Can we trust the polls? Experts weigh in ahead of the 2026 midterms
With the 2026 midterms less than three weeks away, news outlets and election forecasters have been disseminating the latest polling data at a rapid clip. How much faith should be placed in these early forecasts?

Since the 2016 presidential election upended formal predictions, election-watchers and paranoid partisans have grown more suspicious of national pollsters.
Response rates have declined substantially in recent decades, according to the Pew Research Center, a nonpartisan research organization that studies public opinion and social trends. Those who do participate may not accurately represent the diversity of the electorate, the findings show, while respondents are increasingly less candid about their political views — and others opt out entirely.
With the 2026 midterms less than three weeks away, it’s been all about the polls, with news outlets and election forecasters disseminating the latest data sometimes almost like gospel.
But in an era when elections are won and lost on extremely thin margins, how much faith should be placed in these early predictions?
It depends on the circumstances and who you ask, experts say. To some, they provide a fast-moving data point in a noisy political landscape; to others, they are an irritating fixation distracting from the issues that matter most.
The polls got it wrong in 2016 when election forecasters put Democratic hopeful Hilary Clinton’s chance of winning the nation’s top job “at anywhere from 70% to as high as 99%” over President Donald J. Trump, according to a post election report by the Pew. She was also a heavy favorite to win states like Pennsylvania and Wisconsin that Trump ultimately won.
The results prompted a reckoning in the polling industry over how surveys are conducted, how their results are interpreted and how confidently forecasters predict election outcomes.
“Really the challenge is trying to figure out who to poll and who is likely to show up,” said Katherine Haenschen, an associate professor in Northeastern University’s departments of communication studies and political science.
Haenschen contends that the “polling misses” in recent years have stemmed from pollsters failing to reliably predict who ultimately turned out to vote on Election Day. In 2016, for example, pollsters struggled to adequately account for white voters without college degrees, a group that helped propel Trump to victory, election-watchers have said.
Post-mortems pointed to methodological concerns, such as “nonresponse bias,” which refers to the possibility that people who participate in a poll have different political views from those who don’t, potentially skewing the results. Observers coined the phrase “shy Trumpers” to refer to voters who privately supported Donald Trump but were reluctant to tell pollsters so. A POLITICO/Morning Consult study suggested the phenomenon didn’t really exist.
Still, many election observers maintain that the public opinion polling methods more or less capture public sentiment with accuracy — and within a margin of error.
“All the polls were predicting a very close election regardless, and they were wrong by a couple points here or there,” Nick Beauchamp, an associate professor of political science at Northeastern University, said of Trump’s narrow 2016 victory.
The American Association for Public Opinion Research, a professional association for survey researchers and polling experts, has reviewed polling performance across many election cycles. Mary Losch, the group’s president, said the broader takeaway is that pre-election polls can offer a useful snapshot of a race, but their ability to predict the outcome diminishes when candidates are separated by only a few points.
“For close congressional races … even relatively small polling errors can change who appears to be ahead,” Losch told Northeastern Global News. “Polls are often very good at identifying whether a race is competitive, but distinguishing between candidates separated by only a point or two is inherently more difficult.”
Losch said that there is “no single methodological approach that consistently produced more accurate” predictions, but noted that some approaches were “somewhat more accurate” on average. They include higher-volume pollsters, those weighting for party identification, those using detailed likely-voter models and Republican-affiliated pollsters, she said.
Still, a string of polling errors in the wake of the 2016 election have continued to undermine faith in the process.
In 2024, the Des Moines Register’s final Iowa poll, conducted by veteran pollster Ann Selzer, showed Democratic nominee Kamala Harris leading Trump by 3 percentage points in a state Trump ultimately won by around 13 points. The result stunned observers, given Selzer’s reputation for accuracy, and renewed concerns over an industry that some political observers say can’t quite seem to reliably capture Trump’s political appeal as it actually exists.
Northeastern journalism professor Dan Kennedy said the media has become too attached to polling and the political “horse race” it attempts to forecast — often at the expense of reporting on the issues and the candidates themselves.
Studies have shown that the rise of polling aggregators like Nate Silver’s FiveThirtyEight, a nonpartisan, data-driven political site that is no longer in operation, and RealClearPolitics, an election aggregator and media company, has led to a greater reliance by journalists on statistical models and election forecasts, even as debate over the limitations of certain models intensifies. Critics worry that this emphasis comes at the expense of scrutinizing individual polls and reporting on the issues that shape voters’ decisions.
“We should be reporting on the genuine desires of the public and which candidate is speaking best to those desires — or, rather than making that value judgment, how candidates are responding to those desires,” Kennedy said.
He added that experts have long warned that polling could become less reliable as people moved to cell phones, ditched their landlines and “became increasingly reluctant to respond to pollsters.”
The last few election cycles have also seen the introduction of prediction markets, which add another dimension to the business of forecasting elections. Platforms like Polymarket and Kalshi are setting odds on midterm races in prediction markets that experts expect to closely track traditional polling.
“I would expect them to line up closely with polls like they always do, if only because 95% of the market info comes from the polls in the first place,” Beauchamp said.













