|
Activity Number:
|
55
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, August 2, 2009 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #305302 |
|
Title:
|
The Prior Selection and Approximations for the Nested Error Regression Model: Estimation of Finite Population Mean for Small Areas
|
|
Author(s):
|
Santanu Pramanik*+
|
|
Companies:
|
NORC at the University of Chicago
|
|
Address:
|
, College Park, MD, 20740,
|
|
Keywords:
|
Laplace Approximation ; Hierarchical Bayesian ; Unit level model
|
|
Abstract:
|
In this paper we applied unit level linear mixed model to estimate the finite population means for small areas. As an inferential procedure we used Bayesian approach that needs specification of prior for the hyperparameters. Following some objective criteria, we propose a prior distribution for the ratio of variance components, along with a standard uniform prior on the regression coefficients. To approximate the posterior moments of small area means, we apply Laplace method. Our choice of prior avoids the extreme skewness, usually present in the posterior distribution of variance components. This property leads to more accurate Laplace approximation. Our simulation study shows that the resulting approximate Bayes estimators (with new prior) of small area means have good frequentist properties such as MSE and coverage rate.
|
