No abstract

No abstract

No abstract

No abstract

The posterior distribution for parameters of a data distribution is usually the major objective of a Bayesian statistical analysis. Relatively little attention has been given to the fact that either a prior or a posterior distribution implies a marginal distribution, which may be called a "predictive distribution," for outcomes of any sample not yet observed. Predictive distributions have been mainly applied to design problems, such as determination of optimal sample size. I n this paper tentative suggestions are made for applications to statistical inference, especially problems of appropriateness, selection, interpretation, and validation of formal models.