In US presidential elections, there’s the main contest – the one between the candidates – and then there’s the race among polling and modeling organizations to accurately predict the outcome. In my academic work, I’ve been engaged in election modeling over many election cycles – and the 2024 presidential contest was no exception. Heading into that election, most polls anticipated a Harris win for the popular vote, with a more mixed picture of the Electoral College outcome. Our research team’s model told a different story.  

‍

Every Electoral Vote Correctly Predicted

After all the votes were counted, the model that my colleagues and I developed was the most accurate. Using Verasight data, we were able to correctly predict every Electoral College vote, and we got within one percentage point of the popular vote. (We predicted Trump would win 50.3 percent of the two-party popular vote; he won 50.7 percent.) Our main forecasting innovation was to measure both presidential approval ratings and economic conditions for each state, rather than for the nation as a whole. And it worked. 

‍

The media took notice

Although we’re happy to stay behind the scenes, focusing on data and modeling, it was rewarding to be recognized in the media in this election cycle:

The Telegraph (UK – November 9, 2024): “The race between Kamala Harris and Donald Trump, if the polls were anything to go by, was extremely close. Except, it wasn’t. Mr. Trump was destined to win – according to one group of political scientists who forecast the exact result, state-by-state, nearly a month ago.” 

The Guardian (UK – November 28, 2024): “Rather than focusing on the candidates’ support nationwide or in the swing states, Enns and his co-authors built a model that combines two types of data: presidential approval ratings from all 50 states using data from Verasight, the survey firm he co-founded, among others, and a Federal Reserve Bank of Philadelphia index measuring state-level real income, manufacturing and labor market conditions. Both sets of data were compiled more than 100 days before the vote.” 

‍

Data made the difference

Verasight wasn’t the only source of data used in our model, but it was a critical factor. We used it for the same reason so many Verasight clients depend on it: Our focus is data quality, and we recognize the challenge of conducting trustworthy research amid a sea of bots, “professional” survey takers, and inattentive respondents. We have worked to build innovative, practical ways to engage with reliable, authentic survey takers to generate high-quality data.  

For a closer look at our 2024 election predictions, our article in PS: Political Science & Politics is available here. To see how our research stacked up with other well-known polls and models, view the table below from “The PollyVote Forecast for the 2024 US Presidential Election”, an article also slated for publication in PS: Political Science and Politics by Andreas Graefe, professor of management at Macromedia University of Applied Sciences in Munich, Germany.  

And if you have any questions about how you could put Verasight data to work in support of your research projects, just write directly to me to start the conversation, using my email address linked below. We’d love to help. 

‍

Peter Enns
Co-founder and Chief Data Scientist, Verasight 

Professor of Government and Professor of Public Policy
Robert S. Harrison Director, Cornell Center for Social Sciences
Cornell University

‍

* * *

PS: As noted above, every successful research project is the result of a team effort. My team for this year’s 2024 election predictions included:

• Jonathan Colner, New York University
• Anusha Kumar, Yale University School of Medicine
• Julius Lagodny, ELPATO (Germany)

In 2020, Julius and my forecast correctly predicted 49/50 states.