|
Activity Number:
|
486
|
|
Type:
|
Invited
|
|
Date/Time:
|
Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #307960 |
|
Title:
|
Bayesian Nonparametric Modeling for Well-Calibrated Location and Scale Inference with Skewed and Heavy-Tailed Data
|
|
Author(s):
|
David Draper*+
|
|
Companies:
|
University of California, Santa Cruz
|
|
Address:
|
Department of Applied Mathematics and Statistics, Santa Cruz, CA, 95064,
|
|
Keywords:
|
bootstrap ; Dirichlet-process ; mixture modeling
|
|
Abstract:
|
The bootstrap is a popular frequentist nonparametric technique for creating interval estimates, which claims to produce well-calibrated intervals no matter what the underlying distribution F is. However, since it's based solely on the empirical cumulative distribution function, it has no information about F beyond the largest data value Y(n). When n is moderate and F is heavy-tailed and/or heavily skewed, this can ignore much of the "weight" of the underlying distribution, leading to (extremely) poor calibration for location and scale functionals. In this talk I will describe the use of Bayesian nonparametric (Dirichlet process mixture) modeling to produce well-calibrated location and scale intervals, even when n is quite small and the (unknown) data-generating F is quite skewed and/or heavy-tailed.
|
- 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 |
