Nonparametric Regression Methods for Modeling Longitudinal Data
                    Hulin Wu

 Frontier Science & Technology Research Foundation 1244 Boylston
 Street, Suite 303, Chestnut Hill, MA 02467
 Email: wu@sdac.harvard.edu

Longitudinal data such as repeated measurements taken on each of a
number of individuals arise frequently in many biomedical and
clinical studies. For such data, mixed-effects models provide a
useful and flexible likelihood framework in which population
characteristics are modeled as fixed effects while between-subject
variations are modeled as random effects. Linear and nonlinear
mixed-effects models have been widely used in many applications.
However, if parametric models are not available for the
longitudinal data analysis, we may need to resort to nonparametric
regression techniques. In this talk, I will introduce
mixed-effects modeling ideas in nonparametric regression methods,
in particular, for local polynomial regression and regression
splines methods, in order to build an efficient inference model
for longitudinal data. The theoretical properties of the
estimators will be discussed. Both continuous and discrete
longitudinal data from AIDS clinical trials will be presented to
motivate the methodology development and illustrate the
application of the proposed methods.