Combining methods to forecast the 2004 presidential election: The pollyvote, Working paper

We present an evaluation of a project to forecast the 2004 presidential election by applying the combination principle, a procedure which in other contexts has been shown to reduce error. This involved averaging within and across four categories of methods (polls, Iowa Electronic Markets quotes, quantitative models, and a Delphi survey of experts on American politics) to compute a combined forecast of the incumbent’s share of the two-party vote. We called it the Pollyvote, signifying “many (methods). ” Both approaches reduced error. With the Pollyvote, the mean absolute error was reduced by one third relative to the next most accurate method, the Iowa Electronic Markets, when tested across the 163 days preceding the election. Gains were achieved at all forecast horizons that we tested. On the morning of November 2, the Pollyvote had Bush winning 51.5 percent of the two-party vote, which came within 0.2 percent of the outcome (51.3%)