Professor’s model perfectly predicted Trump victory

A Cornell professor's election forecasting model correctly picked Trump's win this year in all 50 states - and would have correctly predicted 95% of states in every election since 2000.

The model was developed by Peter Enns, professor of government in the College of Arts and Sciences and professor in the Cornell Jeb E. Brooks School of Public Policy, along with Julius Lagodny, Ph.D. '22, and was first deployed in the 2020 presidential election, where it correctly predicted the winner in every state except Georgia.

"Our forecast uses data from 100 or more days before the election, while most of the major forecasts update their data every day until the election," Enns said. "This would seem like it would give other forecasts an advantage, but this was not case. As the states were being called on election night, I noticed our forecast aligned with each outcome. By the end of election night, we had all of the states right. Five states were called later and we got those right too."

Enns' model focuses on economic conditions and president approval ratings as key factors (which are common in many prediction models), but the model extrapolates individual information from national polls back as far as 1980, using statistical modeling and additional information to develop numerous simulations and final forecasts on a state-by-state level.

Enns, Lagodny and colleagues Jonathan Colner and Anusha Kumar published a paper about the model in October in an election forecasting issue of Political Science and Politics, a journal of Cambridge University Press. Their model was the most accurate of the group, both in terms of the overall popular vote and correctly predicting that Trump would win 312 electoral votes (see page 7 of the PDF).

"If we think about this campaign, Trump did not follow conventional campaign strategy, and Harris did. Harris had more money, her campaign events generated excitement, and there were controversies at the last Trump rally in New York," Enns said. "This would have been the most likely election to not follow our forecast, but it still did."

Because of the model's success, the research team used historical data to attempt predictions on past presidential elections, back to 2000, and were accurate for 95% of states, Enns said.

The model could have multiple applications for campaign strategists and media outlets, Enns said - helping to predict which potential candidate would do better, where resources should be targeted in specific swing states and the types of issues resonating most with voters. Enns said it would also be possible to extend the model to specific congressional districts.

Enns is working on a volume chapter, with doctoral student Claudia Miner, looking at long-term shifts in the relationship between voting and two factors - voters' concern for government overreach and whether they are hostile toward people of other races.

And Enns - who is co-founder and chief data scientist at Verasight and the Robert S. Harrison Director of the Cornell Center for Social Sciences - said his team plans to continue with this forecast for future presidential elections.

"If the forecast continues to be accurate, then that's great, but if it doesn't, then the factors that have historically predicted the outcome aren't working anymore and we will learn that something is changing dramatically," he said.

Kathy Hovis is a writer for the College of Arts and Sciences.

Public link

Please use the above public link if you want to share this noodl on another website.