|
Activity Number:
|
514
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 7, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #302127 |
|
Title:
|
Improved Confidence Regions for Seemingly Unrelated Regression Models
|
|
Author(s):
|
Kent Riggs*+
|
|
Companies:
|
Stephen F. Austin State University
|
|
Address:
|
Department Of Mathematics and Statistics, Nacogdoches, TX, 75962-3040,
|
|
Keywords:
|
Seemingly Unrelated Regression ; Variance-Covariance Reduction ; Bartlett-type correction
|
|
Abstract:
|
We employ maximum likelihood (ML) methods under the seemingly unrelated regression (SUR) model to derive five new confidence ellipsoids for covariate coefficient vectors of interest. The new confidence ellipsoids include a Bartlett-type corrected percentile on a Wald statistic and a Wald statistic with the confidence coefficient determined by a parametric bootstrap. Lastly, we study the behavior of our new confidence ellipsoids and the ordinary least squares (OLS) ellipsoid in a Monte Carlo simulation that demonstrates the improvement of several new confidence ellipsoids over the OLS method. For the configurations studied in our simulation, we determine that the Wald statistic with a Bartlett-corrected confidence coefficient is preferred in that it has close to nominal coverage and relatively small volume. Finally, we apply two of the new confidence ellipsoids to a real data set.
|
