Forecasting presidential elections: Accuracy of ANES voter intentions

Abstract

Despite research on the accuracy of polls as tools for forecasting presidential elections, we lack an assessment of how accurately the ANES, arguably the most used survey in political science, measures aggregate vote intention relative to the actual election results. Our ANES 1952–2020 results indicate that the reported vote from the post-election surveys accurately measures the actual vote (e.g., it is off by 2.23 percentage points, on average). Moreover, the intended vote measure from the pre-election surveys reasonably accurately predicts the actual aggregate popular vote outcome. While outliers may exist, they do not appear to come from variations in the survey mode, sample weights, time, political party, or turnout. We conclude that political scientists can confidently use the intended vote measure, keeping in mind that forecasting the popular vote may not always reveal the actual winner.