Election odds are fun to discuss, but they aren’t a good indicator of how South Carolina’s primary will turn out.
Sportsbook odds are skewed by how much money that bettors place on different candidates and how sportsbooks balance their liabilities behind the scenes. During heavy trading on one candidate, the other’s odds can increase to attract more money to that side of the line and balance potential payouts with revenue.
Prediction markets do a much better job of converting customer guesses to probabilities. However, prediction market accuracy depends on the users making predictions. Like sportsbooks, prediction markets are vulnerable to market manias. Prediction markets can also reflect highly publicized polls instead of making unique predictions.
Polls are the only method where a professional can mitigate audience biases and sample issues. Accurate polling doesn’t happen without proper technique, and there are chronic mistakes that pollsters must learn from after the 2016 and 2020 presidential elections. But a professional poll remains the best source of election data.
2024 presidential election odds and their flaws
Election betting is popular overseas, but it’s illegal in the United States. American election betting lost its popularity with the rise of scientific polling driven by George Gallup’s successful prediction of the 1936 election. Other forms of gambling also replaced election betting, and it hasn’t made a comeback since.
Today, election odds are sometimes roped in with prediction markets. Sportsbooks aggregate many guesses about the future. However, sportsbooks give bettors lines they want to bet on instead of reporting accurate predictions.
Oddschecker listed U.S. presidential election odds before and after the New Hampshire primary. The odds changes told no coherent story about the primary’s impact on the general election:
|Donald Trump +120
|Donald Trump +102
|Joe Biden +200
|Joe Biden +198
|Gavin Newsom +2600
|Michelle Obama +2000
|Michelle Obama +2700
|Nikki Haley +3100
|Robert Kennedy, Jr. +2800
|Gavin Newsome +3300
|Nikki Haley +3100
|Robert Kennedy, Jr. +4100
|Kamala Harris +8900
|Kamala Harris +7900
Before and after New Hampshire, Nikki Haley’s odds didn’t change, and the two parties’ frontrunners’ odds barely moved. However, Michelle Obama, who is not running and has a long-running disdain for how politics is practiced, has odds listed. Additionally, her odds improved after receiving zero delegates in New Hampshire.
In 2020, Trump bettors kept pouring money on a win even as Biden secured his victory. In response, sportsbooks had to increase Biden’s odds and decrease Trump’s odds. The implied probabilities made Biden’s chances look worse even as he was securing critical electoral college wins.
Odds movements tell an interesting story about where bettors are willing to put their money. But sportsbooks are an entertainment product, and it shows.
PredictIt markets and audience effects
Unlike sportsbooks, prediction market probabilities can be taken at face value. PredictIt prices allow users to buy trades from $0.01 to $0.99. Correct predictions earn $1 per share. Incorrect predictions earn nothing. This pricing structure translates share prices into percentages.
Prediction markets are a good way to translate user opinions into percentages. However, the wisdom of crowds only works if the crowds have relevant information about the question they’re answering.
|Trump - 47 cents
|Biden - 46 cents
|Biden - 45 cents
|Trump - 46 cents
|Haley - 5 cents
|Newsom - 6 cents
|Newsom - 4 cents
|Haley - 5 cents
If prediction market users are making decisions based on publicly available polls, then the prediction markets won’t reveal new information. It’ll just be similar to the polls that most users read before they trade. But a well-sourced prediction market could answer questions that pollsters miss.
For example, professional pollsters have to find ways to tell whether respondents are likely Democrats or Republicans. G. Elliott Morris’ book, “Strength in Numbers,” found that in 2016, pollsters missed education and social trust as key factors in predicting votes.
Many missed social trust again in 2020 and over-predicted Biden’s margin of victory in some states. Morris also notes that “Nate Silver’s model had Biden up by eight points [in Wisconsin] when he only won by 0.6.”
A prediction market doesn’t have to manually code those voters. It just has to ensure that a representative amount of money from those voters is included in the market. Prediction markets are potentially interesting checks on polls but can’t be trusted without in-depth questions about their customer bases.
Polls, shortcomings and useful predictions
Before George Gallup correctly called the 1936 presidential election, Literary Digest sent a poll out asking its readers who the president would be. Gallup weighted his survey respondents’ answers based on their representation in the country’s population, not just who was in one magazine’s readership. That led to his correct prediction and contributed to the popularization of weekly scientific polls.
Done well, scientific polls can predict voter trends invisible to sportsbooks and prediction markets alone. It’s why the changes in the Republicans’ average national primary polls before and after New Hampshire from FiveThirtyEight are insightful and only include small changes. Little changed before and after Trump’s New Hampshire victory because he was already the clear Republican frontrunner:
|Trump - 67.8%
|Trump - 68.5%
|Haley - 12.3%
|Haley - 12.2%
However, there are conditions that polls have to fill to be accurate after poor predictive patterns in 2016 and 2020. In “Strength in Numbers,” Morris found that cold calling and internet surveys overrepresent urban voters. Urban voters are more likely to answer surveys conducted over the phone or the Internet, and urban voters are more likely to be Democrats.
Pollsters also have to expand the margin of error. The margin of error is sampling error, the error introduced by sampling deviations. Morris also lists nonresponse error, measurement error, coverage error or the survey’s weighting algorithm as potential sources of error. He suggests that pollsters disclose these additional sources of error. For consumers of polls, he suggests they double the margin of error as a back-of-napkin calculation.
While polls aren’t perfect, a flawed pollster is more credible than a prediction market that doesn’t disclose key audience metrics or a sportsbook that’s offering entertainment in lieu of insight. It’s better to follow Morris’ recommendation to become a better consumer of polls than to look elsewhere for electoral wisdom.