|
Activity Number:
|
29
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 3, 2003 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #302106 |
|
Title:
|
Marginalized Regression Models for Categorical Data with both Serial and Long-Range Dependence
|
|
Author(s):
|
Jonathan S. Schildcrout*+ and Patrick J. Heagerty
|
|
Companies:
|
University of Washington and University of Washington
|
|
Address:
|
Department of Biostatistics, Seattle, WA, 98195,
|
|
Keywords:
|
marginalized model ; longitudinal ; categorical ; likelihood
|
|
Abstract:
|
Marginalized regression models permit likelihood-based estimation of marginal regression parameters with longitudinal response data. Marginalized latent variable models structure dependence with random effects while marginalized transition models use past responses to describe dependence. Specifically, a random intercept model assumes that correlation among responses is induced via a single scalar latent variable that is shared by all observations. Such models assume equal short-range and long-range dependence. In contrast, a first-order transition model uses local dependence on past outcomes to describe serially correlated data. Our goal is to introduce a model for long series of categorical response data that combines the marginalized random intercept model with the first order marginalized transition model and therefore permits estimation of marginal regression parameters with a dependence model containing both short-range serial and long-range components. We compare operating characteristics of the parameter estimates derived from this model to those from previously developed estimating procedures.
|
