|
Activity Number:
|
106
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #301561 |
|
Title:
|
Benchmark Estimation--Motivation and Basics
|
|
Author(s):
|
Steven MacEachern*+ and Subharup Guha and Mario Peruggia
|
|
Companies:
|
Ohio State University and Ohio State University and Ohio State University
|
|
Address:
|
404 Cockins Hall, 1958 Neil Ave., Columbus, OH, 43210,
|
|
Keywords:
|
MCMC ; variance reduction
|
|
Abstract:
|
While studying various features of the posterior distribution of a vector-valued parameter using an MCMC sample, systematically subsampling the MCMC output can only lead to poorer estimation. evertheless, a 1-in-k subsample is often all that is retained in investigations where intensive computations are involved or where speed is essential. In these computationally intensive settings, we seek to create a discrete representation of the posterior distribution which is superior to the empirical distribution of a random sample of points drawn from the posterior. Benchmark estimation is one technique which yields this better representation. It relies on a number of estimates that are based on the best available information (the entire MCMC sample), and uses these to improve other estimates made on the basis of the subsample. The approach is implemented by creating weights for the subsample which are quickly obtained as the solution to a system of linear equations. Benchmark estimation leads to substantial gains at a negligible computational cost.
|
