To the untrained eye, prediction markets suggest the winner of the Minnesota Senate race is almost guaranteed: Kalshi odds give Democrat Peggy Flanagan a 90% chance of winning. Looking at the polls, the race might seem closer. Averages from VoteHub have Flanagan leading by just under six percentage points, a solid but not safe margin.

Those numbers may seem contradictory. In fact, they’re really just two ways to capture a similar picture.

Prediction market odds have drawn more attention this election cycle, with prime placement in news coverage, campaign materials and political ads. Some people have accused platforms like Kalshi and Polymarket of blowing it when the candidate they favored lost. Others, armed with screenshots of a candidate’s odds plummeting as results come in, have held up prediction markets as evidence of foul play.

Despite their “truth machine” marketing, prediction markets have their limitations, just like polls. But part of the backlash against them is rooted in misguided assumptions about what they tell us — and what they don’t.

When it comes to the midterms, the odds on Kalshi and Polymarket are generally pointing in the same direction as polls. While a 75-25 split in the prediction markets can look like a massive blowout next to a 5-point polling margin, those numbers are in fact largely aligned. That’s because they measure different things. A poll margin is the size of a candidate’s lead, while market odds are the chance of that candidate winning on Election Day. Compare the two across key midterm races, and the polls reflect most of where the markets land.

Some prediction market experts are expressing concern about how people interpret these odds. “We're trying to teach people to accept the new way of thinking about politics and outcomes and forecasting,” John Phillips, co-founder of the politics-focused prediction market platform PredictIt, said during a prediction market conference last week. “This is probability as opposed to margin of victory.”That said, there are some races where polls and markets genuinely disagree.

Governor races in Nevada and Oregon are polling as toss-ups, while prediction markets favor the Democrats. In South Carolina and Nebraska, polls show the Republican Senate candidates with slim leads, while markets imply a roughly 70% chance of victory for both.

Polls and prediction markets can diverge for a number of reasons. Traders might be reacting to news in real time, deviating from polls that capture a snapshot of sentiment. Users might place wagers informed by privileged information. (Kalshi and Polymarket have both investigated numerous cases of insider trading around the midterms.) Some might just be going off of vibes. But it’s hard to fully account for the differences. Prediction markets show how traders behave, not why.

Even when they’re aligned, polls and prediction markets answer different questions. A poll estimates who leads and to what degree at a moment in time, with uncertainty reflected in a margin of error. A prediction market translates financial wagers into an estimate of a candidate’s likelihood of winning, with uncertainty baked into the odds themselves.

At the recent prediction market conference, veteran political journalist Chuck Todd expressed frustration about the general public’s difficulty making this distinction. “Would you get in an airplane that only had an 80% chance of landing?” he said. “Then suddenly people start thinking about probability differently.”

But insiders sometimes talk about prediction markets like they have the predictive powers of a crystal ball. Last year, Polymarket CEO Shayne Coplan told 60 Minutes that prediction markets are “the most accurate thing we have as mankind right now.”After Donald Trump won the 2024 presidential election, Kalshi CEO Tarek Mansour declared that his platform “proved to be the most accurate source of predicting election outcomes.”The certainty with which these executives talk about their markets when they seem to get it right helps explain why some people feel misled when they don't.

The night before Wisconsin’s gubernatorial primary in August, a Kalshi market gave progressive candidate Francesca Hong a 95% chance of winning. When she ended up losing by less than one percentage point, critics pointed to the election as a prediction market misfire.

Following Brazil’s presidential election, Kalshi was subjected to similar criticism for giving Flávio Bolsonaro an 80% chance of victory before he was defeated by Luiz Inácio Lula da Silva. On X, Mansour said it was one of many examples of a “dude not understanding probability.”In an ensuing back-and-forth, the executive conceded that claims of prediction markets “calling” the 2024 US presidential result may have overstated the data, contributing to confusion over probabilities.

These debates aren’t new. Election forecasters have been assigning probabilities to election outcomes for decades, feeding polls along with so-called fundamentals — such as partisan lean, incumbency, fundraising and candidate strength — into statistical models. Nate Silver helped popularize the practice in 2008, when his FiveThirtyEight model correctly favored the presidential winner in 49 out of 50 states and then went 50 for 50 four years later.

But in 2016, almost every major model favored Hillary Clinton over Donald Trump. The New York Times’ Upshot model gave Clinton an 85% chance of winning going into Election Day, but swung sharply toward Trump as returns rolled in. The change was depicted in real time by the Needle, a speedometer-like graphic oscillating to reflect each candidate’s likelihood of winning.

FiveThirtyEight was more cautious than most, but also underestimated Trump at a 29% chance. In the aftermath, a debate broke out over whether the models had failed or if the public had misperceived a 71% chance as a sure thing.

This election cycle, Polymarket and Kalshi are largely in line with the forecasting models. On the biggest question — which party will gain control of the Senate — most models surveyed by Bloomberg give the Democrats a 62% to 79% chance of winning 51 or more seats. As of Friday, Kalshi’s odds fell in the middle of that range at 71%.

A common argument in favor of prediction markets is that they can react to shifts in voter sentiment instantaneously, while polls are inherently on a lag. Following a controversial attack ad released by Vivek Ramaswamy, the Republican candidate for Ohio governor, traders on Kalshi and Polymarket boosted the odds of his opponent, Democrat Amy Acton. In polling aggregators and many forecasts, the race still appears just as close as it was a few weeks ago.

“When you're talking about a routine election, when nothing terribly unusual is happening, then polls and models that are based on polling data and other kinds of data do a pretty good job of generating forecasts,” said Rajiv Sethi, an economics professor at Barnard College who studies prediction markets. “But when something really unusual happens, then markets are able to adjust to it and respond to it.”Polls are still one of the most important inputs for prediction markets, Sethi said. Traders follow their results closely as they place their wagers. On the flip side, a growing number of models factor prediction markets into their forecasts, creating a circular relationship that makes it tricky to draw a clear line between one and the other.

Election prediction markets are getting more eyeballs than ever as they make their way into political s, campaign communications and news coverage. Some experts are sounding the alarm about their downstream effects on democracy.

Prediction markets have been used to cast doubt on the integrity of elections after they conclude. Amid the Los Angeles mayoral primary, Kalshi asked paid political influencers to remove posts that suggested candidate Spencer Pratt was being robbed, pointing to his once-strong odds in a Kalshi market as evidence, as Semafor first reported.

Election integrity experts are bracing for more claims of fraud based on prediction market odds as the results trickle in. “We’re anticipating prediction markets playing a very big role in rumoring on Election Night,” said Priya Dhawka, a researcher at the University of Washington’s Center for an Informed Public.

A recent study from the Brennan Center for Justice found that prediction market odds impact voters’ views of the electoral landscape. That’s raised concerns about the possibility of market manipulation, as strategically placed bets can significantly shift the appearance of a candidate’s odds, potentially shaping where donors give and how people vote.

Without proper regulation, election markets are vulnerable to becoming self-fulfilling prophecies, said Sethi, the Barnard economist. That sets them apart from something like a market on weather, he said.

“I’m not going to make it rain simply by thinking that it’s more likely to rain,” he said. “But elections have the property that people’s beliefs about what’s going to happen can actually affect what happens.”