Validity Testing for Coverage Properties of Small-Area Models for Cell Proportions

• Activity Number: 37
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Aaron Gilary*+ and Jerry Maples and Eric Victor Slud
• Companies: U.S. Census Bureau and US Census Bureau and US Census Bureau
• Keywords: Arcsine square-root transformation ; Beta-binomial regression ; Confidence Bound ; Parametric Bootstrap ; Prediction Interval ; Transformed Fay-Herriot Model

Abstract

In Gilary, Maples, and Slud (2012), we compared three possible models for small proportions in small survey domains: Fay-Herriot (Fay and Herriot 1979), GLMM (Jiang and Lahiri 2006), and Beta-Binomial (Prentice 1986). The comparisons used bootstrap-based condence intervals which were justied by asymptotic theory or established only for large samples. Here we build upon that work by conducting simulations of small area data in moderate-sample settings, for two purposes: to evaluate the performance of dierent analysis methods when using each of the three simulation models; and to assess the validity of predictors and Wald-type condence interval coverage properties for each method.