|
Activity Number:
|
399
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #308388 |
|
Title:
|
Model Selection: Multiplicities and Approximations
|
|
Author(s):
|
James Berger*+
|
|
Companies:
|
Duke University
|
|
Address:
|
Inst of Statistics and Decision Science, Durham, NC, 27708-0251,
|
|
Keywords:
|
BIC ; multiplicity ; reproducability ; multiple testing ; asympotics ; marginal likelihood
|
|
Abstract:
|
Two of the biggest hurdles in dealing with model uncertainty are dealing with multiplicity and computation. Issues of multiplicity in testing are increasingly being encountered in practice, and failure to properly adjust for multiplicities is being blamed for the apparently increasing lack of reproducibility in science. The first part of the talk will discuss different types of multiplicities that are being encountered and methods for handling them. A major computational hurdle for Bayesians is computation of model likelihoods. Because of the difficulty of this computation, BIC is often used as an approximation. Unfortunately, BIC has a number of problems. The second part of the talk will cover new approximations that show considerable promise in significantly improving on BIC, even potentially applying to situations where the model size grows with the sample size.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2007 program |
