|
Activity Number:
|
248
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
General Methodology
|
| Abstract - #301975 |
|
Title:
|
Latent Variable Models for Survival Discrete Time Indicators
|
|
Author(s):
|
Irini Moustaki*+
|
|
Companies:
|
Athens University of Economics and Business
|
|
Address:
|
Department of Statistics, 104 34 Athens, , , Greece
|
|
Keywords:
|
latent variable models ; survival indicators ; EM algorithm
|
|
Abstract:
|
We discuss latent variable models for the case where some of the manifest indicators measure survival time. Survival data can be either censored or not censored. Both types of data will be discussed. In a latent variable framework survival indicators will usually occur with other types of variables such as categorical or metric. Therefore the models presented here will be discussed for the mixed case. We also allow for covariate effects to affect the manifest indicators and the latent variables in the model. The assumptions made are that the latent variables together with the observed covariates are responsible for the association among the manifest indicators and that the latent variables are independent. Interaction effects among the latent variables are also included in the model. The estimation of the model is based on the maximization of the marginal distribution of the manifest indicators using the EM algorithm. Goodness-of-fit measures and model selection criteria will be discussed together with ways of computing latent scores for the population elements on the identified latent dimensions. The methodology will be illustrated with a real example.
|
