|
Activity Number:
|
340
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #305955 |
|
Title:
|
A Robust Likelihood-Based Approach to Nonlinear Measurement Error Models with Application to Radiation Dose Effects on Leukemia-Specific Hazard Rate among A-Bomb Survivors
|
|
Author(s):
|
Antara Majumdar*+ and Randy L. Carter
|
|
Companies:
|
University at Buffalo and University at Buffalo
|
|
Address:
|
Department of Biostatistics, 264A Farber Hall, Buffalo, NY, 14214,
|
|
Keywords:
|
measurement error model ; instrumental variable ; maximum likelihood ; Monte Carlo EM algorithm ; radiation effects ; leukemia survival data
|
|
Abstract:
|
We propose a robust likelihood-based method to estimate the parameters of measurement error models. We combine three key points: First, the errors-in-variables problem can be viewed as a missing data problem. Second, the measurement error problem can be solved with availability of an instrumental variable. Third, the likelihood based on piece-wise exponential survival data is the same as the likelihood based on Poisson distribution. We construct a robust multivariate likelihood of the manifest variables utilizing all the above information and use the MCEM algorithm to maximize the observed data likelihood. Our method accommodates other covariates that are error-free, as well as facilitates testing of regression parameters. This work was motivated by the ongoing research at RERF. We illustrate our method by analyzing the incidence of deaths due to leukemia among A-bomb survivors.
|
