Polls got it wrong when they showed Hillary Clinton defeating Trump in 2016. But pollsters say surveys showing Biden over Trump are more trustworthy.

WASHINGTON Joe Biden can look at the polls and smile.
Cautiously.
A double-digit advantage in numerous national surveys, solid leads in a number of key battlegrounds, and competitive showings in states Donald Trump carried handily in 2016 suggest the presumptive Democratic nominee is the favorite to win in November.
But the overwhelming majority of polls four years ago indicated Trump would lose as well. So why put much faith in the 2020 polls that show the former vice president consistently on top?
David Burgess of Kittery, Maine, said he stopped believing polls after the 2016 presidential election.
“They predicted Hillary Clinton would win, and she didnt, Burgess said while taking a stroll through downtown Portsmouth, New Hampshire, with his miniature schnauzer, Taavi. “Voters are like an iceberg. (With polls), you just see the tip of the iceberg. You dont see the rest of the iceberg. You dont know who theyre going to vote for.”
As America heads into the final weeks of the presidential campaign, pollsters say they’ve learned lessons from 2016’s failings that they believe will make this campaign season’s polls more accurate. And while they sympathize with voters’ frustrations, they defend their work as needing minor tweaks, not a fundamental overhaul. 
More:Exclusive USA TODAY poll: Biden widens his lead, but Trump keeps the edge on enthusiasm
“The public understandably walked away from 2016 feeling like polls were broken. And there’s some truth to that,” said Courtney Kennedy, director of survey research at the Pew Research Center. “But it’s not the case that 2016 meant that polling writ large doesn’t work anymore.”
Charles Franklin, director of the Marquette Law School poll, one of many that showed Clinton with a lead over Trump, said 2016 is “a reason to be very cautious.”
“It taught us the lesson that there really isn’t safety in numbers because it is possible for a systematic error or change in the last minute of the election to make everybody wrong and that’s what we saw in 2016,” he said.
What went wrong in 2016?
Actually, pollsters got it mostly right four years ago.
They had Clinton winning the popular vote by about 3 percentage points. She won by 2.1 points. And they were right about the outcome in most states. But their research did not capture the full picture of voter sentiment in the upper Midwest that provided Trump with the margin of victory in the Electoral College.
More:Some young Black voters not enthusiastic about a Biden presidency
Polls in Michigan, Pennsylvania and Wisconsin showed Clinton winning consistently in the months leading up to Election Day.
Of 104 published polls that surveyed voters in those three states from August to the election, 101 had Clinton winning, two were tied, and one (in Pennsylvania) showed Trump with a slight lead. Many fell within the margin of error but 15 had Clinton up by double digits at some point. 
Trump ended up winning all three states by whisker-thin margins: a combined 77,744 votes out of 13,940,912 cast, or about half a percentage point. The 46 electoral votes those three states provided Trump the winning margin, stunning those who predicted a Clinton victory. He won 306 electoral votes to Clinton’ 232.
Ronna McDaniel, who chairs the Republican National Committee, dismissed the 2020 polls given what happened four years ago.
More:Bernie Sanders supporters reluctantly turn to Joe Biden, fueled by their dislike of President Trump
“These polls right now are not even significant,” she told FOX Business channel recently. “One hundred and fifty polls were done between now and the election in 2016 that showed Donald Trump losing, and the ultimate poll is Election Day.”
How polls are changing
Both Kennedy and Franklin say two large factors complicated the accuracy of the 2016 polls: Many state surveys tended to over-sample college-educated voters (who favored Clinton) and many failed to capture the late-deciding voters (who generally swung to Trump).
Neither says those problems couldn’t recur this year but they’re hopeful 2020 polls will provide a better yardstick of voter opinions for several reasons:
Educational attainment: Pollsters have been encouraged to increase the sample of non-college graduates, who not only tend to favor Republicans over Democrats but who are also less likely to want to participate in polls.
Patrick Murray who runs the New Jersey-based Monmouth University Polling Institute, which had Clinton winning in its three Pennsylvania polls, said they’ve begun weighing educational attainment more in polls “and were already seeing that impact has helped a bit.”
Late deciders: This may be harder to capture since most polls don’t interview voters in the week before the election. But the number of voters unable to make up their minds until the last minute is expected to be considerably less in 2020 than four years ago, when the historic unpopularity of Trump and Clinton had many hemming and hawing up to the very end.
A Monmouth poll in June indicated that nearly nine in 10 voters had already made up their minds about which candidate they’ll back in November..
More:Young Black voters say they aren’t enthusiastic about a Joe Biden presidency
“Thats a suggestion that theres not a lot of room for movement,” Murray said. “But of course the one reason why we want to be a little careful about taking what were seeing in the polls today and projecting it to November is that there are still certain states where it would take a very small amount of movement to change the outcome.”
Third party candidates: The lack of a prominent independent in the presidential race this year means fewer choices and clearer options.
Pollsters say the candidacy of Libertarian Gary Johnson, who polled at 10% as late as August 2016, probably contributed to the relatively large number of late-breaking deciders since a significant bloc of voters were seeking alternatives to Trump and Clinton. Johnson faded toward the end of the campaign, finishing with less than 3.3% of the vote.
