|
Activity Number:
|
403
- Recent Advances in Small Area Estimation
|
|
Type:
|
Topic-Contributed
|
|
Date/Time:
|
Thursday, August 12, 2021 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Survey Research Methods Section
|
|
Abstract #317050
|
|
|
Title:
|
Bayesian Hierarchical Spatial Models for Small Area Estimation
|
|
Author(s):
|
Gauri Sankar Datta* and Hee Cheol Chung
|
|
Companies:
|
US Census Bureau/University of Georgia and Texas A&M University
|
|
Keywords:
|
Conditional autoregression;
Intrinsic autoregression;
Simultaneous autoregression;
Fay-Herriot model;
Unsampled small areas;
Current Population Survey
|
|
Abstract:
|
The Fay-Herriot model is an important model to produce reliable small area statistics. It treats model errors or random effects as independent random variables with a common normal distribution. Lack of good covariates to estimate means of geographically contiguous areas often introduces spatial dependence among the random effects. We consider various popular spatial models as alternatives to the Fay-Herriot model. We consider Bayesian estimation using these models based on a class of noninformative improper priors. We assess effectiveness of the spatial models based on simulation study and an application. Our study shows considerably superior performances of the spatial models over the regular Fay-Herriot model in absence of good covariates. Often some small areas have no samples and no direct estimates. Spatial models generate predictions of unsampled area means by borrowing from neighboring residuals better than the synthetic regression means that result from the independent Fay-Herriot model. For all the Bayesian models we considered, even in the absence of data from some areas, posterior distributions based on improper priors are shown to be proper under some mild conditions.
|
Authors who are presenting talks have a * after their name.
