Activity Number:
|
252
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, August 13, 2002 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Section on Bayesian Stat. Sciences*
|
Abstract - #301137 |
Title:
|
Bayesian Analysis of a Survival Model
|
Author(s):
|
Subhashis Ghosal*+ and Sujit Ghosh
|
Affiliation(s):
|
North Carolina State University and North Carolina State University
|
Address:
|
220 Patterson Hall, Box 8203, 2501 Founders Drive, Raleigh, North Carolina, 27695, U.S.A.
|
Keywords:
|
survival analyis ; covariate ; censoring ; Dirichlet process ; mixture ; consistency
|
Abstract:
|
We consider a mixture of exponentials as a model for survival distribution in presence of a vector of covariates. Conditionally, on the value of a parameter $\lambda$, survival distributions are assumed to be independent exponential with parameters $\lambda\exp(-z_i'\beta)$. The unknown parameter $\lambda$ is assumed to follow a distribution $H$, which is completely unknown. Observations are assumed to be randomly right-censored. This model has an interesting property that the mean lifetime of observations, rather than the hazard rates, are directly related to the covariates. We do a Bayesian semiparametric analysis of the model by putting a Dirichlet prior on $H$. The posterior can be easily computed by MCMC algorithms. The resulting posterior is consistent under certain conditions.
|