|
Activity Number:
|
254
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #305685 |
|
Title:
|
Testing Lack-of-Fit of Heteroscedastic Nonlinear Regression Models with Local Linear Smoothers
|
|
Author(s):
|
Chin-Shang Li*+
|
|
Companies:
|
St. Jude Children's Research Hospital
|
|
Address:
|
332 N. Lauderdale Street, Memphis, TN, 38105,
|
|
Keywords:
|
bandwidth selection ; boundary effects ; fit comparison ; local linear kernel ; quasi-likelihood estimator
|
|
Abstract:
|
A data-driven test is proposed for assessing the appropriateness of heteroscedastic nonlinear regression models by using local linear regression smoothers in which no boundary-corrected kernels are needed to resolve boundary effects. The bandwidth is selected based on the asymptotically optimal bandwidth under the parametric null model. This selection method leads to the data-driven test. It is shown that the test statistic is normally distributed under the null hypothesis and the test is consistent against any fixed alternative. The resulting test can be applied for the lack-of-fit of a postulated generalized linear model and is compared to existing tests. A real-life dataset is used to demonstrate the practical use of the proposed test.
|
