Activity Number:
|
419
- Bayesian Computation and Spatial Modeling
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, July 31, 2018 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Section on Bayesian Statistical Science
|
Abstract #329216
|
|
Title:
|
Bayesian Small Area Estimation of Multinomial Outcomes
|
Author(s):
|
David R Judkins* and Stas Kolenikov and Raphael Nishimura
|
Companies:
|
Abt Associates and Abt Associates and Abt Associates
|
Keywords:
|
Multinomial Poisson;
MCMCglmm;
SAE
|
Abstract:
|
Too date, there have only been a few papers on SAE for multinomial outcomes. We were motivated to tackle this problem by a desire to produce state-level estimates of jail inmate profiles from a survey that was only designed to produce national estimates. There are periodic censuses of jails that focus mostly on staffing and less frequent surveys that focus on inmate experiences. An example of the type of statistic required is the 4-tupple of primary reason for being held in jail, where reasons A through D are mutually exclusive. We applied the MCMCglmm R package with some surround-coding to tackle this problem. In this paper, we report on the performance of the software on simulated data where the truth was known. We used proper priors on all parameters and worked to overcome some problems where the MSE of posterior draws was actually greater than the variances of direct design-based estimators in many states.
|
Authors who are presenting talks have a * after their name.