Benchmarking Small-Area Estimates: A Minimum Discrimination Information Approach and Other New Perspectives

• Activity Number: 305
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Andrew Vesper*+ and Ryan Janicki
• Companies: Harvard University and U.S. Census Bureau
• Keywords: small area estimation ; benchmarking ; entropy ; Bayesian ; simulation study

Abstract

In sample surveys, often there is insufficient sample size to obtain reliable direct estimates for certain domains. Precision can be increased by introducing small area models which borrow strength by connecting areas and incorporating auxiliary covariate information. For small area models, estimates at a lower geographical level typically will not aggregate to the estimate at the corresponding higher geographical level. Benchmarking is a statistical procedure which is used to adjust model-based estimates to satisfy constraint requirements. This paper briefly outlines two new approaches for constructing benchmarked estimates: minimum discrimination information (MDI) and a fully Bayesian model conditional on the constraint. Simulations assess the performance of several benchmarking procedures for Fay-Herriot models, offering new insight into benchmarking in practice: the proposed methods benchmark both first and second moments (where competitors often do not), procedures equivalent to the MDI estimator outperform competitors, and estimates at higher levels of aggregation should not necessarily be held fixed when benchmarking.