|
Activity Number:
|
139
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #308195 |
|
Title:
|
A New Bayesian Procedure for Nonparametric Regression Estimation and Pointwise Confidence Bands
|
|
Author(s):
|
Linda Zhao*+
|
|
Companies:
|
University of Pennsylvania
|
|
Address:
|
3730 Walnut ST, Philadelphia, 19104,
|
|
Keywords:
|
|
|
Abstract:
|
We construct approximate Bayesian confidence intervals for the values, f(x), of a nonparametric regression function. The method uses a blockwise construction which reduces the problem to that of confidence intervals for single coordinates of a multivariate normal mean. The prior adopted for the multivariate mean problem is a modification of the Harmonic prior. In this multivariate problem the Bayesian estimates perform well although the Bayesian regions for single coordinates tend to overcover in a crucial part of the space. A frequentist modification is then suggested in the multivariate problem and the results are carried over to the original regression problem. The resulting nonparametric regression confidence bands have outstanding frequentist coverage properties over a wide range of empirical examples.
|
