|
Activity Number:
|
228
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #301776 |
|
Title:
|
Estimation of a Weibull PHM with Time-dependant Stochastic Covariates from Incomplete Data Using a Variant of the ECME Algorithm
|
|
Author(s):
|
Dragan Banjevic*+
|
|
Companies:
|
University of Toronto
|
|
Address:
|
5 King's College Rd., Toronto, ON, M5S 3G8, Canada
|
|
Keywords:
|
proportional hazards ; Markov process ; stochastic covariates ; incomplete data ; EM algorithm ; reliability
|
|
Abstract:
|
A method for estimation of parameters of a proportional-hazards model (PHM) with time-dependent stochastic covariates from incomplete covariate data is discussed. It is assumed that the failure-time process and the covariate process follow a nonhomogeneous multidimensional Markov process. A variant of the ECME algorithm of Liu and Rubin for simultaneous estimation of both the PHM parameters and Markov process transition rates is considered. Convenient recursive formulas are derived necessary to perform E-step in the calculation, as well as to calculate the observed information matrix, required for direct minimization in Newton-Raphson method and for estimation of the standard errors. The recursive formulas also include solution to an extremely important problem in the calculation, that is, to the scaling of very small probabilities and expectations that may cause underflow in the computation of the likelihood function. The method is applied to real data from industry obtained from equipment condition monitoring and associated failure times.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2004 program |
