|
Activity Number:
|
506
- New Ideas in Inference
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 6, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Business and Economic Statistics Section
|
|
Abstract #310967
|
|
|
Title:
|
Whither Wald? an Alternative to the Delta-Wald Test
|
|
Author(s):
|
Joe Hirschberg* and Jenny Lye
|
|
Companies:
|
University of Melbourne and University of Melbourne
|
|
Keywords:
|
likelihood profile;
Fieller method;
confidence curves
|
|
Abstract:
|
The Wald test for nonlinear restrictions of parameter estimates from regression models is widely employed due to the ease of application to the results of the unrestricted model and the wide availability of computer routines to apply the delta method for the approximation of the standard error. However, it has been shown that in some cases the specification of equivalent restrictions can lead to highly varying results which has resulted in the admonition in many textbooks that these test results should be used with care or not used at all. This paper proposes an alternative approach that employs a constrained optimization approach to the formation of a probability model that can be used to estimate a p-value that will not be subject to variations caused by differing specifications. This approach can be programmed in such a manner as to rely solely on the results of the unrestricted model estimation in the same manner as the existing delta method software.
|
Authors who are presenting talks have a * after their name.
