New election forecast model predicts Harris win

Kamala Harris and Tim Walz. – © AFP SAUL LOEB

A new model updates US election forecasts. This is by drawing on daily data, driven by the odds market. The model comes from a Northwestern University data scientist. Using these data, the new forecasting model predicts how the Electoral College will vote

Similar models correctly predicted the 2020 presidential election and 2021 runoff elections for two Georgia senate seats.

The model updates the odds of a win by former President Donald Trump or Vice President Kamala Harris each day. With this level of precision, the new media and other interested parties can see how single events — such as a debate, campaign activities or legal rulings — might affect the potential outcome of the U.S. presidential election.

According to Northwestern’s Thomas Miller, who developed the platform, the presidential debates will be critical and “Trump’s legal events also are critical. There have been massive changes in forecasted Electoral College votes in recent months associated with current events and campaign activities.”

Miller is the faculty director of the master’s in data science program at Northwestern’s School of Professional Studies. Viewers can follow his daily predictions and accompanying analysis on his site, The Virtual Tout.

Miller’s system uses data from PredictIt, prediction markets in which users bet real money on political races. Miller then uses pricing data as input to his forecast for how the Electoral College will vote. 

Using his daily forecasts, Miller can gauge responses to singular news or campaign events. When Trump received a sentencing delay for his New York “hush money” conviction, for example, Harris’s campaign experienced a predicted drop of 68 forecasted electoral votes, according to Miller’s model.

Miller’s system currently predicts the Harris-Walz ticket will win the November election with 289 electoral votes. Candidates need 270 votes to win the presidency.

Miller’s models proved considerable accuracy during the 2020 presidential election — only predicting one state (Georgia) incorrectly. Miller has updated the algorithm from this error, uncovering innate biases that led to a Democratic Electoral College vote prediction that was 12 votes lower than the actual Electoral College vote.

Miller’s 2020 model was still more accurate than Nate Silver’s FiveThirtyEight. While FiveThirtyEight predicted 348 electoral votes for Biden-Harris, Miller predicted 294 electoral votes for the Democratic ticket. Ultimately, the Electoral College casted 306 votes for Biden-Harris.

Miller says popular election forecasting systems, including FiveThirtyEight and The Economist’s predictions project, are inherently flawed because they use data from opinion polls. According to Miller, these data are old, compared to fast-moving news cycles.

Miller also notes another key advantage of prediction markets over political polls: Large groups of investors who stay in the markets until Election Day. These groups grow larger as Election Day approaches. Miller relies on prediction markets that have tens of thousands of investors, with thousands of shares traded each day. Typical opinion polls involve between one and two thousand respondents, with new respondents recruited for each poll.