Activity Number:
|
106
|
Type:
|
Contributed
|
Date/Time:
|
Monday, August 12, 2002 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Bayesian Stat. Sciences*
|
Abstract - #301385 |
Title:
|
An Approach to Capturing Data Information Contribution in Bayesian Data Analysis
|
Author(s):
|
Suman Bhattacharya*+ and Robert Weiss
|
Affiliation(s):
|
Johnson & Johnson Pharmaceutical R&D and University of California, Los Angeles
|
Address:
|
920 Route 202, P.O. Box 300, Raritan, New Jersey, 08869-0602, USA
|
Keywords:
|
Observe Fisher's information ; Leverage ; Generalized linear model ; Conjugate prior ; Non-informative prior
|
Abstract:
|
To evaluate case information contribution, we developed one interpretable case influence measure (CIM), which assesses the information contribution of each observation to the posterior in a Bayesian analysis framework. In order to have a better understanding of CIM, I shall present a decomposition of this measure. I shall show that the CIM measure is a generalization of the leverage that is a measure for detecting influential observations. Both algebraic and numerical examples will be presented to demonstrate how our measure works in linear and generalized linear models. Parallel comparisons are drawn between our measure and available Bayesian case diagnostics measures for all numerical examples.
|