Quasi-BLUPs for Reducing Over-Shrinkage in Small-Area Estimation

• Activity Number: 174
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Avinash C. Singh*+ and Pin Yuan
• Companies: NORC at the The University of Chicago and Human Resources and Skills Development Canada
• Keywords: Alternatives to EBLUP ; Over-shrinkage ; Zero or Negative Estimated Variance Component

Abstract

In small area modeling, estimation of second order parameters (such as variance components or correlation coefficients in time series or spatial models) is often challenging because the estimates may turn out to be inadmissible or unreasonable. Estimated variance components may become very close to zero or negative (in which case it is truncated to zero or modified to get a positive estimate) due to model misspecifications or due to large sampling errors. The resulting SAEs tend to exhibit over-shrinkage to synthetic estimates and may be far from the direct estimator. We propose quasi-BLUP estimation in which suitably pre-specified values are used for variance components for computing SAE but the MSE estimates are adjusted for using working values which may not be consistent. Empirical results in the context of Canadian LFS show that such estimates have desirable properties.