Susan Collins defied the polls. Here’s what they may have gotten wrong.

If you closely followed public polling in the race, U.S. Sen. Susan Collins’ resounding victory on Tuesday may have been a surprise. The margin surprised campaigns and insiders as well.

The Republican won a fifth term by House Speaker Sara Gideon, D-Freeport, leading in every independent and public poll conducted in 2020. Close to Election Day, Collins sat 5.5 percentage points behind Gideon in a RealClearPolitics average and the incumbent had just a 33 percent chance at keeping her seat, according to a Decision Desk HQ model.

But his odds were overshot in polls in Florida and Wisconsin. The Democrat had been favored in both states, but Trump won Florida by 2.5 points while Biden won Wisconsin by less than a point. The polling misses in this Maine U.S. Senate race were among the worst in the country, though surveys are meant as snapshots.

In Maine on Wednesday, pollsters were reflecting on their work after Collins won by eight percentage points after a bruising election in which $130 million flowed into Maine to boost Gideon in campaign contributions and outside spending. Collins got only $76 million.

Theories about the survey misses revolved around late movement of undecided voters and underestimates about the share of ticket-splitting voters in Maine, which was won by both Biden and Collins by nine-point margins statewide. Trump won the 2nd Congressional District by a comfortable eight points as both of Maine’s Democratic U.S. representatives won easily.

Traverse Burnett, an analyst with Digital Research, a Portland firm that partnered with the Bangor Daily News on two polls this year, attributed Collins’ win to undecided voters breaking heavily for the incumbent. There were signs of that, but they were mixed in with red herrings.

For example, the final public poll of the race, which was released on Sunday by Emerson College, found Gideon ahead of Collins by six points with 4 percent of voters undecided. But when those 22 undecided voters were asked who they leaned toward, every single one said Collins. It was a signal in retrospect, though it looked at the time like an aberration.

Collins said throughout the race that her campaign’s figures compiled by longtime pollster Hans Kaiser showed a closer race than public polls did. On Wednesday after her victory speech, she cited the warm reception she received on a bus tour as a source of encouragement, not the polling numbers.

Her campaign did not comment on what it saw at the end, but Senate Republicans’ campaign arm only released two polls in 2020. One showed Collins up by eight points in June and another showed the two tied in September. A Republican strategist familiar with the race said while there were surveys showing both Gideon and Collins slightly ahead, those watching the Maine election were expecting a narrow race likely to come down to ranked-choice voting.

Polls showed Gideon would benefit in that case, since supporters of progressive Lisa Savage, an independent who won 4.9 percent of votes in Tuesday’s election, overwhelmingly favored Gideon over Collins. The senator pondered that scenario in a Monday radio interview alongside former U.S. Rep. Bruce Poliquin, who was ousted in a ranked-choice race in 2018.

On the Democratic side, a source familiar with polling said there were signs of danger for Gideon as figures tightened late in a way that suggested that it was due to a late blitz of Republican attacks against Gideon for her handling of 2018 sexual misconduct allegations against former Rep. Dillon Bates, D-Westbrook, who denied the allegations but resigned. Gideon spokespeople did not respond to a request for comment.

Beyond the issue of late movement in the race, pollsters in Maine and across the country struggled to capture the energy of rural, working-class voters in 2016 and 2020, said Dan Shea, a government professor at Colby College who supervised four Maine polls during the cycle.

He hypothesized voters who are not “hard partisans” and thus more likely to split their tickets could also be less likely to answer polls, though he noted that is just one potential explanation.

“We did not see that coming,” Shea said. “We did not see the extent of split-ticket voting.”