|
Activity Number:
|
382
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #308992 |
|
Title:
|
Model Selection for Multivariate Smoothing Splines with Correlated Random Errors
|
|
Author(s):
|
Eren Demirhan*+ and Hao Zhang
|
|
Companies:
|
North Carolina State University and North Carolina State University
|
|
Address:
|
1001 Avent Hill, APT. B7, Raleigh, NC, 27606,
|
|
Keywords:
|
COSSO ; SS-ANOVA ; Variable Selection ; Mixed Models ; Correlated Data
|
|
Abstract:
|
Model selection in nonparametric regression is a difficult problem, which becomes more challenging for correlated data such as longitudinal data and repeated measurements. In the framework of smoothing spline analysis of variance, we propose a unified approach to simultaneously selecting variables and estimating model parameters and covariance structures. The new method, as a generalization of the component selection and smoothing operator (Lin and Zhang 2006), imposes a soft-thresholding penalty on functional components for sparse estimation and take into account covariance structure at the same time. We propose an efficient algorithm in the framework of mixed effects which can be implemented by any standard software package. In particular, extensive work is done on the selection of tuning parameters. The performance of the new method is demonstrated through simulations and examples.
|
