May 20, 2025
Research Highlight: Public Polling
Researchers in the Program in Survey and Data Science have been doing work to improve public polling accuracy.
“Tried and true” methods of weighting adjustment may not be sufficient for adjusting for the kinds of selection bias that arise due to problems like partisan nonresponse to political polls (i.e., where the probability of responding to a survey is a function of the unique thing that you are trying to measure in that survey).
In a recent article in Public Opinion Quarterly, Brady West and Rebecca Andridge (The Ohio State) demonstrate how new statistical measures of this type of “non-ignorable” selection bias can be used to effectively adjust estimates and correct the bias in these estimates that standard weighting adjustments cannot. They published software which researchers can use to implement these measures and adjustments.
