|
Activity Number:
|
387
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #308925 |
|
Title:
|
Regression Models for Paired Censored Exponential and Gamma Distributions
|
|
Author(s):
|
Michael P. Jones*+ and Don R. McNeil
|
|
Companies:
|
Macquarie University and Macquarie University
|
|
Address:
|
Psychology Department, North Ryde, 2080, Australia
|
|
Keywords:
|
Bivariate censoring ; Survival ; Models
|
|
Abstract:
|
Background: Models for nonindependent observations are well-studied. While some situations are well covered, such as repeated failures, others, such as twin studies, are not. Aim: Develop models suitable for paired observations where one or both pair members (y, z) might be right-censored. Method: For both Exponential and Gamma distributions we consider 1) a Bayesian approach eliminating nuisance parameters by assuming a non-informative prior and 2) derivation of distributions of ratios (r=y/z). Likelihood functions are derived and maximized numerically. Results: (1) requires only a single parameter to represent the within-pair dependence while the ratio approach eliminates nuisance parameters altogether. Simulation studies show the asymptotic Normality of the MLEs for both (1) and (2) but suggest greater statistical power from (1). Conclusion: The proposed methods are simple.
|
