Trial-Heat Forecasts of the Presidential Vote

Abstract

This research examines Gallup poll trial-heat forecasts of the two-party presidential popular vote for the incumbent presidential party. First, several existing forecasting equations are updated and evaluated. Trial-heat results at six points throughout campaigns from 1948 to 1988 are then examined. These trial-heats are used in several ways to produce presidential vote forecasts: (a) in raw form as direct forecasts, (b) alone in regression estimated forecasts, and (c) in conjunction with economic growth in regression estimated forecasts. As Lewis-Beck and Rice found in 1985, the earliest and most accurate trial-heat forecasts are those using early September trial-heats and second-quarter real growth in the gross national product. These forecasts are also more accurate than forecasts based on previous models. The early September trial-heat/economy forecast equation has an average "within-sample" error of only ±1 percentage point (adjusted R 2 = .94, SEE = 1.5) and a mean "out-of-sample" error of ±1.1 percentage points. The early September trial-heat/economy equation correctly "predicted" the winning presidential candidate in ten of the eleven elections from 1948 to 1988, missing only in the near dead heat of 1960.