WASHINGTON—Election forecasters suffered their own “what happened?” moment in 2016 after Donald Trump swooped in to take the Oval Office.
Longtime forecaster Larry Sabato said the industry had got “drunk on polls,” confident they were enough, even in the new era of Trump and social media.
“There were so many polls from so many sources, and polling averages had been so accurate in number—not just national elections, but state, even local, elections—that we assumed that was enough,” Sabato said during a midterm election seminar at the National Press Club in Washington on Aug. 28.
“And it was a terrible assumption.”
Sabato, the founder and director of the University of Virginia (UVA) Center for Politics, has been in the forecasting business since the 1970s, when few polls existed and prediction methods were more rudimentary.
Taking his mea culpa from 2016, Sabato and his team from UVA joined forces with global market research firm Ipsos to bring more metrics into their prediction modeling on a new website, Political-Atlas.com.
Correctly forecasting the midterms will be the group’s first big test and includes predictions based on polling, expert analysis, and social media chatter, said Cliff Young, president of Ipsos, which is “knee deep” in elections around the world, he says.
He said a rise around the world of anti-establishment sentiment since 2013 has changed the game.
“And in a world of change where voter calculus is changing, our tried-and-true models don’t work as well,” Young said.
Within its polling component, Young said Ipsos will conduct about 10 million interviews a week leading up to the midterms.
But, he said, the most interesting input into the system is social media trends.
“We’re using an artificial intelligence-based series of algorithms to track real time social media sentiment,” Young said. Top issues can be gauged, as well as the “velocity” of a campaign through mentions per hour.
He said the company has used it “to good effect” in three elections now: the Ontario provincial election, Mexico presidential election, and the upcoming election in Brazil.
As is the case with all midterms, the president drives them, Young said.
“It’s really about Trump, Trump, Trump,” he said. Democrats’ top concern, aside from Trump, is health care, whereas for Republicans, it’s immigration.
In the last several weeks, Trump has been tweeting out his endorsements and, later, touting his record for wins. So far, he has held rallies in Florida, Pennsylvania, Ohio, and West Virginia to support his candidates.
Trump is about to launch into a punishing pre-midterm schedule to push his candidates over the finish line, according to administration officials on a press call on Aug. 21.
The officials said as many as one-third of attendees at Trump’s Make America Great Again rallies are not Republicans—and that’s who they want to turn into voters.
In recent races, most of Trump’s endorsements have lined up with wins. He started tweeting his support for special election House candidate Troy Balderson on July 21. After 11 Trump tweets and one last-minute rally on Aug. 4, Balderson narrowly won on Aug. 7. He now faces the same opponent, Danny O’Connor, in the midterms on Nov. 6.
“The Republicans have now won 8 out of 9 House Seats, yet if you listen to the Fake News Media you would think we are being clobbered,” Trump wrote on Twitter on Aug. 8. “Why can’t they play it straight, so unfair to the Republican Party and in particular, your favorite President!”
The current House breakdown is 238 Republican to 191 Democrat seats. To have the 218 seats needed for control, Democrats need to flip at least 23 GOP-held seats.
The post-WWII average in midterms is a 26-seat House loss for the president’s party, according to Kyle Kondik, managing editor of Sabato’s Crystal Ball, an election analysis site.
“An average performance would be enough historically for [Democrats] to win the House, but there are always complicating factors,” he said, adding that there is enough certainty that it wouldn’t be a “slam-dunk.”
The group most likely to cause upsets is young suburbans, said Kondik.
“Some are already starting to describe this year as sort of the ‘year of the suburban, college-educated women’s revolt,’” he said.
“We know from polling data that the president is very unpopular amongst that group, and that group is voting pretty heavily.”
However the results shake out, the House will already experience significant turnover this year—61 seats have no incumbent running, 40 of which are currently held by Republicans and 21 by Democrats.
