Citizen Forecasting 2020: A State-by-State Experiment 

Citizen Forecasting 2020: A State-by-State Experiment 

by Andreas E. Murr and Michael S. Lewis-Beck 

We use ‘citizen forecasts’ to predict the US Presidential Election taking place on 3 November 2020.  This approach asks citizens to forecast which presidential candidate will win in their state and the nation as a whole, and predicts the winning candidate to be the one which most citizens say will win. 

Previous studies have shown that ‘citizen forecasts’ predict better than any other approach in Great Britain and the United States.  But the timing of the data collection forced most of the studies using citizen forecasts to forecast elections after they occurred.  To date, only two ex ante citizen forecasting papers to have appeared before a national election.  Both of them forecasted British General Elections.  We replicated this forecasting approach in the upcoming US Presidential Election.  We asked workers on Amazon.com’s Mechanical Turk in July to predict which candidate will carry a state and which candidate will win the Presidency.  Our citizen forecasting model predicts Trump to win both 50.4% of the popular two-party vote and 357 of the electoral votes.  Given the long lead time and the unusual election context, this research provides a challenging test for the accuracy of ‘citizen forecasting’ in the United States.


The editorial team at PS: Political Science & Politics is pleased to release of the 2020 Presidential Election Forecasting symposium. This year, due to the amazing work done by Ruth Dassonneville and Charles Tien, we have 12 forecasts and three critical reviews. We hope teachers, journalists, and the general public will read, reflect, and comment.  For the latest updates, follow PS on Twitter.

Be the first to comment

Leave a Reply

Your email address will not be published.


*