|
Title
|
Room
|
|
New Developments on Model Selection
|
M-Consulate
|
|
Date / Time
|
Sponsor
|
Type
|
|
08/07/2001
2:00 PM
-
3:50 PM
|
Section on Statistical Computing*
|
Invited
|
|
Organizer:
|
Jianqing Fan, Hong Kong Chinese University
|
|
Chair:
|
Jianqing Fan, Hong Kong Chinese University
|
|
Discussant:
|
|
|
Floor Discussion
|
3:45 PM
|
Description
Most statistical problems are related to model selection. This is true for both parametric models or nonparametric methods. Traditional stepwise procedures ignore stochastic errors accumulated during the selection process. Hence, the confidence intervals constructed based on the final selected model may not have the right coverage. Further, computation can be quite expensive. Several techniques including penalized least-likelihood and Bayesian model selection have been proposed. This session presents some of the state-of-art of modern model selection techniques.
|
