Estimators with Combined Probability and Nonprobability Samples Using Small Area Models

• Activity Number: 126 - Interpreting Nonprobability Samples: Discoveries and Challenges
• Type: Topic-Contributed
• Date/Time: Monday, August 9, 2021 : 1:30 PM to 3:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Michael Yang* and Nada Ganesh and Evan Herring-Nathan and Vicki Pineau
• Companies: NORC at the University of Chicago and NORC at the University of Chicago and NORC at the University of Chicago and NORC at the University of Chicago
• Keywords: nonprobability sample estimation; statistical matching; propensity weighting; small area estimation; doubly robust

Abstract

Through case studies and simulations, the authors have evaluated a range of estimation methods for combining probability and nonprobability samples. Our earlier evaluations show that the Small Area Modeling method, a doubly robust method developed at NORC, tends to achieve greater bias reduction than the other methods, especially for the modeled response variables with large known biases associated with the nonprobability sample. Meanwhile, Statistical Matching and Propensity Weighting also demonstrate good properties. This paper expands our earlier studies to explore hybrid methods that integrate Statistical Matching, Propensity Weighting, and Small Area Modeling. Specifically, it reports comparative evaluations of four methods: (1) Matching Imputation Weighting without Small Area Modeling, (2) Matching Propensity Weighting without Small Area Modeling, (3) Matching Imputation Weighting with Small Area Modeling, and (4) Matching Propensity Weighting with Small Area Modeling. Evaluations of these methods are based on estimates of bias, confidence interval coverage, and effective sample size.