From Polling to Prediction Markets: How Reliable Can We Predict Elections?
This latest episode of the Harvard Data Science Review Podcast takes a closer look at election polling: how polls are conducted, what they can tell us about voters, and why measuring public opinion has become increasingly challenging. In advance of the Harvard Political Analytics 2026 conference, hosts Liberty Vittert Capito and Xiao-Li Meng are joined by two conference speakers: Anthony Salvanto, CBS News’s director of elections and surveys, and Michael A. Bailey, author of Polling at a Crossroads: Rethinking Modern Survey Research. From declining response rates and changing turnout patterns to weighting, online panels, and election-night projections, Salvanto and Bailey explore what goes on behind the numbers—and why modern pollsters are increasingly also modelers. They discuss the assumptions and human judgment embedded in polling, the importance of data quality, and the challenge of measuring and communicating uncertainty when traditional measures such as the margin of error tell only part of the story. The conversation also considers the growing role of prediction markets and what they can—and cannot—add to our understanding of elections and public opinion. Ultimately, the episode asks a larger question: In a rapidly changing polling landscape, how can researchers and the public better understand what the numbers really mean? Our guests: Michael A. Bailey is the Colonel William J. Walsh Professor of American Government in the Department of Government and McCourt School of Public Policy at Georgetown University. His research focuses on applying statistical techniques to answering questions at the intersection of political science, policy, law, and economics. Anthony Salvanto is the executive director of elections and surveys at CBS News. He oversees the CBS News Poll and all surveys across topics and heads the CBS News Decision Desk that estimates outcomes on election nights.