Which method provides the most accurate election forecasts?

Polly can show you which forecasting method is best by using the PollyGraph.

The PollyGraph shows the mean absolute error across the six elections from 1992 to 2012, across the remaining days to election. That is, at each day, the PollyGraph shows the error that you would have received by relying on that component method’s (already combined) forecast until Election Day.

Assume you pick a method 100 days before the election, and you decide to stick with it. If you had picked a polling average, you would have received an average error 2.8 percentage points. In comparison, had you trusted the PollyVote, you would have received an error of only 1.0 percentage point. In other words, the only 1.0 percentage point. In other words, the PollyVote reduced the error of the combined polling average by 64%.

Interestingly, polling averages are among the least accurate methods available, particularly for longer-term forecasts. Among the most accurate individual forecasting methods are citizen forecasts, which yielded an average error of 1.2 percentage points. While this is a very low error, note that it is still 20% higher than the error of the PollyVote.

Note that forecasts from expert judgment and index models were not available for all years. Therefore, these numbers are not fully comparable. Again, note that the research findings suggest that the PollyVote’s performance would have been even stronger if forecasts from these methods had been available also for historical elections.