A Bayesian Approach to Predicting the Electoral Vote Totals for the 2008 Presidential Election

• Activity Number: 479
• Type: Contributed
• Date/Time: Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Steven E. Rigdon*+ and Sheldon H. Jacobson and Edward C. Sewell and Wendy K.T. Cho and Christopher J. Rigdon
• Companies: Southern Illinois University Edwardsville and University of Illinois and Southern Illinois University Edwardsville and University of Illinois and Arizona State University
• Keywords: Dirichlet distribution ; Electoral college ; dynamic programming

Abstract

During the U.S. presidential race of 2008, there were hundreds of polls conducted in each state to determine which candidate is most likely to win the state. We conducted a Bayesian analysis of these polls to determine the probability of each candidate winning each state. Our model included four categories for each voter's intention: Obama, McCain, undecided, and all third-party candidates (lumped together). We then used these state-by-state probabilities, along with the dynamic programming algorithm of Kaplan and Barnett (2003), to determine the posterior distribution for the number of electoral votes, and from that, the probability of each candidate winning the election. Nearly every day during the two months prior to the election, we monitored the polls and recomputed the state-by-state probabilities and the posterior distribution for the number of electoral votes for each candidate.