Multilevel Regression and Post-Stratification with Misreporting and Selection Bias

• Activity Number: 548 - Total Survey Errors in the Combination of Probability and Nonprobability Samples
• Type: Invited
• Date/Time: Wednesday, July 31, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Douglas Rivers*
• Companies: Stanford University
• Keywords: Survey Sampling; Nonresponse; Selection bias; Multilevel regression; Post-stratification

Abstract

One of the most challenging aspects of election polling is the modeling of turnout. Self-reported turnout exceeds actual turnout in both pre- and post-election surveys, but it is unclear how much is due to selection bias (enthusiastic voters are more likely to participate in election surveys) and misreporting (due to social desirability bias and other factors). Multilevel Regression and Post-stratification (MRP) has been successfully used to correct for selection bias when nonresponse is ignorable in the sense of involving selection on observable variables only. However, non ignorable selection and misreporting are not handled by conventional MRP models. This paper proposes sensitivity analyses that can be performed with MRP models to indicate the direction and size of biases in MRP estimates. These techniques are applied to data from polling in the 2018 U.S. midterm elections.