Early voting: The coronavirus pandemic has injected uncertainty into the 2020 election as to who will vote and how. But it is expected to ratchet up the amount of early voting in many states.
If more respondents have cast their ballot when pollsters contact them in late October, it should be easier to gauge true voter sentiment since they will have already made their decision, pollsters say.
Likely voters: Pollsters are slowly but steadily moving to a model using public voter records to identify likely voters rather than a “random-digit dial” system that relies on respondents to report their voting participation patterns.
“People forget that large swaths of our fellow citizens don’t vote. But a lot of them still participate in polls,” said Kennedy, who led a review of the 2016 polling for the American Association for Public Opinion Research. “So there’s a big (gap) that even the best pollsters struggle to model away.”
While changes are in progress, Murray said it’s wise not to dramatically revamp polls that are already largely on target.
“There are some errors that well never be able to account for because they happen idiosyncratically in each election. Theyre different each time,” he said. “So you just want to be careful that you dont over-correct for your last mistake because thats not the one thats going to happen this time around.”
The ‘protest vote’ in 2016
Despite the 2016 stumble, most polls in 2018 showing Democrats retaking the House and Republicans keeping the Senate proved accurate. Even so, pollsters caution that surveys are simply snapshots not predictors and that some are better than others at revealing voters’ deeper attitudes on issues and candidates.
But polls can influence voting behavior.
More:Joe Biden tops Donald Trump in fundraising for second straight month
Brian Schaffner, a Tufts University political science professor, studied how some Bernie Sanders voters in 2016 ended up casting a ballot for Trump. Those progressives might have thought Clinton would make a better president than Trump but they also believed she would win anyway and couldn’t stomach the thought of voting for her, he said.
“It felt OK to cast a protest vote or just not turn out,” Schaffner said. “Some people felt more at liberty to cast a vote that they think wasn’t going to matter.
FiveThirtyEight, which analyzes opinion polling, warns not all polls are the same.
It recommends consumers look at who’s being polled (adults, registered voters, likely voters), check the track record of the pollsters, and pay close attention to the margin of error. A poll that shows Biden up by two points over Trump and has a four-point margin of error means Biden could be up as much as six points or be trailing Trump by two.
Keeping ‘their opinions to themselves’
Ellen Chaput, a nurse who lives in Portsmouth, said she hopes polls showing Biden with a lead over Trump are accurate. But she doesnt believe them.
Theyve got it wrong before,” she said. I dont pay any attention to them.
Even if Biden is ahead right now, things can change, she said.
Helaine Dandrea, a pharmaceutical consultant from Staten Island, New York, said polls often reflect the biases of the people who conduct them.
Dandrea, who is backing Trump for re-election, said polls that show Biden with a solid lead could be underestimating the presidents strength. Some Trump supporters may be unwilling to tell pollsters they are backing the president because they dont want to face the inevitable backlash from the other side, she said.
People tend to be afraid because theres a lot of aggression, she said. People tend to keep their opinions to themselves.
Jim Menard, a retiree from Salisbury, Massachusetts, who backs Biden, said he can see why Trump voters might want to keep their preference to themselves.
More:’Driven by re-election’: John Bolton book accuses Donald Trump of seeking foreign help for political gain
I imagine they are kind of embarrassed to say they support him, Menard said.
Menard suspects polls showing Biden with a comfortable lead are correct because be believes pollsters changed their methodology to get more accurate results, he said.
Theyve gotten better at these polls since the last time, he said.
Important to focus on ‘where things stand today’
Franklin said it’s also important to look beyond the top-line numbers of who’s in front and examine the deeper data that explores why voters feel the way they do.
Marquette’s June poll, for example, shows Biden with an eight-poll lead in Wisconsin: 49%-41% with a margin of error of 4.3 percentage points. That’s up from the three-point lead Biden had in May due mainly to declining support for Trump and growing opposition from independents. 
The June poll shows 50% of Badger State voters approve of Trump’s handling of the economy (down from 54% in May), while 44% approve of his handling of the coronavirus crisis (the same as the previous month). But only 30% said they approved of his handling of protests over the death of George Floyd.
“If there’s a valuable lesson to learn from ’16, it’s to put less weight on what may happen in the unknown future and put more weight on where things stand today,” Franklin said.
More:About that running mate: 72% of Democrats in USA TODAY poll say it’s ‘important’ Biden pick woman of color
Murray, whose Monmouth poll released earlier this month is one of nearly a dozen in June showing Biden with a double-digit lead nationally, said the polling today centers on how voters view Trump because many have yet to learn much about the former vice president. That could change by October as the public gets to know more about Biden, he said.
I think the polls are telling us a story about what’s going on and how people are dug in,” he said. “It doesn’t tell us how the Electoral College is going to turn out right now so thats why you should continue to take the polling with a grain of salt if youre looking ahead to November.”
Just as they did in 2016, polls in 2020 once again show Trump losing in Michigan, Pennsylvania and Wisconsin.
The grain of salt? The same Marquette University Law School Poll that has Biden up by eight points in June showed Clinton up by nine four years ago.