Bayesian Computation with R Multiparameter Models推荐.pdfVIP

4 Multiparameter Models 4.1 Introduction In this chapter, we describe the use of R to summarize Bayesian models with several unknown parameters. In learning about parameters of a normal population or multinomial parameters, posterior inference is accomplished by simulating from distributions of standard forms. Once a simulated sample is obtained from the joint posterior, it is straightforward to perform transforma- tions on these simulated draws to learn about any function of the parameters. We next consider estimating the parameters of a simple logistic regression model. Although the posterior distribution does not have a simple functional form, it can be summarized by computing the density on a fine grid of points. A common inference problem is to compare two proportions in a 2 × 2 contin- gency table. We illustrate the computation of the posterior probability that one proportion exceeds the second proportion in the situation in which one believes a priori that the proportions are dependent. 4.2 Normal Data with Both Parameters Unknown A standard inference problem is to learn about a normal population where both the mean and variance are unknown. To illustrate Bayesian computation for this problem, suppose we are interested in learning about the distribution of completion times for men between ages 20 and 29 who are running the New York Marathon. We observe the times y1 , . . . , y20 in minutes for 20 runners , and we assume they represent a random sample from an N (μ, σ) distribution. If we assume the standard noninformative prior g (μ, σ2 ) ∝ 1/σ2 , then the posterior density of the mean and variance is given by g (μ, σ2 1 1 2 |y) ∝ (σ2 )n/2+1 exp − 2σ2 (S + n(μ − y¯) ) , n 2 where n is the sample s