What’s the Difference Between Polls, Betting Odds, and Prediction Market Prices?

Polls are often conflated with betting odds and prediction markets. It’s a common mistake — here's everything you need to know about the three

RELATED POSTS

As prediction markets gained new prominence during the 2024 presidential race, prediction market prices have been conflated with polls and betting odds. It’s a common mistake amplified by misunderstandings about how betting odds work and a popular conspiracy tying betting markets to a certain Trump victory.

 

The key differences between polls, odds, and prediction market prices include:

 

  • Polls are an estimation of how voters will likely vote 
  • Betting odds reflect potential payouts 
  • Prediction market prices show what a market’s likelihood of an event’s occurrence

 

There are some nuances to separating betting markets from prediction markets. The easiest way to set them apart is to understand why we rely on pollsters instead of bookies.

Polls are about voters

In 1936, George Gallup used weighted responses to correctly predict Fredrick Roosevelt’s victory over Alf Landon. Literary Digest’s poll was based on polls mailed to readers, which undercounted low-income voters. Even after polls – including Gallup’s – incorrectly predicted Thomas Dewey’s victory over Harry Truman in 1948, scientific polling had become a mainstay of American political reporting. 

 

Voting polls are reflections of how voters will actually vote. So, when Donald Trump and Kamala Harris poll at 49% each, that implies that 49% of voters will vote for Harris, another 49% for Trump, and 2% will be undecided or voting for a third party. 

 

However, weighting isn’t a perfect science. In 2016, polls overestimated Hillary Clinton’s projected electoral victory. The polls that were the most wrong failed to weigh responses based on education, a crucial fault line among American voters. 

 

The 2020 presidential election saw changes in polling methods, but new errors cropped up too. One of them was that polls reached too few Republicans and Trump voters. 

 

In his book Strength in Numbers, G. Elliott Morris offers a few theories as to why. One is that Democrats were more likely to isolate during COVID and were therefore more likely to be at home to receive calls from pollsters. Another is that low institutional trust was a new variable unique to the 2016 and 2020 races and wasn’t properly weighted in the last two elections.  

 

Still, these polls model voting patterns directly instead of offering probabilities about who will win. Betting markets technically offer probabilities, but bookies’ odds are not reliable.

Odds are about payouts

Sportsbook odds express payouts through a few formats. For example, +200 odds pay $200 for every $100 bet. Odds can also be converted to probabilities. So, +200 is equal to 33.3%, and -200 is equal to 66.7%. 

 

However, that doesn’t mean that betting odds are precise. A study of political betting markets in the United States found that most of the time, the betting favorite won the election. The study covered presidential elections from 1884 to 1940, and the betting favorite won 73% of the time. 

 

The betting markets weren’t precise in the way polls have conditioned us to expect. In 1936, FDR beat Landon with all but eight electoral college votes and 60.8% of the popular vote. According to betting odds, FDR had a 66% chance of winning the election three days before Election Day. The betting markets didn’t see a blowout coming. 

 

Compare that to the 1896 election, in which McKinley won with only 4.4% more of the popular vote than his opponent. In a much closer race than 1936, election markets passed 66% 97 days before the election. Broad patterns informed newspaper readers about the state of the race, but they were imprecise.    

 

Modern sportsbooks offer another complication when it comes to offering election odds. They are a reflection of payouts and the potential liabilities that sportsbooks must pay to the winning side. Sportsbooks adjust their odds to balance their liabilities on both sides of the line. They don’t want to lose money if they’re wrong about the winner.

 

On Election Day in 2020, FiveThirtyEight gave Biden an 89% chance of winning. Biden’s sportsbook odds were 60-70%. Slate reported that the day after the election – when Biden had already won and Trump’s odds were +525 – 80% of new bets were still coming in on Trump. It was not the rational behavior that markets purport to encourage.    


Polls offer a more precise look at voter movements, but polls rely on models that could misread the electorate. Betting markets are imprecise but can capture trends. 

 

Where does that leave prediction markets, which are simultaneously betting platforms, financial platforms, and forecasting mechanisms?

