UNH pollster missed the mark for Clinton-Trump
Pollsters across the country are trying to discover why they failed to predict Donald Trump’s victory.
In New Hampshire, the University of New Hampshire Survey Center, which does polling for WMUR-TV, faces an even more uncomfortable truth — it predicted an 11-point victory for Hillary Clinton in this battleground state.
Meanwhile, the average of all seven firms polling in the state at the time said Clinton would win the presidential race by just three-tenths of a percentage point, according to RealClear Politics.
In the end, Clinton won 47.6 percent of the vote, compared to 47.2 percent for Trump.
“We’re actually trying to figure that out right now, because the polling this time across the board for us was quite unusually off and I want to figure out exactly what’s going on,” said Andrew Smith, director of the survey center.
Ryan Williams, a veteran, Republican media strategist, said the entire election cycle exposed the WMUR/UNH Granite State Poll as badly flawed.
“These were always going to be close races. All the internal results I saw was the races for President, Senate and governor here were all too close to call, yet UNH kept putting out polls showing it as a double-digit lead for the Democrats,” Williams said.
“It was like they were polling a parallel universe. The campaign professionals on both sides of the aisle knew they were consistently wrong.”
In swing state after swing state, pollsters also overestimated Clinton’s strength.
The Trafalgar Group, a Republican polling firm, was the only one that had Trump ahead in Pennsylvania or Michigan; there was no poll that had Trump winning Wisconsin.
Last Tuesday, this Rust Belt trifecta was what led to Trump’s stunning, Electoral College win.
“I think it was an important polling miss. It would really be glossing over it to say that it was a typical year,” said Courtney Kennedy, director of survey research at the Pew Research Center.
Some reasons offered for the misfire include failing to account for “shy” Trump supporters who wouldn’t tell pollsters they were for him; the growing use of cell phones; and the universe of new voters Trump mobilized who didn’t get into any pool of likely voters.
“I think we’re going to need some more time to figure that out. But when we say there was a poll failure; there was a catastrophic, across-the-board poll failure; public polling, private polling by both campaigns,” said Chris Galdieri, assistant professor of politics at St. Anselm College.
Random digit dialing
The UNH Survey Center stood out in other races, too. The last UNH survey had Democrat Colin Van Ostern beating Republican Chris Sununu by 11 percent, 55-44 percent. Sununu beat Van Ostern by 2 percent.
“We had a sample that was much more favorable to Democratic candidates than what actually bore out, so we want to see if there was any systematic bias in the framework,” UNH Director Smith said.
“We do random digit dialing so every cell and every house has a chance of being called.
“We need to see if there was any differences in devices used to identify likely voters, or if there is anything in how people felt toward the parties or their willingness or unwillingness to state their support for the candidates they supported.”
Why were the other polling outfits such as Suffolk University for the Boston Globe and UMass Lowell for 7News so close to the final outcome?
“I’m just speculating because they don’t share their methodology, but the biggest difference in a number of polls we looked at is that many of those polls did weight by party registration or party identification and they used listed samples of registered voters rather than random samples of all adults in the state, and that’s something I’ve wanted to avoid in the past,” Smith said.
“Up to 10-15 percent of the electorate register on the election day, so they would not have shown up in the samples.”
Greg Moore is a longtime Republican strategist and state director of Americans for Prosperity, a fiscally-conservative group that has hired dozens of firms to do polling in key states.
Moore said if Smith wants to continue the “purity” of no weighting of his sample he must include more than the 800 likely voters that UNH employs for its polls or risk having a wildly-unreliable result.
“If you are doing random-digit dialing and polling even 800 then you are setting yourself up for potentially a bad sample,” Moore said.
“If UNH is going to be the pollster of record in this state you would hope they would look at their methodology to improve their results.”
Senate Democratic Leader Jeff Woodburn of Whitefield said he did find it to be a very challenging environment to get accurate results and saw mistakes not in just UNH polls.
“As I looked at all those polls and looked at the cross tabs, I saw the lowest Trump numbers were in the North Country and that’s not what I was seeing on Election Day,” Woodburn said. “Trump won Coos County which is pretty much unheard of and so did Kelly Ayotte. You always wonder who answers these (polling) phone calls?”
2016 included many UNH hits
In many ways, this was a rough campaign season for the Granite State Poll.
On the day before the election, Republican Senator Kelly Ayotte’s Campaign Manager Jon Kohan issued an unusually harsh tweet to the last UNH poll that had Democratic Gov. Maggie winning, 52-47%; Hassan won by only 978 votes, 48-47.9%.
“For those covering the new (absurd) @UNHSurveyCenter poll, remember these wise words from @AFJacobs, spox for Gov. Hassan.”
Aaron Jacobs, Hassan’s communications director, had recently advised, “In a small state with lots of polls in the field simultaneously and voters’ phones ringing off the hook, it’s difficult to get a representative sample.”
Back in less-tense times, Robby Mook, Clinton’s national campaign manager, had this to say about a WMUR-UNH poll on May 5, 2015 courtesy of WikiLeaks that hacked Campaign Chairman Jon Podesta’s email server.
The poll showed Clinton’s support had fallen by seven points in a head-to-head with both Senators Bernie Sanders and Elizabeth Warren; Warren ultimately did not run.
“Although I still hate UNH poles (sic) and what they have done to my blood pressure for years,” said Mook who ran Sen. Jeanne Shaheen’s successful Senate campaign in 2008.
Later the same day Mook emailed “Madam Secretary” as he addressed Hillary Clinton regarding the same UNH survey.
“As always, this poll doesn’t have a good history of accuracy so we need to take it with a grain of salt,” Mook wrote Clinton.
UNH Survey Director Smith said he’ll take time with this polling autopsy to make sure 2016 wasn’t a once-in-a- generation anomaly.
“If we see something here that we figure we can pinpoint we may make some changes, but the other thing I want to wait and see is how the industry reacts and the analysis of polling across the country,” Smith said.
“I don’t want to change something that has been historically very accurate and worked very well, if this was just an unusual one-time instance.”
In that vein, Smith said there’s been one parallel in modern history to this wildly unscientific election result.
“This reminds me of the 1980 election. President Jimmy Carter was leading in the polls the week before the election, pollsters were being told voters didn’t trust Ronald Reagan,” Smith summed up.
“Then on Election Day, Reagan won even bigger than Trump did this past week.”
Since 1984, American University professor Allan Lichtman has predicted the winner of every presidential race using 13 specific keys. On Sept. 23, Lichtman said the keys pointed to a “very, very narrow” Trump win. He said polls were not predictions but “snapshots.”
“The bigger picture is that the outcome is utterly inexplicable, based on conventional analysis. Some of the pollsters are good friends, and good people, but during the next election, I hope we send them to a very nice Pacific Island for a very nice vacation,” Lichtman quipped.