For the past few weeks, I've been following presidential election predictions published by Thomas Miller, a data scientist at Northwestern University. Miller bases his predictions not on polls but on prices displayed on what he believes is America's most trusted political betting site, PredictIt. Since June 8, Miller has updated his forecast once a day. The new numbers will be posted on its homepage, the Virtual Tout, at midnight EST.

Warning to election watchers: On Oct. 29, Miller unveiled a new website that continuously displays voter counts that PredictIt odds suggest each candidate will win. PredictIt updates its prices once per minute. Miller immediately translates these odds into a prediction for the number of electoral votes (EVs) each candidate will win. “We show where the election is going in almost real time,” he says. “It’s like having your own personal presidential election ticker tape.”

For now, you can click on virtualtout.io to see what this author thinks is the best data to hamper the wild race for the White House in 2024. Believe me, it's addictive.

As I've described in previous stories, Miller used PredictIt in 2020 to make highly accurate decisions on both the presidential race and the two subsequent Georgia Senate runoffs. Miller likes PredictIt because, he says, it's extremely responsive to campaign events. By carefully studying how PredictIt works, Miller developed a model that converts the prices posted on the site into the population vote shares each candidate receives on a moment-to-moment basis. To get there, Miller makes a number of adjustments. For example, he noted that prices have a slight Republican bias because most PredictIt bettors are men who also frequently bet on sports. So his frame corrects the rightward tilt. Likewise, the data scientist found that when a participant is far ahead in PredictIt, the odds change far too abruptly in his favor. Reason: When the race looks like a sure thing, bettors flock to back the “winner,” creating a bandwagon effect that drives prices to “landslide” extremes well beyond the likely margin of victory.

Miller also examined all elections since 1960 and found a close relationship between a candidate's share of the total number of votes cast nationwide and his share of the 538 EVs that decide the winner. He has developed a proprietary framework that converts popular vote counts derived from PredictIt prices into electoral votes.

As of Oct. 29 at 3 p.m., Miller's “ticker tape” showed this: Trump/Vance 345, Harris/Walz 193. Next to the totals for each candidate, you'll always find a red or green arrow showing where you stand compared to their count at midnight and the amount of the change. In our snapshot, Harris/Walz is at +1 and Trump/Vance is at -1. “Trump is still way ahead,” says Miller. “But prices are moving in the vice president's favor.” He points out that Trump hit his all-time high of 367 on October 26, versus 171 EVs for Harris, marking a huge turnaround from late September, when the former president was by a wide margin remained behind.

The story continues