Bayes Sampling Designs for Selection Procedures
Abstract
From k independent populations P1,...,Pk, which belong to one parameter exponential family ?Ftheta!, theta in omega reflex subset contained in R, random samples of sizes m1,...,mk, respectively, are to be drawn. After the observations have been drawn, a selection procedure will be used to determine which of these k populations has the largest value of theta. Given a loss for selections at each parameter configuration, given n past observations, and given a prior for the k parameters, a Bayes selection procedure can be found and its Bayes risk can be determined, where both depend on m1,...,mk. Let the sample sizes be restricted by m1 + ... +mk = m, where m is fixed. The problem of how to find the optimum (minimum Bayes risk) sample design subject to this constraint is considered, as well as m-truncated sequential sampling allocations. Results for normal and binomial families, under the '0-1' loss and the linear loss, are presented and discussed. An introduction to Bayes selection procedures in included.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1997
- Accession Number
- ADA332578
Entities
People
- Klausche J. Miescke
Organizations
- Purdue University