2020 US Election Forecast
Forecasting the 2020 Electoral College Winner: The State Presidential Approval/State Economy Model
Peter Enns and I predicted 49/50 states + D.C. correctly. Our only miss was Georgia.
The model was published in PS. We discuss our results and the potential implication of presidential approval being disconnected from reality in a Washington Post Monkey Cage article. We also published an article in the University of Miami Law Review in which we discuss how legal and social science researchers can use early forecasting to estimate the effects of campaigns, media, and laws.
I talked about our model and the American Presidential Election 2020 to the German media outlet “Marktforschung.de” (in German) and Argentinian newspaper “Infobae” (in Spanish). Our work was also cited in “Le Figaro” (in French).
Our model focuses on fundamentals and presidential approval. Instead of predicting the national vote share we predict the outcome of the Electoral College. To receive State-level presidential approval rates we used MRP models using 86 polls and almost 90,000 respondents between 1980-2020 to estimate the State-level presidential approval rate in each state. Additionally, we use economic indicators on state-level as well as some political variables like home-state of the presidential candidates. 104 days before the election, our model predicted about a 4 in 10 chance that Donald Trump would be re-elected and about a 6 in 10 chance that Joe Biden is the next president. The figure below shows the distribution of the 70,000 simulated Electoral College outcomes. The modal value was a victory of Trump by a margin of one vote. However, in ~60% of the simulations, we predicted that Joe Biden would win more than 50% of the Electoral College votes.
The graph below shows the distribution of our simulations on state-level. Higher histograms indicate greater importance for the Electoral College. The only state we missed was Georgia.