|
Activity Number:
|
74
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 6, 2006 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #306648 |
|
Title:
|
Semiparametric Models and Sensitivity Analysis of Longitudinal Data with Nonrandom Dropouts
|
|
Author(s):
|
David Todem*+ and KyungMann Kim and Jason P. Fine
|
|
Companies:
|
Michigan State University and University of Wisconsin-Madison and University of Wisconsin-Madison
|
|
Address:
|
Department of Epidemiology, B601 W. Fee Hall, East lansing, MI, 48823,
|
|
Keywords:
|
exponential family distribution ; functional estimators ; global tests and extreme statistics ; incomplete longitudinal data ; nonparametric mixture ; uniform convergence
|
|
Abstract:
|
We propose a family of semi-parametric non-response models to adjust for informative dropouts in the analysis of longitudinal data. The approach conceptually focuses on generalized linear mixed effects models with an unspecified random effects distribution. A novel formulation of a shared latent class model is presented and shown to provide parameters that have a meaningful interpretation. We show how to use the non-identifiability of some model characteristics to construct new global tests of covariate effects over the whole support of the sensitivity parameter. Simulations demonstrate a large reduction of bias for the nonparametric model relative to the parametric model at times where the dropout rate is high or the dropout model is misspecified. The methodology's practical utility is illustrated in a psychiatric data analysis.
|
