and show that inferences may or may not be sensitive to the choice of prior. In this example, we illustrate the “brute-force” method of summariz- ing a posterior where the density is computed by the “prior times likelihood” recipe over a fine grid. One way to generalize the family of conjugate priors is by the use of mixtures, and we illustrate the use of a mixture of beta distri- butions to model the belief that a coin is biased. We conclude by describing a Bayesian test of the simple hypothesis that a coin is fair. The computation of the posterior probability of “fair coin” is facilitated using beta and binom functions in R. 3.2 Normal Distribution with Known Mean but Unknown Variance Gelman et al. (2003) consider a problem of estimating an unknown variance using American football scores. The focus is on the difference d between a game outcome (winning score minus losing score) and a published point spread. We observe d1 , ..., dn , the observed differences between game outcomes and point spreads for n football games. If these differences are assumed to be a random J. Albert, Bayesian Computation with R, Use R, DOI 10.1007/978-0-387-92298-0 3, © Springer Science+Bu