Pollsters go back to drawing board

Pollsters who missed Donald TrumpDonald John TrumpNFL freezes policy barring players from protesting during anthem McConnell spokesman on Putin visit: 'There is no invitation from Congress' Petition urges University of Virginia not to hire Marc Short MORE’s surprise electoral victory are headed back to the drawing board.

Most polls predicted a Hillary ClintonHillary Diane Rodham ClintonProminent Putin critic: If Trump turns me over, I'm dead Dems unveil slate of measures to ratchet up pressure on Russia Trump tweets old video of Clinton talking up 'a strong Russia' MORE victory on Election Day, with some positively bullish on the likelihood of a Clinton White House.

ADVERTISEMENT

The final poll from the Reuters State of the Nation project put the odds at 90 percent that Clinton, the Democratic nominee, would win; The Huffington Post’s model predicted a 98.2 percent chance of Clinton winning; and The New York Times’s Upshot put them at 85 percent.

Both thought that Clinton would win the states of Pennsylvania, Michigan and Wisconsin.

The RealClearPolitics average of polls one day before the election had Clinton ahead by 3.2 percentage points over Trump, the Republican nominee.

The results put egg on the polling industry’s face and left it searching for answers. Clinton did win the popular vote; she held less than a percentage point lead over Trump at press time.

But she lost Pennsylvania, Michigan and Wisconsin, which few if any pollsters had forecast, while also losing the states of North Carolina and Florida. She was blown out in Ohio.

Pollsters note that their surveys weren’t a complete whiff.

National polls predicted a low-single digit Clinton popular-vote victory.

The majority of swing states also ended up in line with the margin of error in most polls. 

But many of the highest-profile models fell short, as did many of the polls used as the backbone for the most bullish predictions. 

FiveThirtyEight pollster Nate Silver, who preached a healthy skepticism about the models that predicted a certain Clinton election, had Clinton with a 71 percent chance of winning. 

Just one polling company, the Republican-leaning Trafalgar Group, polled Trump ahead in Michigan and Pennsylvania. Besides that poll, conducted days before the election, not one other poll since August had Trump ahead of Clinton. 

In Wisconsin, Clinton swept all 19 polls conducted since June, according to RealClearPolitics. 

Democratic pollster Celinda Lake blamed a “herd mentality” for some of the problems.

“In 2012, we had national polls that were off more [than 2016]. But they were off in different directions, so there wasn’t that herd mentality,” she said. 

Pollsters also say they erred with their predictions of who would come out to vote on Election Day.

Exit polls from 2012 found that 38 percent of voters were Democrats, compared to 32 percent who were Republicans. 

Most pollsters anticipated a similar electorate in 2016. But that’s not what happened.

Instead, voter turnout dropped on both sides. Exit polls showed Democrats with about a 2-point lead in terms of who showed up on Election Day, compared to the 6-point edge they held in 2012.

Lake said another possible problem was the historic unpopularity of Clinton and Trump.

She noted that 18 percent of voters didn’t like either candidate. 

Those voters historically stay at home, split their vote or back the candidate viewed as most qualified. 

This year, a good portion broke for Trump, hurting Clinton.

To do better in 2020, pollsters told The Hill they need to adopt a “mixed-mode” approach that embraces a host of different polling methods and outside data, and that would ensure people are reached not just by landlines but through smartphones and online.  

John Zogby, an early adopter of online polls, cast the skepticism of online polls as unwise, noting that internet access among likely voters is near the level of landline telephone access at its peak. 

“There has to be a greater appreciation for the technology and a greater understanding of social changes,” he said. 

“When I started in 1984, two out of three people you got on the phone were willing to take a survey, and one of those were eager,” he said.

There’s also a push to study accurate ways to include external data sources like social media conversation as yet another metric. 

“All these things can tend to make the polls more accurate, as well as bringing outside sources of data in to give us a little more texture,” Republican pollster Patrick Ruffini said.  

“We’re consuming too much horse-race polling and not enough of what is the mood of the country and what voters are actually thinking. … It’s not a replacement for polls but it improves the overall prediction when you throw it into the model.”

All these changes will be more expensive — which could be a problem.

Expense is also one major reason why pollsters typically stop polling in the days before an election — it may not be worth the money, since the data will have limited use by the time its synthesized. 

But this year, 13 percent of voters told exit pollsters they decided their vote in the campaign’s final week, with Trump winning that group by 5 percentage points. 

“People don’t finalize their Christmas shopping until Christmas Eve. Why would they finalize their voting decisions? Human nature tells us a lot of people wait until the last minute,” Ruffini said. 

A final fix? Zogby said consumers need to quit thinking the polls are a sort of guarantee of who is going to win.

After the 2012 election, in which Silver was deified for his calculations, Zogby suggested many put too much faith in how accurate the polls could possibly be.

“We need to take the word ‘prediction’ out of this business,” he said.

“If I’ve got to lose 20 pounds between now and Nov. 8, the scale will tell me how I’m doing. … But it’s not going to tell me what my weight will be on Nov. 8.”