|
Activity Number:
|
205
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2003 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Stat. Sciences
|
| Abstract - #301299 |
|
Title:
|
A Bayesian Approach to Latent Trajectory Models for Longitudinal Data
|
|
Author(s):
|
Sujata M. Patil*+ and Trivellore E. Raghunathan and Jean Shope
|
|
Companies:
|
University of Michigan and Institute for Social Research and University of Michigan
|
|
Address:
|
Biostatistics, Transportation Research, Ann Arbor, MI, 48109-2150,
|
|
Keywords:
|
Bayesian ; trajectory ; longitudinal ; growth curves
|
|
Abstract:
|
Longitudinal data are often used to study individual growth curves or 'trajectories.' These trajectories are not directly observed and require estimation. The analyses considered here explore the predictive relationship between latent trajectories and a future event. Analysis methods generally proceed in two stages. In stage 1, individual latent trajectories are estimated and summarized by two or more latent trajectory variables. For example, in polynomial models the coefficients for linear or quadratic terms are the latent trajectory variables. In stage 2 the latent trajectory variables are used as predictors of a future event. In the traditional approach, the stage 2 model conditions on estimated trajectories, ignoring measurement error and affecting inference. In a competing approach, parameter estimates and variances are corrected for measurement error. We develop a Bayesian approach that accounts for all uncertainties and allows for extensions to the model. We apply these three methods to study trajectories of adolescent alcohol use as predictors of vehicle crashes incurred during young adulthood. Results from a simulation study examining the three methods are also reported.
|