“Those are the races where you might expect to see the really big shifts from 2016 that we’ve seen in the special elections, because there’s no power of incumbency to mitigate that,” Kondik said. “So besides the big-picture, national indicators, the fact that there’s just a lot of open seats and the Republicans are defending about two-thirds of them, is also a good sign for the Republicans.”
Sabato’s prediction is that Republicans will most likely lose the House, as has been the case with all but three incumbent parties in the last 100 years—1934, 1998, and 2002 midterms.
He said the strong economy should mitigate losses for Republicans, combined with a lack of military action that often moves the needle, but, with many caveats, he expects Democrats to take the House.
“But then you have the extraordinary and totally unprecedented presidency and persona of Donald Trump—which may outweigh anything else,” he said.
“Since the rules have been broken once, recently, it’s always possible that the midterm rules of engagement will be broken too.”
The Senate looks safe for Republicans, who hold a slim 51 to 49 majority, according to Geoffrey Skelley, who is tracking the Senate and gubernatorial races for the 2018 Political Atlas.
Democrats are defending 26 of the 35 Senate seats up for grabs.
“The story of defensive is important here, because for Democrats in the Senate, they’re also defending 10 seats that are in states that Trump won,” Skelley said. “And five of those are seats that Donald Trump won by at least 18 percentage points.”
On the gubernatorial map, it’s the opposite—Republicans are defending 26 of the 36 governorships.
Skelley predicts an “almost certain” net gain by Democrats, “it’s really just a question of how many.”
He is expecting a close contest in states such as Florida, Michigan, Ohio, Wisconsin, Iowa, and Nevada.
“A lot of these gubernatorial contests are going to take place in important states—a lot in the midwest that are very important when you think about the next redistricting cycle,” he said. “Michigan, Wisconsin, Ohio, these are all states where Republicans were really able to draw very, very sharp gerrymanders that favored their party.”
Getting It Right
With the added information being fed into prediction models, Kondik is confident a repeat of the 2016 presidential predictions won’t happen in the midterms. He said he’ll be looking to see if one of the three methods his team is using (polling, experts, social media) performs notably better or worse than the others.
In the last three House election cycles he has predicted, he has missed between six to 10 seats, and suspects this one will fall into the same range—which he doesn’t consider a failure.
“If there’s a situation where many think that Republicans will lose the House and they end up netting 10 seats or something—yeah, that would be a big failure,” Kondik said.
“I think that if there is a big miss this year, it might make us even more skeptical of traditional polling, or poll-based modeling … and I think it will also lead us to question whether there is a shy Republican/shy Trump vote out there.”
Toss-Up Races in the Midterm Elections
The University of Virginia Center for Politics has made predictions on the upcoming midterms. Below are the states where they predict the races could go either way, as of Aug. 28.
* indicates incumbent Republican is running.
** indicates incumbent Democrat is running.
UVA Crystal Ball estimates 32 House seats could go either way, as of Aug. 23. They predict that all toss-up races with incumbents running are Republican seats.
California (10th*, 25th*, 39th, 45th*, and 48th* districts)
Colorado (6th* district)
Texas (7th* and 23rd* districts)
Florida (26th* district). Primary is Aug. 28.
Kansas (2nd and 3rd* districts)
Iowa (1st* and 3rd* districts)
Minnesota (1st, 2nd*, 3rd*, and 8th districts)
Illinois (6th* and 12th* districts)
Kentucky (6th* district)
Ohio (1st* and 12th* districts)
Michigan (8th* district)
West Virginia (3rd district)
Virginia (2nd* and 7th* districts)
Pennsylvania (1st* district)
New Jersey (3rd and 7th* districts)
New York (19th* and 22nd* districts)
Maine (2nd* district)
UVA Crystal Ball estimates six Senate seats could go either way, as of Aug. 23.
Arizona* Primary is Aug. 28.
Florida** Primary is Aug. 28.
UVA Crystal Ball estimates 10 Senate seats could go either way, as of Aug. 23.
Alaska (Independent incumbent)
Florida* Primary is Aug. 28.