Sign up for the daily CJR newsletter.
“I know we’ve had some big polling misses in primaries recently, and that has everyone on edge,” G. Elliott Morris, a journalist and pollster who writes a Substack called Strength in Numbers, told me. Polls showed wide leads for progressive candidates in some key Democratic primaries over the summer: Abdul El-Sayed was predicted to win Michigan’s Democratic Senate primary by double digits but only eked out a victory. Francesca Hong’s big lead in Wisconsin’s gubernatorial primary turned into a loss for the Democratic Socialists. Peggy Flanagan, a progressive from Minnesota, beat her opponent by nearly twenty points despite polls calling the race “deadlocked.” These misses drew national headlines questioning whether polls have gotten less credible, and whether we are in for more of the same inaccuracy in the general elections, too.
That fear may be misplaced—experts I spoke to highlighted the volatility of primaries compared with general elections thanks to low turnout, shifting tickets, and candidates’ still refining their message—and Morris told me that primaries are so plagued with inaccuracies that “you should literally expect them to miss the mark.”
Still, journalists should be careful about how they report on polls in the upcoming midterms. “One of the first things we look for is whether the poll is aligned with a partisan cause,” William P. Davis, the director of election data analytics for the New York Times, told me. “Campaign pollsters are often regarded within the polling industry as having pretty good track records, but you have to understand the motivations behind why this polling information is being released.” Often, their goal is to build a specific narrative—like that their candidate is doing better than other polls suggest—in an effort to gain momentum and encourage donors. Davis believes campaign polls should be used as background reporting.
Specificity and transparency are crucial—and the more that a pollster reveals, the more confident journalists can be about the quality of the survey. “Did they release all of the questions that they asked, and the full question wording? Did they call voters? Did they text voters? Is this a panel of voters that was built using probabilistic means? All of those can lead to pretty different results,” Davis said.
“It’s even better if you can get a cross-tabulation document, the results of the poll by different demographic groups, or a statement by the pollster, some sort of description of the composition of their survey,” Morris noted. It’s also important to realize that a single poll seldom warrants a full story. He suggests averaging polls or consulting forecasting models that feed individual polls into a larger dataset for analysis.
Context is also crucial. “If the candidate’s up by one or two percentage points, it’s really not good practice to describe those things as leads,” Scott Keeter, a senior survey adviser at the Pew Research Center, told me. “And by the same token, when we’re looking at trends in polling from one poll to the next, it’s very easy to overinterpret shifts in the margin between the candidates.” It gets even more misleading when reporters try to interpret survey figures in terms of the likeliness of the result. “The public doesn’t process those numbers very well and can read much more into it,” he said. “They can also think that an 80 percent chance of winning means they’re going to get 80 percent of the vote, which of course is not true. We strongly recommend against converting polling leads into some probability that a candidate will win.”
But polling is an art as much as a science. “There is no such thing as the single best way to conduct a survey in 2026,” Morris said. “An accurate poll now has much less to do with how you sample people, how you talk to them or whether or not you’re interviewing them online or over the phone.” Instead, he said, it’s about accounting for the gap between those who say they’ll vote a certain way and those who actually show up to the polls—a factor known as “nonresponse bias.”
Polls can also provide essential information about the problems and priorities of citizens. “We want to understand why people are doing what they’re doing,” Keeter said. “What is the story of the election? What are the issues that are motivating people? How are they evaluating the candidates as people and as likely leaders?” Paraphrasing Sidney Verba, the political scientist, Morris said that polls were “a microcosm of democracy” and that public opinion surveys, in their most ideal form, are “a way to represent everyone.”
One new development this time around: the rise of prediction markets like Kalshi and Polymarket, which have presented themselves as election forecasters. CNN, for instance, has given odds from Kalshi as win probabilities in a segment called “The Odds,” where pundits also reminded viewers about the inaccuracy of polls in 2018 and 2024. Dustin Gouker, an expert on gambling and prediction markets who covers the industry in his newsletter, Event Horizon, told me that it’s important to understand that prediction markets aren’t standalone sources but a reflection of what people make of “all the data we have: all the polling, all the news and historical trends.” But the experts I spoke to emphasized that markets lack the margins and the nuances of a well-executed poll. “Journalists should not be consulting prediction markets if they want to talk about the voters,” Morris said. “If you care about what people think, you should just ask them directly.”
Has America ever needed a media defender more than now? Help us by joining CJR today.
