Marginal Analysis of Longitudinal Count Data from a Teenage Driving Study

• Activity Number: 170
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Zhiwei Zhang*+ and Paul S. Albert
• Companies: Eunice Kennedy Shriver National Institute of Child Health and Human Development and Eunice Kennedy Shriver National Institute of Child Health and Human Development
• Keywords: correlation ; estimating equation ; latent variable ; over-dispersion ; random effect

Abstract

This research was motivated by the Naturalistic Teenage Driving Study (NTDS), which followed 42 newly licensed teenage drivers continuously during their first 18 months of independent driving. We consider fitting a mixed effects Poisson regression model to counts of kinematic events. Because of the size of the NTDS dataset (thousands of trips for most drivers), standard methods such as maximum likelihood can be difficult to carry out. We propose a marginal approach based on estimating equations that relate model parameters to certain marginal quantities. Implementation of this approach involves several considerations with implications on the performance of estimators and the computational requirement. In addition, a within-cluster resampling technique can be used to deal with informative cluster sizes. The proposed approach is evaluated in simulation experiments and applied to the NTDS.