Variable Selection Strategies for Effective Quota Sampling and Propensity Weighting: An Application to SARS-Cov-2 Infection Prevalence Estimation

• Activity Number: 310 - Making Finite Population Inferences from Nonprobability Samples
• Type: Topic-Contributed
• Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Yan Li* and Barry Graubard and Michael Fay and Sally Hunsberger
• Companies: University of Maryland, College Park and National Cancer Institute, DCEG, Biostatistics Branch and Biostatistics Research Branch, Division of Clinical Research, NIAID and NIAID
• Keywords: balanced distribution; propensity score; quota variable

Abstract

To inform policy makers with timely results, quota sampling, as a time- and cost-efficient method, is often conducted to collect data. To control for selection bias of a quota sample, all variables that predict the outcome ideally are used to form the quota sampling strata so that their distributions match the census of the target population. In practice, however, census data may not have information on all predictive variables for quota sampling, whereas more predictive variables are available in a probability reference survey and collected in a quota sample for pseudoweight construction. This paper studies: 1)bias of simple sample mean using different quota variables; 2)bias reduction by propensity pseudoweighting that uses various predictive variables to model propensity. We propose variable selection strategies for pseudoweights construction combined with quota sampling to improve the accuracy and efficiency for the population mean estimation. The adjusted quota sample estimators by the proposed strategy are evaluated analytically and numerically. These methods are applied to estimate SARS-Cov-2 infection seroprevalence using a quota sample and the BRFSS surveys as a reference.