Small-Domain Estimation with Limitations on the Direct Estimate

• Activity Number: 436
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Wesley Basel*+ and Jasen A Taciak
• Companies: US Census Bureau and US Census Bureau
• Keywords: small area estimation ; poverty ; SAIPE ; Amercian Community Survey

Abstract

The Small Area Income and Poverty Estimates program of the U.S. Census Bureau currently produces poverty estimates for school districts(SD) using a tiered approach. A Fay-Herriott approach is used to produce poverty count estimates at the county level, which are then allocated to the SD level using both survey and administrative records data. Preliminary results indicated that a direct SD model could produce more precise estimates, but an improvement in the handling of zero-poverty estimates in the survey data is required. For the current county model, less than 10% of the single-year direct estimates from the American Community Survey result in zero realizations. But at the SD level, areas and associated samples are much smaller, and proportions of zero realizations are higher. This paper constructs and evaluates a basic Empirical Bayes model of county poverty, where the linking function uses an inverse logit transformation to limit the support of the random effect to the (0,1) interval. Evaluation against the current log-level model and alternative models will be included. This basic model should serve as a benchmark for further specification research.