Every week this fall, it seemed like some new poll predicted that voters favored Democrat Sara Gideon over longtime incumbent Republican Susan Collins in the U.S. Senate race by anywhere from 4 to 7 percentage points.

Yet Mainers woke Wednesday to find Collins beating Gideon by almost double digits. By Wednesday afternoon, The Associated Press reported Collins had won and garnered 51 percent of the vote to Gideon’s 42 percent. Most national polls had Joe Biden beating President Trump on average by about 52 to 42 percent before the election Tuesday, while the actual tally Wednesday afternoon had Biden at just over 50 percent and Trump at more than 48.

“I keep asking myself what went wrong?” said Barb Childs, 54, of Waterboro, after seeing election results that were in stark contrast to pre-election polls. “Something needs to change to make the polling more accurate.”

Polls throughout the campaign that showed Democratic challenger Sara Gideon leading Susan Collins “missed the mark.”

For the second U.S. presidential election in a row, voters have seen election results veer dramatically from poll numbers, prompting questions about their value. Some pollsters and election experts say there are so many more polls now – because they draw valuable publicity for the colleges or marketing firms that conduct them – and not all are done with the same thoroughness or the same methods, so the accuracy varies widely.

It also may have been tougher to gauge the voting plans of conservatives over the last two presidential election cycles, as the country has grown more divided. Some surveys have shown conservatives and Trump supporters are less likely to talk politics to anyone, including pollsters, than Biden supporters and Democrats, for fear of the reaction they might get, according to several political science professors and pollsters interviewed for this story.

The University of New Hampshire Survey Center did one poll this fall that found that more than 60 percent of Trump supporters would not put a Trump bumper sticker on their car or sign on their lawn for fear of vandalism. Only about 35 percent of Biden supporters said the same thing, according to the survey results.

“We live in a world where somebody who says they support Trump may get called a racist or a sexist, and that makes a lot of people not want to say who they’re voting for,” said Lonna Atkeson, a political science professor at the University of New Mexico who focuses on elections and public opinion.

Trump supporters and conservatives may also be less likely to participate in polls because of a distrust of the institutions, including colleges, that conduct them and of the media that so widely report on them, said Ken Cosgrove, who teaches political marketing at Suffolk University in Boston. Cosgrove said political scientists often cite the “Bradley Effect,” named for former Los Angeles Mayor Tom Bradley, who got much higher poll numbers than actual votes because voters thought it was more “socially acceptable” to say that they were voting for him, Cosgrove said. The inverse seems to be happening with Trump and other Republican candidates, where people might be deciding the socially acceptable thing to do is not to disclose their support.

“The culture we live in now is so emotional, disrespectful, that a lot of people might say, ‘I’m going to do what I want to do, vote for who I want, but I’m not going to say so in public,’ ” said Cosgrove.

WHAT WENT WRONG?

An ABC-Washington Post poll that found a 17-point edge for Biden in Wisconsin wasn’t even close. But the race sure was.  Associated Press photos

Among the most off-base polls this fall were an ABC-Washington Post poll that gave Biden a 17-point lead in Wisconsin, a state that he won by less than 1 percent on Wednesday. A poll by Quinnipiac University in Connecticut gave Biden a five-point lead in Florida and a four-point lead in Ohio, states that Trump has won.

“A full examination of what went wrong with polls this year is going to take awhile. At the moment, I still need to see the final election results and final exit poll results, and without those, I’m not able to make even preliminary hypotheses about what exactly the issues are,” Doug Schwartz, director of the Quinnipiac University Poll, said in an email to the Press Herald.

In Maine, a poll by Portland-based Pan Atlantic Research found Gideon with a 7-point lead over Collins, after surveying 600 Maine voters online between Oct. 2 and Oct. 6. The results of a Colby College survey of 879 people released in late October showed Gideon with a 3-point lead over Collins. It also showed Biden with a 51-38 percent advantage over Trump in Maine. As of Wednesday afternoon, Biden was leading Trump 53-44 in Maine. The poll was conducted Oct. 21-25, over the phone and online.

