Although most pundits rely on polls to predict who will be our next President, the Presidential election is not a popularity contest--it's a partition problem contest. That is, the goal is to get a majority of electoral votes, and each state gives all of its votes to one of the two candidates (with a couple of exceptions). Most states are red or blue, but there are a few swing states, and who gets elected depends on which way these states swing.
A nifty OR model by Sheldon Jacobson (from the University of Illinois), Steven Rigdon and Ed Sewell (both from Southern Illinois University Edwardsville) take advantage of the winner-take-all state elections to predict who will end up with the majority of electoral college votes. They use state polling information in a dynamic programming algorithm to determine the probability distribution for the electoral college votes using Bayesian estimators. As of now, the predict that Barack Obama will win the election with a probability of 87%.
