A Bayesian Model for the Analysis of Quantal Response Data
Abstract
Inference on the failure probabilities of ordered binomial trials conducted at M differing stress levels is considered. It is shown that a general joint prior may be constructed as a mixture of ordered M variate Dirichlet distributions, which possesses marginals of nearly arbitrary shapes. Posterior marginals at both observational and non-observational stresses are shown to consist of sums of beta distributions. Recursive relationships are developed that permit the rapid and exact computation of the posterior marginal distributions. The model is attractive for use in successive Bayesian analyses.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 05, 2003
- Accession Number
- ADA418744
Entities
People
- William W. Mcdonald
Organizations
- Naval Surface Warfare Center Indian Head Division