As the primary season hit its final stretch in August, a flurry of high-profile misses threatened to undercut public confidence in polling going into the midterm elections.
The polls were off by 10 to 25 percentage points in high-profile primaries in Wisconsin, Michigan and Minnesota, raising sharp questions about poll credibility and compounding voter distrust that has lingered since misses in the 2016 and 2020 presidential elections.
Pollsters, however, are standing firm in their belief that blunders in primary polls are no reason to believe surveys should be distrusted as the general election in November nears.
The primary misses “have zero implications for the general election — like, literally zero,” said Tom Jensen, the director of Public Policy Polling, a Democratic polling firm. “It’s just a completely different election with completely different dynamics.”
To understand why these pollsters remain confident, it helps to understand why primary polling is so difficult — and why general election polls have typically been far more reliable.
Primary polling has always been difficult.
Not every primary poll misses the mark. In Massachusetts, for example, polls in the final four weeks of this year’s Democratic Senate primary were off by only a single percentage point on average. But a race like that is the exception, not the rule. Between 2016 and 2024, primary polling was off by an average of eight to 11 percentage points in each of the major election years.
The 2026 results were in line with those recent years, with primary polls off by an average of 10 percentage points in the month before the elections.
Big misses in primary polling have not led to similar misses in general elections, however. In 2024, primary polls were off by 11 percentage points — their worst year since at least 2016 — while general election polls were off by three. Missing by a few points on average is considered normal for polls, which carry asterisks in the form of margins of error.
“Primary polling is difficult because several things can go wrong at once,” said Jason Corley, a founder of the right-leaning polling firm Quantus Insights. “We can model the wrong electorate, reach systematically different voters from those who ultimately participate, miss late movement or get some combination of all three.”
Primary electorates are hard to predict.
To measure how candidates are faring in a race, pollsters make decisions about who is likely to vote. Primary turnout is inherently harder to model than general election turnout because of the smaller number of voters. In 2022, primaries saw around 20 percent turnout, while more than 45 percent of the eligible electorate voted that November.
Turnout also varies widely. In the 2025 New York City mayoral primary, nearly all of the polls showed former Gov. Andrew Cuomo in the lead, but Zohran Mamdani won thanks in large part to young voters, who turned out at unexpectedly high rates.
Even with accurate turnout models, pollsters may face “nonresponse bias,” where one candidate’s voters are more willing to take a poll than others. A pollster can correctly estimate, for example, that highly educated white voters will make up a quarter of the primary electorate. But if the voters who respond to the poll lean more toward one candidate than those who do not respond, the poll will show that group’s preference skewing toward that candidate, leading to a flawed topline result.
“You can get the electorate broadly right on paper and still have a big problem if the right mix of voters aren’t answering the survey,” said Ryan O’Donnell, the executive director at Data for Progress, a left-leaning polling organization.
Primaries are volatile.
Primaries also differ from general elections because the races themselves can change dramatically in the final weeks.
Candidates — even high-profile ones — can drop out at random times in the cycle. In Wisconsin, Francesca Hong built large leads in the polls to become the Democratic nominee for governor as two of the most prominent candidates, Mandela Barnes and Lt. Gov. Sara Rodriguez, exited the race in the final month of the election. David Crowley, another major candidate, suspended his campaign in early July only to re-enter the race less than two weeks later.
The polls conducted during this upheaval were measuring entirely different fields from the one that existed on Election Day. Even the two polls conducted after the field was settled both showed Mr. Crowley well behind Ms. Hong. In the end, Mr. Crowley consolidated support from various groups and narrowly won the race, with surveys fielded during the last three months of the primary off by more than 25 percentage points on average.
Even in elections with far more stability, other factors, like presidential endorsements, can significantly change how voters view a race.
Undecided voters also drive a great deal of the volatility. On average, 18 percent of voters were undecided in the month before the primary races this year, among polls that explicitly reported this share. That leaves plenty of room for voters to consolidate behind a candidate.
General election polls are more sound.
With November now the focus, general election polls, which have historically been less prone to error, are taking center stage. The errors in these elections tend to be smaller and more predictable than the ones in primaries, even as there have been some notable polling misses in general elections.
One reason is that general elections have far fewer persuadable voters, with many already knowing which party they will support well before they’re tuned in to the individual candidates. In the 2024 Senate races in Michigan and Wisconsin, just 5 percent of voters, on average, were undecided in polls fielded in the month before the election, and the polls, on average, were within three percentage points of the final outcome.
General elections also tend to feature more polls from high-quality pollsters that are not connected to partisan organizations, while primaries are often dominated by internal polls. Those internal polls, which are occasionally the best information available for a primary, are often released by a sponsor — usually a campaign or an affiliated group — to promote favorable narratives for specific candidates. In 2026, nearly 45 percent of primary election polls were internal ones, while they currently make up less than 30 percent of general election polls.
In 2026, major independent pollsters like The New York Times/Siena Poll and CNN/SSRS did not conduct any polls for primary races. Instead, those organizations have already released several polls for general election contests.
A more stable electorate, paired with higher-quality surveys, makes huge misses — like those recently seen in Wisconsin and Michigan — far less common in general elections.
“There’s not going to be a 13-point miss or a 20-point miss for the generic ballot in the general election in November,” said Chaz Nuttycombe, the executive director of State Navigate.
Pollsters have something to prove.
In light of the primary misses, many pollsters are taking stock as they look to address skepticism about the effectiveness of public polling.
“If you have a bad poll, you have to understand why these things happen so that we can avoid it,” said Spencer Kimball, the director of polling at Emerson College.
Mr. Kimball said he often reviews Emerson’s decisions after high-profile misses, like last year’s poll for the New Jersey governor’s race, which underestimated the support for Mikie Sherrill. He said he would conduct a similar review after Emerson’s poll of the Democratic primary in Michigan greatly overstated Dr. Abdul El-Sayed’s lead over Representative Haley Stevens.
The goal, for Emerson, is simple.
“I don’t care who wins — I don’t even know their names,” Mr. Kimball said. “All I want is for the poll to be accurate.”
The post Primary Polls Had a Bad Year. What Does That Mean for November? appeared first on New York Times.