Prediction markets aren’t polls, but they’re better than sportsbooks

Prediction markets reflect the probability of an event occurring. A 50-cent contract reflects a 50% chance of a payout. Instead of being managed by a central authority like a sportsbook, a prediction market arrives at its price as its traders buy and sell contracts. A prediction market reflects what its traders think. 

 

Prediction markets are not polls because they don’t ask voters how they’re voting. Instead, prediction markets ask who will win the election and aggregate those opinions into clear probabilities.

So, a 10-point lead among voters is dramatically different from a 10-point lead on a prediction market. A 10-point lead in a poll suggests that 10% more voters will cast a vote for that candidate. In contrast, a 10-point lead in a prediction market suggests a 10% higher chance of victory. 

 

Consider a weather forecast. If there’s a 50% chance of rain, how much does a 60% chance change how you view the likelihood of rain? 

 

That doesn’t mean skeptics of prediction markets’ accuracy have unfounded concerns. Prediction markets are only an aggregation tool. Their accuracy depends on their users. 

Mythbusting: Misunderstandings about prediction markets

Prediction market prices are only as insightful as their traders. University of Chicago economist Dr. John Birge emphasized that prediction markets can reflect a group’s biases if the group is insufficiently diverse. 

 

Prediction market traders are a self-selecting group who are willing to learn how to trade on the election or who just want to bet on it. Traders don’t have to match the demographic information of likely voters, but traders must cancel out each other’s errors to produce an accurate forecast. 

 

Traders don’t have to be politically partisan to produce imprecise election forecasts, either. Over-relying on certain media narratives would reflect common stories rather than an accurate forecast. Betting patterns like preferring to bet on favorites make a prediction market’s price fuzzier than a prediction based solely on a user’s past accuracy.  

 

After the shooting at Trump’s July 13 rally, Trump’s Polymarket chance of victory spiked from 59.5% to 67.5% over the next 12 hours. His chance peaked at 71.5% before sharply falling back to 61.5% after the Republican National Convention. That surge suggested a brief enthusiasm for Trump contracts followed by a sell-off at a reliable high point for the contracts. Trump lost the lead after Harris entered the race, but has become a clear favorite in the final weeks of the race.    

 

In contrast, Metaculus is a prediction market platform that doesn’t use money. Instead, users forecast the probability of events, and their guesses are weighted based on how accurate they’ve been in the past. 

 

Metaculus gives Trump a 54% chance of winning, a functional tie. Metaculus shows the same trend of Trump’s improving chances throughout October. But the reputation-based platform reflects a closer race than real-money markets subject to common betting patterns, unclear ideological commitments among its traders, and other sources of statistical noise.    

Prediction markets and polls aren’t supposed to match

Analysts also shouldn’t expect prediction market prices and polls to match, especially after poll movement. Sept. 11-12 Reuters/Ipsos polls found Harris leading by four to five points nationally. Mid-October polls found her ahead by only two to three points. Several swing states where Harris previously led have also polled at ties.  

 

As Trump gained in the polls, more traders bought his contracts on prediction markets. Trump’s price rose from the mid-forties to the low sixties across prediction markets. One large user on Polymarket doesn’t negate the legitimate price movement in Trump’s direction. 

 

On the other side, prediction market prices aren’t “proof” of an impending Trump victory. Prediction market prices reflect what traders believe at the time, but those beliefs change. Trump used to be priced at 71 cents on July 16. Trump was at 46 cents on Aug. 16. Clearly, trader sentiment isn’t prophecy.

Each prediction market also uses an external confirmation mechanism to settle its market. If you’re betting on Kalshi, they use the Office of the Presidency as the source to determine who will settle its presidential market. Polymarket will use the Associated Press, Fox News, and MSNBC. If the three companies don’t call the race for the same candidate, Polymarket will settle its market based on the candidate inaugurated. 


Prediction markets attempt to reflect reality, not recreate it. It’s wrong to claim that prediction market price changes outweigh the real votes that people cast. Mapping conspiracy theories onto prediction markets is not a good way to leverage prediction markets’ popularity.   

Buy JNews
ADVERTISEMENT

Join the

Prediction News Community

Featuring prediction market
analysis, data insights
plus
comprehensive industry reporting

Prediction Platforms

Who will win the 2024
US Presidential Election?

Loading..

Loading..

Loading..

Loading..

Loading..