“We missed the mark, but I think it’s important to remember, a poll is a snapshot of a particular point in time, campaigns evolve and turnout matters,” said Dan Shea, chair of the Colby College Department of Government and lead researcher on the poll. “I think our poll, as well as others across the country, failed to capture the passion of rural voters.”

Shea said, for instance, he was surprised to see how large Collins’ margins of victory were Wednesday in small towns in very rural areas. He said polling and other data indicated Collins might win by 2-1 margins in places in Aroostook County, but instead she won by 3-1 or more in some. That could indicate the pollsters needed to find more rural voters to poll in certain areas.

Shea also said his poll and others might not have taken into account a factor that became apparent after the election – that Maine is bucking a national trend toward party-line voting. He said political experts have believed for a few years that “split-ticket voters,” including people who vote for candidates from different parties for president and senate, did not exist in large numbers anymore. But in Maine, some areas that voted for Biden also voted for Collins in significant numbers, including Lewiston-Auburn and Augusta, Shea said.

There is no one way that a political poll is conducted, and nobody is in charge of regulating them or assuring quality control for all of them. So many different types of organizations do political polls in many different ways. Many do them because they attract publicity for their year-round business doing marketing surveys or polls for clients, said Andrew Smith, the director of the Survey Center at the University of New Hampshire in Durham.

Campaigns and candidates use polling to gauge likely votes, but those polls are usually more focused than the ones put out publicly through the media. The public first became aware of political polling because of a firm called Gallup, which began doing polls in the 1930s. But Gallup was, and is, a company that makes money by polling for clients on a wide variety of topics. Many companies today do political surveys among many other types of paid research.

One method often used by pollsters today includes online surveys, especially ones that recruit people from gaming or shopping websites and offer cash or other incentives for participating, Smith said. Some pollsters call people on landlines, which almost guarantees they’ll get older people. Some use computer-generated “robocalls” to ask people simple yes or no questions. Some buy lists of cellphone numbers from various sources. Some use targeted online panels, assembled by research companies.

“There are certainly a lot of polls that were way off, but I think you can find some that were pretty close to what happened,” said Smith.

One poll that came closer to predicting the Collins-Gideon race than some was one done by the Portland-based firm Digital Research for the Bangor Daily News. The survey of 466 likely voters was done between Sept. 24 and Oct. 4 and showed 44 percent of the respondents planning to vote for Gideon and 43 percent for Collins. As of Wednesday afternoon, Gideon had 42 percent, while Collins had 51 percent. That poll also had Biden ahead of Trump in Maine by 51 to 40 percent, when the numbers Wednesday were 53-44.

Digital Research had also done a poll in August showing Gideon with a 43-35 percent lead and 14 percent of respondents listed as  “not sure,” said Traverse Burnett, research director for Digital Research. But their later survey showed just 7 percent not sure, and it appears that most of the “not sure” voters went for Collins, Burnett said.

Smith thinks the political polling industry is at a crossroads, as groups try to figure out the best way to get accurate information from voters. He said pollsters had similar challenges when transitioning from in-person polling to all-phone polling. Now, the challenge is finding the best way to use a combination of polling on cellphones and online, or some other method.

The cost of polling keeps going up, which makes some pollsters look for cheaper and less effective methods, Smith said. He said a statewide survey of New Hampshire in 2016 cost his center around $25,000, but in 2020, it cost about $75,000, because fewer people today answer calls from numbers they don’t know, so pollsters have to call a lot more people and labor costs go up, Smith said.

The proliferation of polls by different companies and organizations, with varying backgrounds in polling, will likely continue as long as they continue to draw attention.

“Why do these polls? Because they earn media attention for your company or organization,” said Atkeson, the University of New Mexico professor. “Polls are interesting and easy for the media to report, but they don’t get at policy or things that really matter to voters. They’re of very little value to voters.”