Bayesian Statistics Seminar
North Carolina State University
presents
Dr. Jayaram Sethuraman
Florida State University
"Partition based priors to analyze data arising from failure models and censoring models"
ABSTRACT
A special class of dependent data are generated by Failure models and censoring models. In this talk, I will describe more general versions of these models than those that are currently in use. It will be based on initial distribution function which gets modified on the basis of previous observations and other environmental (or censoring) variables The statistical problem is the estimation of this initial distribution. The famous Kaplan-Meyer estimate is the frequentist method Frequentist methods have been used to study failure models this problem, sometimes imposing further conditions. In this talk, I will define partition based (PB) priors, partition based Dirichlet (PBD) priors which generalize Dirichlet priors. The posterior distributions are easy to describe in all these models. Many helpful simplifications occur when one uses PBD priors. I will illustrate this by comparing Bayes estimates with the frequentist estimates. I will also describe the usual way Bayes hierarchical models are described in the literature, with examples. The likelihood and posterior distributions which are stated in these contexts are all in error since the specification of the model is incomplete for a probabilistic treatment of the problem.
Lectures Slides: Slides I and Slides II
Tuesday, October, 21, 2008
4:00 - 5:00 pm
208 Patterson Hall