They know they screwed up. Pollsters have a few ideas why.
It’s possible Donald Trump’s upset victory this week was powered by a surge of late deciders. Or the mysterious group often referred as “shy Trump” voters somehow escaped their radar. Many in the polling industry are also second-guessing their turnout modeling, trying to discern whether there’s a serious flaw that went unnoticed.
No matter the root cause, an industry already reeling from a series of misses in the United States and overseas is engaging is a round of serious introspection. While the data streams required to evaluate whether they modeled the electorate incorrectly — or whether Trump voters disproportionately wouldn’t respond to polls — won’t be available for months, already the nation’s leading professional organization of pollsters is admitting they “clearly got it wrong this time” and pledging to study the causes of the errors.
“It seems like the catastrophic polling error that we’ve been fearing for decades,” said Jon Cohen, the vice president of survey research at SurveyMonkey and a former pollster for The Washington Post and the Pew Research Center. “But it may prove to be less than that.”
The polls underestimated Trump – most acutely in a number of battleground states viewed as leaning in Hillary Clinton’s direction – much as they systematically underestimated Republicans in the 2014 midterm elections. But the polls were off in the other direction in 2012, with national surveys understating President Barack Obama’s margin of victory by about 3 points.
Overseas, the polls badly missed the 2015 election in Great Britain. The polls were closer in Britain’s vote to leave the European Union this year, though they are often blamed to a greater degree because the result stunned so many observers.
The pre-election polls in this year’s presidential election could actually end up closer to the actual result than the polls four years ago, at least on the national level. But there are no laurels for that. Trump won the Electoral College and not Clinton, so it’s viewed as a far more significant polling malfunction.
Even if pollsters don’t yet know precisely what went wrong, they are asking the questions — and cautioning against downplaying the extent of the breakdown this week.
“The pollsters who are out there saying the polls are really okay because they all fall within the margin of error, that is just not credible — and not helpful,” said Democratic pollster Jefrey Pollock, whose firm, Global Strategy Group, worked for the pro-Clinton group Priorities USA.
“It is a mistake to go out and say polling is useless,” Pollock added. “It is just as big a mistake to claim the polls are all okay because they all fall within the margin of error.”
The potential causes for error run the gamut from a secret army of Trump voters lying to pollsters, to a late-breaking wave of voters flocking to Trump after most of the polling had concluded.
Here are the questions pollsters are asking this week:
Did Trump surge at the end?
There’s some limited evidence that’s the case.
National polls certainly closed in the last few weeks of the race — but, in the final days, it appeared Clinton had arrested Trump’s momentum.
Indeed, the national exit poll suggests she did: Trump led by 10 points among voters who said they made up their minds in the final month of the campaign, but his lead among voters who decided in the last week was only 5 points. (Clinton led by 5 points among voters who decided before October, which comprised the majority of the electorate.)
In some of the states where Trump won unexpectedly, however, Trump won more late-deciders. In Wisconsin, where the latest results show Trump ahead by a point, Trump overwhelmingly won voters who made up their minds in the final week, 59 percent to 30 percent. In Pennsylvania, where Trump is also ahead by a point, he won last-week deciders, 54 percent to 37 percent. In Michigan — which hasn’t yet been called with Trump ahead by three-tenths of a percentage point — Trump won those who decided in the final week, 52 percent to 37 percent.
That kind of late movement can thwart pollsters who conduct surveys in the weeks leading up to an election.
“If the accuracy of your final forecast is the most important thing, then there is an incentive to poll right up until the last minute,” said Charles Franklin, who conducts the Marquette Law School poll in Wisconsin.
Franklin, a member of the American Association for Public Opinion Research’s task force that will examine the election polls, said he concluded polling in Wisconsin on October 31, more than a week before Election Day.
“I made a policy decision when we started this that we would release [the final poll] the Wednesday before the election, in part so we could give the campaigns a chance to react,” Franklin said. (That Marquette Law School poll had Clinton ahead by 6 points.)
In fact, the final poll in Wisconsin from any nonpartisan outlet was conducted fully a week before Election Day, leaving no instrument to capture if there was last-minute movement.
If that movement did occur late, however, it might make sense that it would be against Clinton, who was running for a third-consecutive Democratic term.
Republican pollster Dan Judy pointed to Clinton’s vote shares in a number of states, relative to where she finished in the final polling averages. The result? She only scored a point or so higher in most states on Election Day — suggesting undecided and some voters choosing third-party candidates in the pre-election polls drifted to Trump in larger numbers.
Clinton was at 46 percent in the Wisconsin polls, according to HuffPost Pollster. She won just shy of 47 percent of the vote. She was at 46 percent in the Pennsylvania polls and won less than 48 percent of the vote. In Michigan, she was at 47 percent in the polls and won that percentage on Election Day.
