An election forecaster who previously predicted a landslide victory for Vice President Kamala Harris has changed his prediction to former President Donald Trump.

Thomas Miller, a data scientist who accurately predicted the outcome of the 2020 presidential election, previously backed Harris to win in 2024 Newsweek in September that, according to his model, the Democrat was on track to win more than 400 votes in the Electoral College.

The model, which is based on betting odds rather than polling data aggregates, now says Trump is the favorite. The latest forecast is that the Republican could win 345 votes in the Electoral College.

Newsweek has laid out what a potential victory for the former president in the 345 Electoral College could look like. In this scenario, he sweeps the South and Midwest, winning all the states where FiveThirtyEight predicts Harris has only a single-digit lead.

The model first began predicting a narrow Trump victory on October 7, when betting odds began to swing toward the former president.

Miller's model prioritizes betting odds over poll data and also uses historical trends from previous elections to match the data with what he calls “fundamentals,” which is what most other models do.

For the 2024 election, Miller is using the betting site PredictIt as the basis for his model, which was the same data that based his successful 2020 forecast.

However, the data and fundamentals suggest different results. Miller wrote on his website Oct. 20 that “technical and fundamental analysis disagree” as historical trends point to a Democratic victory, which contradicts the betting odds that Trump will win.

This is consistent with what his election observer Allan Lichtman also said about the election. Unlike other forecasters, Lichtman's Key to the White House model relies almost entirely on long-term historical trends and has consistently predicted a Harris victory.

While polls still suggest the race is close, the betting odds remain in Trump's favor. Polymarket odds show Trump leading Harris nationally with a 60.3 percent chance of winning, compared to 39.7 percent for Harris.

Although betting odds have historically consistently predicted the correct winner of presidential elections, they are not based on the same representative statistics as polls and are therefore more susceptible to bias.

Miller told Newsweek that he corrects potential biases in his model that currently favor Trump.

“Prediction markets have a Republican bias,” Miller said. “We assessed the level of Republican bias in 2020 and are currently correcting that level of bias. “What is uncertain, however, is the extent of Republican bias in the prediction markets in 2024. We are conducting studies across all prediction markets, which we hope will shed light on the level of bias in 2024.”

Miller's latest prediction actually represents a slight improvement for Harris; On October 26, the model indicated that the vice president would win just 171 Electoral College votes, the lowest result for a candidate since Republican nominee Senator Bob Dole in 1996.

“The 2024 presidential race has gone from back-and-forth to a Republican landslide, then to a possible Democratic landslide, to a possible Republican landslide, and now to a possible Republican landslide,” Miller said. “Could the prediction markets and associated election forecasts change again with less than two weeks before voting ends on November 5th? Yes. We expect increased trading and high volatility in the final week of the race.”

In the 11 presidential elections since 1980, the only race in which the winning candidate had worse odds than the losing candidate was in 2016, when neither the betting markets nor conventional polls could predict a Trump victory.

Update, 10/30/24, 11:29 a.m. ET: This article has been updated with additional information

Do you have a story that Newsweek should cover? Do you have any questions about this story? Contact [email protected].