This week, The Economist launched its first presidential election prediction model called “The Economist U.S. presidential forecast.”
The model, which as of now predicts Joe Biden has an 83% change of winning the presidency in November, was created by The Economist’s data team in collaboration with Columbia University political scientists Andrew Gelman and Merlin Heidemanns. It uses machine-learning techniques to promote accuracy.
Using factors such as economic conditions, presidential popularity and the amount of time one party has been in power, alongside polling data at the national and state levels, the model produces probabilities for each state and the overall election.
Through a single “best guess,” Biden was predicted to win 53.5% of the popular vote (excluding third parties) and 329 electoral votes. Uncertainties surrounding estimates are also calculated.
According to The Economist, the model’s use of machine-learning allows it to maximize accuracy by emphasizing fundamentals and downplaying the importance of polls until the election looms nearer in the fall.
As polls are taken into greater consideration, the model doesn’t consider their margins of error, but instead corrects for biases, like the tendency of one party to answer more surveys than the other.
So far, the model has produced charts based on best guesses from each day since the beginning of March. Visitors will also find a map that breaks down each state by estimated vote share and win probabilities.
U.S. editor for The Economist, John Prideaux, stated: “The Economist aims to provide the most rigorous analysis and reporting on the 2020 US election. Our U.S. presidential forecast takes this a step further by providing a projection grounded in meticulous statistical analysis and modelling.”