Preelection polls were wide of the mark in two closely watched Democratic primaries this month in Michigan and Wisconsin. In both races, polling showed progressive candidates leading by large margins, but the results proved much closer. In Michigan, former public health official Abdul El-Sayed only narrowly defeated Rep. Haley Stevens to win the party’s nomination for an open U.S. Senate seat. In Wisconsin, left-wing state lawmaker Francesca Hong wound up losing the gubernatorial primary by a razor-thin margin to David Crowley, a moderate who runs Milwaukee County, Wisconsin’s largest municipality.
Many of the polls that forecast blowout victories for Mr. El-Sayed and Ms. Hong were internal surveys that partisan groups selectively released. But independent polls also failed to predict the outcomes, raising questions about how reliable election polls are and whether certain voters are being overlooked.
Problems with polling aren’t new, but the challenges have grown amid an evolving information landscape in which many people no longer own landlines or answer calls from unknown numbers. This electoral cycle even included a purported pollster who admitted to conducting no actual polling after promoting a fraudulent survey – one which Los Angeles Mayor Karen Bass cited in a now-deleted tweet.
Polling misses have been common – and sometimes egregious – during the current primary season. This has a real impact on campaigns, as polls shape media coverage, affect fundraising, and give cues to candidates honing their positions.
Polling misses are particularly common in primaries, which have become increasingly important at a time when most House seats aren’t competitive in the general election. What happened in Wisconsin and Michigan, as well as in Minnesota’s U.S. Senate primary, is what often happens: Polls offered a preelection snapshot that didn’t predict the outcomes of complex, late-breaking races.
The average polling error in statewide primaries is 13 percentage points, according to an analysis by FiftyPlusOne, a political polling website that analyzed 87 Senate and gubernatorial races held in 2014, 2016, and 2024. One in 10 of these races had an error rate of around 20 points. By comparison, polls taken in the two weeks before the 2024 presidential election averaged within 3.3 percentage points of the results, according to the American Association for Public Opinion Research. (In 2016 and 2020, the average error was above five points.)
In primaries, “you’re dealing with much greater unknowns,” says Lee Miringoff, who directs the Marist University Institute for Public Opinion, which conducts surveys for national news organizations. The share of undecided voters is higher, more candidates might be involved, and turnout is harder to model. And in states such as Michigan and Wisconsin, being open to voters regardless of party affiliation “makes it even more treacherous,” he adds.
Tom Jensen, who runs Public Policy Polling in Raleigh, North Carolina, agrees. “It’s much easier to poll in general elections where 90% of people’s preferences are already baked in,” he says. That allows pollsters to concentrate on surveying swing voters and those who don’t vote regularly but can tip close races.
But polling misses in primaries can have a real impact on the races. Polls shape media coverage of candidates and affect their ability to raise money by casting them as viable or not in the eyes of donors. In presidential primaries, they set expectations and determine how results are spun and who has momentum. Candidates also rely on internal polls to hone their positions on issues and build coalitions.
In Maine, where Republican Sen. Susan Collins is seeking a sixth term, Democrat Graham Platner was forced to withdraw in July over sexual assault allegations which emerged after winning his party’s nomination. Earlier in the year, he had been dogged by scandals over his past behavior. Favorable internal polls released by his allies suggested, however, that he remained viable and that his primary opponent at the time, Democratic Gov. Janet Mills, had no path to victory. Critics have noted that a dearth of public polls made it much harder to challenge the spin from his campaign and to scrutinize his candidacy.
“Candidates conduct polls for strategic purposes, and they only release information for strategic purposes,” says Michael Traugott, a professor emeritus at the University of Michigan who studies political campaigns and how they communicate.
This is why we have independent polls, says Dr. Miringoff, a former president of the National Council of Public Polls. News organizations began commissioning preelection polls in the 1950s so that they weren’t reliant on polls released by campaigns. “They felt the public had to have their own source of independent information,” he says.
Today, shrinking news budgets have made that practice more difficult to sustain, particularly for local newspapers and TV stations. Finding voters willing to talk to pollsters is also harder. “Good polling has become very expensive,” says Dr. Miringoff. The void is often filled by online surveys that are seen as less reliable.
Even the most rigorous pollsters struggled to keep up with Wisconsin’s roller-coaster Democratic primary. Ms. Hong, the progressive state lawmaker, held a big lead in virtually every poll taken before the final weeks of the campaign, albeit with a large share of undecided voters. In July, Mr. Crowley dropped out of the race because of his low polling numbers and endorsed another moderate, Lt. Gov. Sara Rodriguez. When her candidacy unraveled, Mr. Crowley jumped back in, even as early voters were already casting ballots for Ms. Rodriguez and other candidates.
Mr. Crowley’s only path to beating Ms. Hong was to mop up all the voters who preferred a more moderate nominee. In the end, that’s what happened, says Mr. Jensen, who conducted a poll in early August for Mr. Crowley that showed him closing the gap among those most likely to cast a ballot. Then “there was a huge surge in the final week when everyone had stopped polling,” he says.
