Nonparametric and semiparametric models impose less restrictive assumptions than parametric models and provide a flexible approach for analyzing longitudinal data. Developing good, efficient estimation methods for these models has attracted substantial recent interest. We show in this talk that polynomial splines provide a straightforward, unifying solution for many such models.
The idea of the spline method is very simple: one approximates the unknown function involved in the model by a polynomial spline and thus operationally changes the problem to a parametric one. As a consequence, methods for parametric models naturally suggest what one should do in a semiparametric or nonparametric setting, such as if one needs to and how to use the with-in subject correlation in estimation.
In this talk I will give an expository overview of the spline method. In particular, I will explain in some details the application of the spline method in estimation of a time-varying coefficient model, a partly linear model, and in nonparametric estimation of a large covariance matrix. The talk will focus on methodological aspects but theoretical basis of the methods will also be mentioned when appropriate.
Meet the speaker in Room 212 Cockins Hall at 4:30 p.m. Refreshments will be served.