|
Activity Number:
|
503
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #309104 |
|
Title:
|
Modeling Bivariate Survival Times by Frank Copula
|
|
Author(s):
|
Rui Qin*+ and Michael P. Jones
|
|
Companies:
|
Mayo Clinic and The University of Iowa
|
|
Address:
|
Division of Biostatistics, Rochester, MN, 55905,
|
|
Keywords:
|
bivariate survival times ; copula ; pseudolikelihood ; negative association
|
|
Abstract:
|
Copula models with relative risk margins are proposed for bivariate survival time data. The comprehensive family of Archimedean copulas provides flexibility in modeling different correlations. Estimation occurs in two stages. First, Cox regression is used to estimate the marginal survival functions. Second, a pseudo-likelihood for the association parameter is constructed by plugging in the marginal estimators and is then maximized over the association parameter. Empirical process theory is applied to establish consistency and asymptotic normality of the two-stage estimator. Simulation is used to study the behavior of the estimator and the asymptotic conclusions in small to medium size samples. The Frank copula exhibits some advantages over the popular Layton copula. An example with a real dataset is given.
|
