I started following Thomas Miller’s political forecasts during the January 2021 Senate races in Georgia. Just a few days before the election, the polls and pundits anticipated a relatively easy win for both Republican candidates. But Miller—a prominent data scientist and professor at Northwestern University—posited a Democratic sweep, and he was right. He also called Trump’s victory in 2024. What’s different about Miller’s models? They’re far more current than the polls, and they’re centered on prediction market prices that are typically a better guide than the voter surveys—plus Miller adjusts the outlook in every Senatorial contest by using all we know about every contest.
Now, Miller is applying an updated version of his model for the current Senate elections. You can find the stats on his website, Virtualtout.io, and I highly recommend paying a visit. Once again, Miller starts by incorporating the betting venues’ data. “I use the numbers from two sites, Polymarket and Kalshi,” he told Fortune. “They’re the ones that trade the most dollars, where most of the action happens.” Miller then calibrates the posted “prices” so that they reflect probabilities, and the odds for both leading candidates tally to 100%. In a further step, he deploys multiple time series analysis to find underlying factors driving the betting prices, and uses trusted models from finance and economics to project how they’re likely to influence voters moving forward.
By the way, he thinks his model works better than AI.
Miller updates the probabilities of a Republican or Democratic win every 10 minutes. The odds that a candidate will win scroll across the top of the Virtualtout.io screen continuously, in a band resembling a stock ticker tape. For each race, Miller also posts the change over the last 24 hours. Miller calls the streaming feature the “NowCast.”
As for those “underlying” factors, one force swamps all others: Donald Trump. “The states don’t move independently of one another,” says Miller. “If one goes up for Democrats, chances are they’ll all go up for Democrats, and vice versa. The entire market is dominated by what Trump does and says hour by hour.” Put simply, when voters don’t like the POTUS’s words or actions, all Republican odds tend to fall. Miller asserts that his model has identified 85% to 90% of what’s causing the variances on Kalshi and Polymarket, and “my interpretation is that it’s overwhelmingly Trump and the Republican positions generally.”
Though he opened Virtuatout.io to public view about a week ago, Miller started running the numbers on September 17. What’s most striking is just how far the GOP chances have faded in just six weeks.
As of around 9 AM on October 30, five days before most polling sites close, Virtualtout.io gives the Republicans a 17.1%, or just over one-sixth probability, of winning 50 seats and hence retaining control of the upper chamber. On September 17, that figure was twice as high at 34.8%, or slightly better than one-in-three. In all five contests where the Democrats can flip a seat, their chances are better than 60%—in most cases, much better.
In North Carolina, where Republican Thom Tillis is retiring, leaving an open seat, the GOP was never remotely close, and now harbors a minuscule 4% shot at victory. In Texas, the Republican odds have dropped since September 17 from 46.15% to 31.9%, in Ohio from 45.4% to 31.8%, and in Alaska from 32.95% to 28.6%.
In only one state is the GOP faring better than in mid-September and that’s Maine. Republican Susan Collins started as an underdog at 32.6%, but has clawed back to 39.9%, adding 1.7 points in the last 24 hours. In Alaska, things are also looking up for the Republicans: After falling to a low of 23.4% on October 23, incumbent Dan S. Sullivan’s prospects have improved by over 5 points to 28.6%.
For the political pros looking for harbingers of big trouble for either side, the towering sign is the Republican woes in traditional strongholds Iowa and Kansas. In the former, Republican Ashley Hinton dropped from 61.75% on September 17 to a virtual tie at 53.4%, a trend mirrored in Kansas where Roger Marshall was looking a likely victor six weeks ago at 75.8%; now, he’s down almost 19 points at 57.1%. In the last day, Hinton and Marshall have shed 1.45% and 2.47 points respectively.
In the one state where Republicans saw a big chance for a flip, the horizon’s receding fast. The GOP chances of beating Abdul El-Sayed in Michigan have slipped from better than a third in mid-September to about a quarter (25.4%).
If the Democrats indeed score a flip in every contest where their probabilities in Miller’s model stand at 60% or better, and hold all the seats they have now, they’ll clinch a majority of 52 to 48, nearly reversing the Republican dominance at 53-47.
In South Carolina, recently rerated on the Fox News Channel from “lean Republican” to “tossup,” Miller still has Darline Graham at a commanding 75%, though she’s slipped 2.4 points in the past 24 hours.
Keep in mind that Miller’s setting odds, not making definitive predictions. A candidate at over 30% still stands a decent chance of winning. And Republicans are over that bogey in Maine, Ohio, and Texas, and close and trending the right way in Alaska. Odds-defying events happen all the time. The problem for Republicans is that they’d have to beat those odds multiple times, plus defend seats they never dreamed they’d have to defend.
It’s still far from over. And one of the best places to get a plausible, non-AI driven snapshot from a data scientist who’s been doing this for six years, is that NowCast on Tom Miller’s Virtualtout.io.