Bayes theory (Q792713)

In this book the author gives a unified treatment of Bayesian theory. Probability distributions with infinite total probability are considered and complications arising from infinite probabilities may be viewed as a main subject. The content of the book is as follows: In the initial chapters a historic review of probability theories is given and the axioms of probability are stated. The analysis is based on the betting theory interpretation of probability. In chapter 3 conditional probabilities are introduced. They are derived from a set of axioms generalizing the axioms of probability. A general form of Bayes' theorem for non-unitary probabilities is proved. In chapter 4 notions of convergence for non-unitary probabilities are considered. Various concepts of 'making probabilities' are discussed in chapter 5, e.g. similarity probabilities. Similarities are subjective judgements relating expected events to past knowledge from which probabilities are objectively derived. Decision theory and uniformity criteria for selecting decisions are the subject of chapters 6,7; the necessity of considering non-unitary probabilities is established. In the following chapters the distributional assumptions are specified. E.g. various priors for exponential families are considered, admissibility of ML estimates for the multinomial case is shown. In chapter 11 conditions for the asymptotic normality of the posterior distribution are given. The final chapters are dedicated to Bayesian robustness (with respect to the prior component) and to nonparametric Bayesian procedures. The presentation of the results is quite concise. Besides its independent value the book gives a general view of a lot of relatively new results in Bayesian theory.