|
Activity Number:
|
281
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 14, 2002 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Nonparametric Statistics*
|
| Abstract - #301840 |
|
Title:
|
Bayesian Approaches to Generalized Additive Mixed Models
|
|
Author(s):
|
Yisheng Li*+ and Xihong Lin
|
|
Affiliation(s):
|
University of Michigan and University of Michigan
|
|
Address:
|
1420 Washington Heights, Ann Arbor, Michigan, 48109-2029, USA
|
|
Keywords:
|
generalized additive mixed model ; Gibbs sampling ; smoothing spline ; Dirichlet process prior
|
|
Abstract:
|
We consider fully Bayesian approaches to making inference in generalized additive mixed models (GAMMs) for correlated data. GAMMs model covariate effects nonparametrically using smoothing splines and account for the correlation using random effects. We first concentrate on Gaussian outcomes. Gibbs sampling is used for estimation of all model components including the nonparametric functions, random effects, variance components and smoothing parameters. We first assume the distribution of the random effects to be normal and then relax the normality assumption by allowing the random effects distribution to be nonparametric using a Dirichlet process prior. We apply the proposed methods to a longitudinal study on the reproductive hormone progesterone, and evaluate their performance through simulations. We extend the proposed approaches to GAMMs for binary outcomes. Simulation results are provided to evaluate their performance. A real data set was analyzed using the proposed methods.
|
