Using Statistical Matching to Account for Coverage Bias When Combining Probability and Nonprobability Samples

• Activity Number: 210 - Contributed Poster Presentations: Survey Research Methods Section
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Survey Research Methods Section
• Author(s): Edward Mulrow* and Nada Ganesh and Vicki Pineau and Michael Yang
• Companies: NORC at the University of Chicago and NORC at the University of Chicago and NORC at the University of Chicago and NORC
• Keywords: nonprobability sample; statistical matching

Abstract

Many methods have been developed to combine probability and nonprobability samples via quasi-randomization, superpopulation modeling, or doubly robust estimation (Valliant, 2020). Yang, et al (2018) observed that when using statistical matching (a quasi-randomization approach) there may be a proportion of probability sample units that do not match to nonprobability sample units. Given this observation, a reasonable conjecture is that this unmatched portion of the probability sample provides a means to assess the coverage bias of the nonprobability sample. Ma and Mulrow (2019) developed an approach that used statistical matching to produce estimates from combined probability and nonprobability samples, and observed its behavior via a case study. We explore this approach further using the simulation approach in Yang, et al (2019) to assess the bias reduction and confidence interval coverage of the matching approach compared to other methods.