Elements of multi-Bayesian decision theory (Q1063346): Difference between revisions

From authors' introduction: This paper presents a framework for a normative theory of multi-Bayesian statistical decision theory. A group of Bayesians, B, must jointly select a, possibly randomized, decision rule \(\delta\). The worth of \(\delta\) to each Bayesian \(\beta\in B\) is determined by his prior or posterior (as appropriate) expected gain (or loss) of utility. An obstacle to a uniquely acceptable solution to the group decision problem is the lack of a generally acceptable basis for inter-Bayesian utility comparisons, making it unreasonable to postulate that a group will possess a group preference ordering, which completely orders the class of available \(\delta\) 's and thereby points to a best choice. A method of assessing, by hypothetical subsampling, a prospective \(\delta\) is given and some of its implications are worked out. The acceptability of \(\delta\) is tested by ascertaining if it would remain the group's choice were it known that it would be implemented by, and change the utilities of, only the members of an as yet unspecified subgroup. Another method of assessment regards the elicited priors as observables which have with the data, a joint sampling distribution given the true state of nature.