Building-Block BLUPs for Aggregate-Level Small-Area Estimation for Survey Data

• Activity Number: 86
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Avinash C. Singh*+ and Pin Yuan
• Companies: NORC and Human Resource and Social Development Canada
• Keywords: Building-blocks ; Aggregate level ; Benchmarking ; BLUP at Target level

Abstract

Often the choice of the level of aggregation in small area (SA) modeling is not governed by adequacy of modeling assumptions but by user needs. This may have serious impact on validity of the underlying exchangeability of area-specific random effects in a LMM for means. The problem doesn't arise with unit level models which, however, are difficult for taking design into account. As a compromise, we propose a building-block (B) model such that random effects at B-level subareas are likely to be exchangeable, and model parameters are estimated at this level or after grouping of B-level areas to avoid the problem of zero sample sizes or zero direct estimates. Target SA estimates are obtained as BLUPs using a reduced model with target SA-level direct estimates. An illustrative example is presented using Canadian Labour Force Survey data with provincial economic regions as target small areas.