For months, polls indicated that Francesca Hong, a democratic socialist, was far and away the front-runner in the Democratic primary for governor in Wisconsin.

Then David Crowley, a moderate and the chief executive of Milwaukee County, won the race, albeit narrowly.

The result comes just one week after Dr. Abdul El-Sayed, a progressive, significantly underperformed polls to eke out a narrow victory in Michigan’s Democratic Senate primary.

Why were the polls so wrong?

Possible explanations range from a genuine, last-minute change in voter sentiment — Ms. Hong, a state representative, faced significant negative publicity in the closing days of the race — to miscalculations by pollsters about the makeup of the primary electorate.

The disparity between polls and results in Wisconsin and Michigan prompted Frank Luntz, a Republican pollster and consultant, to jokingly call for “a total and complete shutdown of primary polling in the Midwestern United States until we can figure out what the hell is going on” in an early-morning X post on Wednesday.

The Wisconsin primary drew a crowded field of candidates after Gov. Tony Evers, a Democrat, announced last summer that he would not seek re-election. Here’s what experts, pollsters and the data say about the results of the Democratic primary to replace him:

What the polls said

The Wisconsin governor’s race was tumultuous. Mr. Crowley had suspended his campaign just over a month ago after polls indicated he had failed to gain traction. Then he re-entered the contest and won.

“The polls were made out of date very quickly” as a result, said Celinda Lake, a Democratic pollster from Michigan.

In a survey taken by Marquette Law School just after Mr. Crowley re-entered the race in mid-July, 38 percent of respondents said they favored Ms. Hong. Mandela Barnes, the former lieutenant governor, was in second place at 16 percent, and Mr. Crowley had just 7 percent of the vote.

But almost as large as Ms. Hong’s vote share was the percentage of undecided survey respondents: 34 percent.

Mr. Barnes dropped out after that poll was released.

In the weeks that followed, surveys continued to show voter support was fractured among Ms. Hong’s numerous opponents.

As recently as last week, a poll conducted by Mr. Crowley’s campaign found him trailing Ms. Hong by 18 percentage points. He trailed Ms. Hong by 22 points in another poll from State Navigate, which surveys in state-level elections.

What actually happened on Election Day

Polls failed to predict the outcome of Tuesday’s election, but they did not significantly overestimate Ms. Hong’s portion of the vote. Most surveys estimated she would win about 40 percent of the vote, and she ended up winning 39 percent.

What changed was Mr. Crowley’s support. In a ballot-counting marathon that stretched into the wee hours of Wednesday morning, he over-performed the Marquette poll’s estimate by more than 30 points.

In the end, Mr. Crowley was less than half of a percentage point — or just under 4,000 votes — ahead of Ms. Hong on Wednesday morning, with more than 95 percent of ballots counted.

Many primary voters tune in at the last minute

Mr. Crowley may owe much of his success to the support of previously undecided voters.

In primaries, news coverage and campaign advertising tend to increase in the last weeks of the race. And voters may not research candidates until minutes before they walk into the voting booth. In some Wisconsin polls, as many as one-third of voters said they were undecided heading into Election Day.

“It is well known that voters in Wisconsin tend to make up their minds, tend to start paying attention to primary elections pretty late in the game,” said Anthony Chergosky, an associate professor of political science at the University of Wisconsin-La Crosse.

Ms. Hong faced a barrage of negative press in the election’s final days, after six-year-old social media posts surfaced in which she called to defund and abolish the police and “cancel Thanksgiving.”

Mr. Crowley was not as well-known to voters before he re-entered the race in July. Mr. Evers campaigned energetically for him, and a super PAC supporting him also spent heavily on ads disparaging Ms. Hong in recent weeks.

The governor’s late endorsement made the race unusual, said Charles Franklin, who conducts the Marquette University Law School poll.

“In our final poll, 36 percent of Democratic primary voters said that endorsement made them more likely to support Crowley,” said Mr. Franklin. “Evers’s job approval with Democrats was also over 85 percent.”

Mr. Crowley may also have consolidated support from moderates after the early establishment-backed candidate, Lt. Gov. Sara Rodriguez, and Mr. Barnes ended their campaigns last month.

Polls expected more young voters to show up

Even after an internal poll found that Mr. Crowley was trailing Ms. Hong last week, his campaign manager predicted that polls would not reflect the primary results.

In a memo to supporters on Aug. 5 obtained by The New York Times, the campaign manager, Garren Randolph, argued that Wisconsin surveys overrepresented voters between the ages of 18 and 45. In some cases, polls expected twice as many young voters as actually showed up to vote in 2018 and 2022. Ms. Hong was strongest among young voters, according to the internal poll.

Across elections, primary voters tend to be older, and Wisconsin is no exception. Older voters and moderates were also more likely to remain undecided until close to the end, Mr. Franklin said, based on his polling data.

Hong’s supporters may have been more eager to respond to polls

Chaz Nuttycombe, executive director of the firm State Navigate, which conducted polls in Wisconsin, suggested that Ms. Hong’s early and enthusiastic supporters may simply have been more willing to respond to polls, thus skewing the results.

“One candidate’s supporters were just more eager to answer the poll,” Mr. Nuttycombe said.

Pollsters have long grappled with what is known as nonresponse bias, where certain voters may be more likely — or less likely — to respond to a poll.

The same issue may have occurred in Michigan. In a post-mortem of that state’s Democratic Senate primary, Adam Carlson, a Democratic pollster, speculated that the “main driving factor” behind the polling miss was selective responsiveness of voters.

“El-Sayed’s supporters were more fired up and willing to take polls down the stretch, but more enthusiasm does not necessarily equal more votes,” Mr. Carlson, the founding partner of Zenith Research, wrote in a post-mortem analysis on Monday.

Republicans and independents were wild cards

Conducting surveys in partisan primaries has long been a challenge for pollsters, requiring them to make assumptions about the partisan composition of who will actually turn out to vote.

That is particularly difficult in states like Wisconsin and Michigan that hold open primaries and where voters are not registered by party.

As pollsters attempt to target likely Democratic primary voters in surveys, they are often unsure about the percentage of Republicans and independents who might vote in a given election.

Mr. Nuttycombe said polls likely underestimated the Republican share of the electorate.

“If Republicans did vote, they went overwhelmingly anti-Hong,” he said.

Mr. Carlson noted that in the Michigan Senate primary his likely voter model — which pollsters use to estimate who will actually vote in an election — probably underrepresented seniors, low-income Michiganders and self-identified Republicans.

Julie Bosman contributed reporting.