Fortune reports:

Data scientist Thomas Miller has crafted a model for forecasting the 2024 presidential election that, he says, is far more reliable than the polling that’s constantly cited in the media as the best guide to the outcome on Nov. 5. Instead, the Northwestern University professor deploys a framework based on the betting odds set by folks wagering their dollars not on the candidate they intend to vote for, but the one they expect to win. This writer began following Miller’s predictions during the 2020 White House contest, and the two Georgia Senate races that followed in early January of 2021. Miller called the former within 12 electoral votes, and correctly posited that the Democrats would sweep both Senate seats when the polls showed the Republicans significantly ahead. His calls for the margins of victory proved right on the mark.

Given Miller’s excellent record in the 2020 cycle, it’s highly instructive to examine his outlook for this year’s race. And as in the case of the Georgia runoffs, his view is shockingly contrarian: While most pundits and prognosticators see a “dead heat” or “toss-up,” Miller’s numbers show the Harris-Walz ticket far in the lead and the wide Democrat advantage settling into a remarkably stable pattern.

Miller’s model centers on the “prices” posted on America’s most active political betting site, PredictIT. It’s a highly liquid market that trades, on average, 37,000 shares per day. His ruling principle: The odds for the two candidates, after eliminating bets on third-party contenders, closely reflect their share of the popular vote. If a candidate’s price on PredictIT is 54 cents, that 54 cents can be used to estimate vote shares, which will be lower, but still still track those odds. Miller also concludes from assessing the results of all presidential elections since 1960 that the popular-vote shares reliably predict the percentage of the electoral tally each candidate will receive. “The core of this model is 60 years of presidential election history,” Miller avows. He stresses his respect for the work of Allan Lichtman of American University, who also bases his widely followed predictions on the history of presidential elections.

“The polls rely on much smaller samples than on PredictIT,” says Miller. “They typically survey only 500 to 1,500. There’s no consistency in the people who are polled. The pollsters are all using different groups of likely voters. They’re are also using different methods, so there’s no single way of analyzing their data. By contrast, tens of thousands are voting with their dollars on PredictIT. If they’re right, they turn 57 cents into a dollar. It’s a substantial return on investment. The participants are motivated to stay in the market until the election’s over to get the payoff, so there’s a consistency to the group of investors who are setting the prices.” [Continue reading…]