In the 2021 Senate runoff elections in Georgia, this writer did a number of stories following forecasts by Thomas Miller, a prominent data scientist who’s a professor at Northwestern University. Miller crunched the numbers and made the contrarian call that Democrats Jon Ossoff and Raphael Warnock would win. In the 2024 presidential race, I tracked Miller’s numbers once again, and he scored by positing a Trump triumph.
I found Miller’s methods so convincing because he takes the novel approach of starting not with polls, but with the biggest prediction markets at the time, then making adjustments using proprietary algorithms that adjust for the sites’ inherent biases. For the 2024 contest, Miller broke new ground by updating his odds of a Trump or Harris victory every 10 minutes, so you could see the latest numbers posted on a continuously moving band resembling a stock ticker.
Now, Miller’s updated his methodology to handicap the Senate elections. (The current ticker is on his site, called The Virtual Tout.) Miller’s calculated the percentage chance that all 35 seats in play will go blue or red virtually in real time.
Once again, Miller started by using the betting odds. “I chose Kalshi and Polymarket because they have the biggest volumes,” he told me. The betting sites post “prices” that don’t translate directly into probabilities. Miller makes the statistical adjustments required to nail the probabilities inherent in the prices. Then, as Miller puts it, “the magic happens.” He runs the probabilities taken from the two betting sites through multiple time series analysis, identifying “underlying factors” that explain 80% to 90% of the changes in the probabilities.
Miller thinks that the underlying factors are likely affected by what Donald Trump does. What POTUS says and does doesn’t move just one race, but all of them. “The states don’t move independently of one another,” he says. “If one moves up for Republicans, chances are they all move up for Republicans and vice versa. The entire market’s driven by what Trump does hour by hour. If voters like what Trump says, the Republican odds rise almost everywhere, when they dislike what he says, the Republican odds drop across the board.”
Miller began posting what he calls “NowCasts” on October 22, but he started applying his methodology on September 19. And since then, the trend toward a Democrat sweep has intensified. On September 19, Miller’s forecast gave the Republicans an almost 36% chance of retaining control of the upper chamber. But as of 11:30 pm on October 23, that probability had fallen over 10 points to just 22.5%. Look at the ticker now and you’ll see that Democrats lead the odds in 18 of the 35 races to 17 for the Republicans. The Dems are looking strong in North Carolina, Ohio, Texas, Alaska, and Maine, while holding firm seats in Michigan (68.6%), Georgia (94.9%).
If the current odds in all those states hold, the Democrats would win 18 out of 35 races and shift the majority from 53-47 for the Republicans to 52-48 for the Democrats.
Miller points out that the numbers are extremely volatile, and could shift big time by election day. Still, the momentum, by his forecast, strongly favors the Democrats. Since October 14 alone, the Dems odds of taking the majority has jumped from 71.5% to 77.6%.
None of this means that Miller’s forecast will prove the most accurate in the days to come. But it provides a highly interesting perspective, positing that as of now, the races aren’t nearly as close as shown by the polls. It also shows statically that it’s all about the POTUS, and that the Trump effect could be proving counterproductive for the GOP.