Department of Statistics Seminar
North Carolina State University

presents

Dr. Raymond J. Carroll

Texas A & M University

"Longitudinal and Clustered Data and Non/Semiparametric Regression"

ABSTRACT

I will review the problem of nonparametric and semiparametric regression for longitudinal/clustered data. In the independent data case, kernel methods and spline methods are essentially asymptotically equivalent for these problems. The same is not true for the longitudinal/clustered data case, with standard kernel methods failing to account for correlation in any sort of effective form. Indeed, such kernel methods and spline methods have different local influence functions ("effective kernels"). I will discuss the search for kernel--type methods that can take advantage of the correlation inherent in such problems, and will show that two such methods exist, both based on versions of pseudo-observations. I will then indicate how the methods can be extended to the time series problem and to the partially linear model for clustered/longitudinal data.

Friday, April, 2, 2004

3:35 - 4:35 pm

206 Cox Hall

Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.