Activity Number:
|
284
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
Sponsor:
|
Section on Survey Research Methods
|
Abstract - #305102 |
Title:
|
A Noninformative Bayesian Approach to Small-Area Estimation
|
Author(s):
|
Glen Meeden and Yanping Qu*+ and Bo Zhang
|
Companies:
|
The University of Minnesota and The University of Minnesota and The University of Minnesota
|
Address:
|
313 Ford Hall, Minneapolis, MN, 55455,
|
Keywords:
|
small area ; polya posterior ; noninformative Bayes ; polar importance sampling
|
Abstract:
|
We propose a new approach to small area estimation by borrowing strength through a Bayesian prior. This new prior leads to a generalized Polya posterior, which works as a constrained Polya posterior estimation when auxiliary information is available. The Bayes like character of the generalized Polya posterior determines the admissibility of our estimator. Our new approach needs no assumption of linearity. It utilizes both the area-specific and element-specific auxiliary information from each small area and outperforms classical regression methods. Variability of prior parameters allow more flexibility in estimating. This approach could be applied to cluster sampling as well. An approximation to the new estimator by using importance sampling with polar coordinates is proposed. It also provides a general tool to generate random samples in a constrained high-dimensional convex polytope.
|