The objective of this research is to provide information regarding the small sample properties of several recently proposed methodologies for estimating marginal model mean parameters. Focus will be on the generalized estimating equation methodology (Liang and Zeger, 1986, Zhao and Prentice, 1990, Prentice and Zhao, 1991) and a likelihood approach to estimation in this setting (Fitzmaurice and Laird, 1993). These models are applicable when a sample of clusters of correlated observations is collected, as in a longitudinal study. The critical need for such methods is when the correlated response measures are discrete or cannot be modelled with a multivariate normal distribution, and there are both time- varying and time-stationary continuous covariates. The impact of missing data on the small sample properties will also be assessed. Large-scale computer simulations will be used to address these questions, with attention paid to computational concerns and required resources. The interactive activities include: teaching a graduate-level course to statistics students; participating in a graduate-level course on longitudinal data; organizing a series of mini-courses on topics related to this research; seminars; informal groups; and visiting local area statistical organizations with students.
