Fall 2002 Seminar Series - September 26, 2002
University of Minnesota
School of Statistics
College of Liberal Arts
A Bayesian Analysis of the Bivariate Censoring Model
R. V. Ramamoorthi
Department of Statistics and Probability, Michigan State University
Thursday, September 26, 2002
4:00 PM, 115
Ford Hall
Minneapolis, East Bank Campus
Social at 3:30 PM, 300
Ford Hall
Abstract
Let (T1, T2) be a pair
of survival times with joint distribution F, and let (C
1, C2) be a pair of censoring variables, independent
of the survival times. Let Zi = min(T
i, Ci), ∆i =
I (Ti ≤ Ci), i
= 1,2. The goal is to estimate F based on independent observations
of ( Z1, Z2, ∆1, ∆2
).
Pruitt showed that a Dirichlet prior for F
can lead to inconsistency of the posterior distributions. That is the
posterior, as the number of samples goes to infinity, may fail to converge
to the true distribution. In this talk I will explore construction of
nonparametric priors that yield posteriors that are both tractable and consistent.
This is part of ongoing work with J. K. Ghosh and
Charles Messan.