Count responses arise frequently in longitudinal panel data situations. When the proportion of zero counts observed is greater than expected under the traditional Poisson regression model, the data are said to have excess zeros. The Hurdle model is commonly used to accommodate such data. In recent years attention has been paid to estimating the effects of time-dependent covariates within a longitudinal study. Both conditional and marginal estimation methods have been adjusted in an attempt to maximize the information allowed through time-dependent covariates. However, the effects of including time-dependent covariates in longitudinal models with excess-zero counts has received little attention.
It has been argued that the most efficient method for marginal parameter estimation in the presence of time-dependent covariates is a method based on the generalized method of moments (GMM). Currently, GMM has not been extended to the case of longitudinal data with excess zero counts. This talk will present such an extension of GMM to longitudinal excess-zero data. Some comparisons to existing methods are made, and an example analysis of self-reported Marijuana usage is discussed.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.