The Referendum that Didn't Happen: The Forecasts of the 2000 Presidential Election

Extract

On August 29, 70 days before the election and more than 100 days before anyone would know who would be the next president, my trial-heat and economy model for forecasting presidential elections predicted that Al Gore would receive 52.8% of the national two-party popular vote. By the actual returns, after all the recounts and court rulings were completed, Gore received 50.3% of the national two-Party popular vote, 2.5 percentage points less than the model predicted. While this error is larger than normal (the mean out-of-sample error from 1948 to 1996 for the model is 1.5 percentage points), three of the last thirteen elections had larger out-of-sample errors. In short, though the performance of the forecasting model this year was well within normal bounds, the model was not quite as accurate as usual.