That reflects an old rule of politics, Judy said — one that hasn’t always applied in the past: Undecideds vote against the incumbent in the end.
“The swing states and ‘Blue Wall’ states that Trump won definitely treated Hillary like an incumbent,” Judy said. “It’s remarkable how close her actual percent was to her final polling average across a lot of the most competitive states. She really was polling like an unpopular incumbent — maybe not surprising given the run for a third term, and Obama’s aggressive campaigning for her down the stretch.”
That was also evident in a key number from the exit poll: the vote preferences of the roughly one-in-five voters who had an unfavorable opinion of both candidates. Among those voters, who made up a historically high 20 percent of the electorate, Trump won by 21 points: 50 percent to 29 percent.
Those voters, who could have been among the last to decide, helped propel Trump despite the fact that more voters viewed Clinton favorably (42 percent) than had a favorable opinion of Trump (39 percent).
“It looks like the biggest chunk in the end chose to take the dive with Donald Trump,” said Joe Lenski, the researcher who oversees the exit poll for Edison Research. “That’s the only group that puts him over. In any other election, if you just had the two favorable numbers, you would say the person with the highest favorable number would win.”
Were there “shy Trump” supporters?
It was a theory POLITICO sought to test in late October, along with the online pollster Morning Consult: Were there people who voted for Trump but wouldn’t admit it to a pollster?
The study found only a slight impact by moving poll respondents from the internet to a phone call with a live interviewer — with larger effects among college-educated white voters.
Perhaps that happened on Election Day: Exit polls — which are an imperfect but immediately available record of who voted, for whom they voted and what they thought about candidates and issues — indicate greater support for Trump among college-educated white voters than the pre-election polls suggested.
“The discussion was all about white non-college men and women,” Cohen said. “But it’s the white college constituency that look dramatically different when you look at the pre-election polls versus the exit polls.”
But it might not be because voters are lying to pollsters — telling them they won’t support Trump but voting for him on Election Day. It might be because they don’t pick up the phone in the first place.
“Differential turnout and participation in surveys is maybe the more worrisome” factor, said Franklin, the Marquette Law School pollster. “If there was some percentage of Trump supporters that refused to do any polling but did go to the polling place, then we’re missing them completely in our samples.”
Who is a likely voter?
Election pollsters are always trying to identify a universe of people that doesn’t yet exist: the future electorate.
Every pollster does that differently: Some allow every voter into their poll that says they are certain to vote. Some make assumptions about who will turn out, including party identification or registration. And most campaign pollsters add what they consider is the most important factor: whether voters have turned out before.
Trump appears to have upended some of those approaches. But it will be months before pollsters know how it happened.
The first clues are trickling in now as the votes are tallied. Once all the votes are tabulated, pollsters will know whether turnout was greater or lower than expected — and, most importantly, where those trends apply.
Was turnout markedly lower in urban centers Clinton needed, like Milwaukee, Philadelphia and Detroit? Early indications, especially in Milwaukee and Detroit, indicate drops in turnout.
Meanwhile, turnout appears higher outside cities and suburbs.
“If you look at these states, and you look at the turnout ratios from four years ago county-by-county, it’s pretty clear the biggest percentage increases in turnout were in the non-urban, non-suburban areas,” said Lenski, who administered the exit polls. “That’s hard to predict both in a pre-election poll and an exit poll.”
In Pennsylvania, for example, Clinton carried more votes out of Philadelphia and the suburbs than President Barack Obama in 2012 — but polls missed the higher turnout in more rural areas of the state.
“If you just showed me Hillary’s numbers in the Southeast, I would have said she would have won by 2 or 3 points,” said Christopher Borick of Muhlenberg College in Allentown. (Muhlenberg’s final poll had Clinton ahead by 6 points in a head-to-head matchup with Trump, and 4 points in a four-way matchup.)
But a complex analysis of the electorate — beyond just from where the votes came — will take months. Pollsters will be able to look at precisely who voted — whether they were regular voters or less-frequent voters drawn out by Trump’s unique candidacy — and who didn’t.
“That will give us the best evidence about new voters, about previous voters who dropped out,” said Franklin. “That will be incredibly valuable.”
For pollsters trying to figure out what happened this week, those voter files — in addition to next year’s Census Current Population Survey — will be worth the wait. Trump’s candidacy rocked the political system, from the Republican primary through the general election. And a Trump presidency could upend how Americans view and interact with their government in a similar way.
“We’ve reacted well to failure before,” said Cohen, the SurveyMonkey pollster. “Polling is too important to go away. The way that we are going to understand what happened in the election, and the contours of where we sit as a country, is through polling.”
Powered by WPeMatico