|
Activity Number:
|
121
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 12:00 PM to 1:50 PM
|
|
Sponsor:
|
General Methodology
|
| Abstract - #300592 |
|
Title:
|
Parameter Estimation for a Proposed Joint Distribution of Correlated Bernoulli Trials
|
|
Author(s):
|
Yufeng Li*+ and Charles R. Katholi
|
|
Companies:
|
University of Alabama, Birmingham and University of Alabama, Birmingham
|
|
Address:
|
153 Wallace Tumor Institute, Birmingham, AL, 35294,
|
|
Keywords:
|
correlated Bernoulli trial ; longitudinal ; maximum likelihood estimation ; Grey code
|
|
Abstract:
|
We have proposed a joint distribution for multivariate correlated Bernoulli trial. It combines the advantages of both Generalized Estimate Equation method and the Bahadur's multivariate distribution. This distribution can be applied to repeated measurements of binary outcome on study subjects and studies of family disease. It has potential application with higher dimensions of longitudinal or repeated data analysis. The parameters have been estimated numerically with quasi-Newton method with line search technique both under no restricted maximum likelihood estimation and restricted maximum likelihood estimation. We adopted the Grey Code to calculate scale parameter from log likelihood function of the distribution. Simulation study shows that the estimation for parameters, marginal probabilities and correlation among response are robustness from initial value selected. We have also developed statistical inference by using likelihood ratio test. We can test null hypothesis that all successive probability are equal. We can also test the hypothesis whether the correlation among response have certain type of structure. We have applied this distribution to a real data analysis.
|
