Weighting and Analyzing Nonprobability Samples for Population-Based Inferences
Virtual | May 7, 2024 | 10:00 am – 1:30 pm ET
Instructors: Lingxiao Wang, University of Virginia and Yan Li, University of Maryland
Making valid population-level inference is a central goal in survey research. While studies with probability sampling are the gold standard to conduct design-based inferences about the target finite population, they are facing substantial challenges such as high costs and reduced response rates in recent decades. As a remedy, nonprobability samples have been increasingly collected in many areas including education, medical studies, and public opinion research. Nevertheless, nonprobability samples cannot well represent the target population due to non-random sampling. Consequently, the naïve nonprobability estimates can be biased from the target population quantities.
Quasi-randomization methods are among the most used approaches to improve representativeness of nonprobability samples. These methods create “pseudoweights” for nonprobability sample individuals using contemporaneous probability surveys as references, which substantially reduce selection bias in estimating target population quantities.
This course will first provide a comprehensive review of the framework for making finite population inferences from nonprobability samples, covering various pseudoweighting methods published in recent literature. Then the attendees will be guided through the specific steps of constructing pseudoweights. To ensure practical application, the course will include software and real-data examples, illustrating how to construct pseudoweights and analyze nonprobability samples, for estimating finite population means and associations.