Emerging Issues in Analysis of Longitudinal Data: Subject-specific Models

June 3, 9:00 am – 12:00 pm, 1:30 – 4:00 pm
Paul Rathouz, University of Wisconsin School of Medicine & Public Health
(http://www.biostat.wisc.edu/People/faculty/rathouz.htm )
In observational studies, longitudinal data often contain covariates that vary both between and within subjects (Neuhaus). Indeed, econometricians have long collected longitudinal data solely to exploit the natural within-subject covariability of exposure and outcome while automatically
controlling unobserved between-subject confounding factors (Wooldridge). Social epidemiologists have exploited covariates varying both between and within subjects to analyze contextual effects (Diez-Roux). This short-course will explore practical aspects of design
and analysis of studies with covariates with both between- and within-subject variability, focusing on readily-available tools and models and on model interpretation. I will draw connections between traditional fixed effects models in econometrics and random effects models in biostatistics. The material will be organized in four sessions along the following lines:
Session 1: 
- Introduction and generalities
- fixed effects models: specification and interpretation
- fixed effects model estimation and testing
- decomposition of time-varying covariates into between-subject and within-subject components
- connection between fixed effects and random effects models
- model estimation and testing revisited
Session 2: 
- Exploiting covariate decomposition
- model parameterization and interpretation
- examples from developmental psychopathology and from econometrics
- subject-specific intercept and slope
Session 3: 
- Exploiting covariate decomposition (cont.)
- power analysis / sample size for longitudinal or clustered-data designs
- interaction terms with covariates varying both between and within subjects
Session 4: 
- Special topics -- possibilities include
- missing data in designs with covariates varying between and within subjects
- outcome dependent sampling designs for longitudinal data
Brief Biography
Paul Rathouz is currently Professor and Chair of the Department of Biostatistics & Medical Informatics at the University of Wisconsin, Madison. Before that, he spent 13 years in the Department of Health Studies at the University of Chicago. He has biostatistical background in
theory, methodology, and practical applications in a variety of areas of biomedical and public health research. Areas of methodological interest include errors in regression covariates, missing data in models for highly stratified, longitudinal or survival (failure time) data, generalized
linear models, methods for multivariate, longitudinal or clustered data, outcome-dependent sampling, and structural equation models. He has served on several NIH study sections primarily in biostatistics methodology and mental health epidemiology, as Associate Editor of Biometrics, and on the elected governing board for the Eastern North America Region (ENAR) of the
International Biometric Society.  His honors include the James E. Grizzle Distinguished Alumnus Award from the Department of Biostatistics at the University of North Carolina, Chapel Hill and the ENAR Van Ryzin Award for best student paper during his graduate program.