Remedies for Informative Sampling in Small-Area Estimation and Imputation

• Activity Number: 284
• Type: Invited
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Emily Berg*
• Companies: Iowa State University
• Keywords: Complex sampling ; Surveys ; Missing data ; Mixed model

Abstract

Small area prediction and imputation often involve explicit model assumptions. If the sample design is informative for the specified model, then predictors of finite population parameters resulting from maximum likelihood (un-weighted) estimators of model parameters can be biased. We study methods to construct unbiased estimators when the sample design is informative for the small area or imputation model. For the case of small area estimation, we show that enforcing design consistency for the finite population mean of a large area has little impact on the mean squared errors of the small area predictors. For the case of imputation, we develop procedures appropriate for a framework in which the missing at random assumption (MAR) is satisfied in the population (PMAR) but not in the sample (SMAR). The imputation methods are demonstrated with both parametric fractional imputation and quantile regression imputation.