Range-Restricted Benchmarking of Pseudo-Hierarchical Bayes Small Area Estimates

• Activity Number: 393
• Type: Invited
• Date/Time: Thursday, August 15, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods*
• Author(s): Ralph Folsom*+ and Akhil Vaish and Avinash Singh
• Affiliation(s): RTI International and RTI International and RTI International
• Keywords: Benchmarking; Range restrictions; Survey weighted HB

Abstract

It is often desirable to benchmark small area estimates so that they are consistent with direct aggregate level estimates, which are often published before the small area estimates. When the aggregate sample is large, Folsom, Shah, and Vaish's (1999) survey, weighted hierarchical Bayes (SWHB) solution for the logistic mixed model, is approximately benchmarked. To achieve exact benchmarking with moderate sized aggregate samples, Singh and Folsom (2001) extended Ghosh's (1992) constrained Bayes theory, noting that benchmarks could be population-weighted moments of the small area estimators, and that the constrained estimators could be subject to range restrictions. For example, the range of benchmarked small area prevalence estimates must be restricted to the [0,1] interval. In this paper, we apply the Singh and Folsom range-restricted benchmarking constraints to county level small area prevalence estimates derived from the pooled 1999 and 2000 BRFSS data for North Carolina. Unconstrained small area estimates for past month cigarette and binge alcohol use are obtained using the SWHB solution. These estimates are then contrasted with the Singh and Folsom exact benchmark results.