Multiple Indicator and Multivariate Discrete State Hidden Markov Models

Melanie Wall
Division of Biostatistics
University of Minnesota

Wednesday, November 8th 2006
3:30pm
Moos 2-690
Minneapolis Campus

Abstract:
In the modeling of several longitudinal processes, it is often of interest to describe changes across time of each variable as well as the relationships among the process across time. If the processes themselves are best modeled as discrete stages and the measurements of these stages are not taken directly, but instead via some observable variable or variables expected to measure the stage with error, then a hidden Markov model may be appropriate where the time process and relationship among the variables is modeled directly on the underlying latent variables or latent stages. For modeling relationships among processes, a multivariate hidden Markov model is developed. Two examples serve as motivation: one which postulates a hidden state process underlying multiple types of medical encounters collected monthly within an insurance claims database where the states characterize individuals' medical use prior to and after alcoholism treatment, and the second example coming from a coronary heart disease clinical trial where it is of interest to model the relationship among several longitudinal processes (heart function, physical symptoms, and
quality of life) all hypothesized as discrete latent variables underlying observed clinical and self-report measurements.

A social tea will be held at 3:00 P.M. in A434 Mayo. All are Welcome.
For more details contact 612-624-4655 or see http://www.biostat.umn.edu/seminar_academic.html