Election Polling in 2020, 2022, and 2024: Lessons Learned to Improve Accuracy

• In Event: Campaigns and Elections

Abstract

Election polls have had a mixed track record of success over the past several presidential and midterm election cycles. Declines in response rates, accurately identifying likely voters, and different data weighting strategies all highlight the challenges that face modern-day pollsters. In this paper, I will test several different approaches to polling to assess which techniques proved to be most accurate in 2020, 2022, and 2024. For example, how accurate is data derived from online, opt-in panels as compared to data derived from a probability sample? What weighting formulas performed best in 2020, 2022, and 2024? What screens worked best to identify likely voters? The data for this project will come from the East Carolina University (ECU) Center for Survey Research (CSR), which has collected election polling data since 2020 in the states of North Carolina, Georgia, and South Carolina. Using the data over these three election cycles from these three states, my paper will assess the accuracy of online, opt-in data (as compared to data from a probability sample), and test various weighting techniques and likely voter screens to determine what approaches to election polling perform best. The results of the paper should yield important insights to benefit the field of election polling and forecasting.

Author

• Peter Francia, East Carolina University