Should you trust political polls? It depends.
By Maggie Lee
It’s Atlanta election season and mayoral candidates are sending emails with talk of “surging” in polls, of “momentum,” of opponents “cratering.”
Those emails are full of confidence, but the broad surprise at the election of President Donald Trump last year suggested to many that there’s something wrong with polling.
Two Atlanta pollsters insist that good polling backed up by research of an electorate is reliable. Not that mistakes aren’t made — in fact, a big national review of 2016 polling found things that went both right and wrong.
But it is difficult, it’s getting more expensive, and media presentations and a voter’s own bias may change how they think of polls.
Mark Rountree is one of the people trying to measure the electoral disposition of a city from a bunch of phone calls, math and figuring out who’s likely to vote. He’s president of Landmark Communications, a Republican firm that’s been around for more than two decades. It’s done three published polls on the Atlanta mayoral race for WSB-TV. The first was in March, the latest in October.
The hard part isn’t the dialing, Rountree said, though you do have to have a good call list.
“Here’s where they make the mistake, it’s almost always the back end, it’s almost always modeling what the election is going to look like that you’re polling,” he said.
That is, modeling the group of people who will turn out and vote, and making sure the responses match that profile, giving the right weight to party and to older folks and younger ones; black and white people, men and women.
“Making that model work is the hard part, because what you’re really projecting is, before an election is held, what is the demographic makeup of the future election that hasn’t even happened yet? That’s why this is hard,” he said.
For the record, Landmark called races correctly 88 percent of the time and got a “B” grade from 538, the data journalism team at the New York Times, according to a review they did in August, 2016. The vast majority of pollsters (national and regional ones) cluster in the middle, with a B or C grade, according to the footnotes.
His company has been collecting data for years and their model looks at things like demographics of the turnout at previous similar elections, at a district’s population, and at a voter’s own history of showing up at the polls.
The modeling of a September poll by Survey USA survey got dissed by a mayoral candidate who didn’t do so well. Peter Aman’s campaign argued that the electorate is likely to be whiter, more female and older than that poll assumed.
“We appreciate the work by SurveyUSA, however, we ask that they adjust their projections to better align with historical performance,” said his campaign manager, Fred Hicks, at the time.
SurveyUSA got an overall “A” from 538.
But the work isn’t getting cheaper or easier, for now, especially in a Democrat-leaning place like Atlanta. For that, blame mobile phones and the law.
By law, pollsters can’t use the same handy machines they use for predialing landlines to call mobile phone numbers. Instead, folks who only have a mobile need to be dialed by hand or via online surveys.
Younger people and African-Americans are likelier to be mobile-only customers than older folks or white folks, said Rountree. Those first two groups that are hardest to get hold of are also likelier to vote Democrat. That can make Democrat polling more expensive, and if it’s done poorly, less accurate.
Then, when it comes to polls, there are what you might call psychological factors in reading them.
“Polls are reliable as long as you’re not using them as a predictor of how the election is going to turn out,” said Mike Hassigner, a Republican political consultant, pollster and Georgia politics blogger.
“That is exactly how people look at them and that is exactly why people don’t trust them,” he said.
A good poll is an accurate snapshot of a certain day, not a projection of what will happen months out.
“Everybody wants every poll to be the predictor of how the election will ultimately turn out,” said Hassinger. “And what we have in the case of the Atlanta mayor’s race is a mathematical certainty that there will be a runoff.”
Most polls count eight candidates; nine counting “undecided.”
There’s also another factor, he said: people tend to think of themselves as a whole sample, when in fact they’re just an anecdote.
“If your chosen candidate is not tops in the polls, your inclination is to doubt the poll,” Hassigner said.
And then there are the things outside a pollster’s control, like news or campaign ad spending that happens between polling day and the election.
“Today numbers can change because somebody gets caught in an embarrassing situation on social media and 48 hours later they’re out of the race,” said Rountree.
And asked about the accuracy of polling, Rountree says that the national presidential ones in 2016 were actually quite accurate.
Indeed, the 2016 national presidential polls were some of the most accurate ever, according to a review by the American Association for Public Opinion Research. Collectively, national polls indicated Hillary Clinton had about a three-point lead; she ended up winning the popular vote by two points.
But it’s states, via the electoral college, that decide elections, not the popular vote. And state-level polling showed a closer race, according to AAPOR, and polls did underestimate Trump’s support in the upper Midwest.
Also, there’s what happens when people see polls on the news.
What happens at the media level, said Rountree, is that a reporter doing a two-minute segment on national news doesn’t have time to report on numbers in dozens of states. The reporter just wants the national number.
And, that AAPOR review said some information that was widely discussed in the media on the eve of the election in November 2016 turned out to be misleading or wrong, such as that upper-Midwest polling, and forecasts by professors and data journalists.
“The day after the election, there was a palpable mix of surprise and outrage directed towards the polling community, as many felt that the industry had seriously misled the country about who would win,” the report says.
The election is on Nov. 7. The runoff is scheduled Dec. 5.
And, for anyone who cares for a glance of history, here are some published polls.
Mary Norwood has had a consistent polling lead, and Keisha Lance Bottoms and Peter Aman made gains between March and October. But taking into account margins of error and “undecided” likely voters, the race seems pretty close as of the last polls.
Links to more information follow.