President Donald Trump leads a cabinet meeting at the White House July 16, 2019. (Photo by Chip Somodevilla/ Getty Images)
Not many people would bet next month’s rent money on the accuracy of the next political horse race poll proffered by the media. Can you blame them?
The perception is that recent polls have failed us spectacularly. Just ask Joe Biden, whom polls showed firmly in control of the Democratic presidential nomination just a few short weeks ago. Or ask President Hillary Clinton.
Poll after poll showed her with a solid lead over her Republican nemesis, Donald Trump, whose own organization entered Election Day 2016 expecting defeat only to emerge as stunned victors.
Leading up to that election, the dominant media message was that Clinton had it sewn up. Few doubted the prevailing conclusion that the former secretary of state, though damaged by her hacked emails, had enough gas to outlast Trump, a reality TV figure famous for ridiculous bombast who was reeling from the disclosure of his profanely sexist hot-mic boast about indecent liberties he’s taken with women.
The morning after, however, questions came in a torrent. How had Hillary’s sure thing reversed itself? It was the second time in five elections that a Democrat favored in the polls won the popular vote only to lose the election.
There were flaws in the 2016 polling. Pro-Trump voters in largely white Upper Midwest states were undersampled, according to an exhaustive post-election postmortem by the American Association of Public Opinion Research. That was significant since Trump’s surprise victories in Pennsylvania, Michigan and Wisconsin — states that had voted Democratic for six presidential elections in a row — all supported Trump and gave him an insurmountable Electoral College advantage.
Reasons the study identified for underestimating Trump: polls oversampled college grads who trended toward Clinton and are more likely to indulge pollsters; some Trump voters did not disclose their support for him to pollsters; and there was a significant regional shift toward Trump in the campaign’s final week.
Nevertheless, the polling wasn’t as errant as it seemed in the immediate aftermath as pundits, the media and Democrats struggled to unpack what went wrong. An average of public polls had predicted that Clinton would prevail by 3 percentage points in the popular vote. She won the popular count by about 3 million votes, or 2 percentage points.
Less attention was paid to state-by-state results that could have shown how that translates into electoral votes, the only tally the Constitution recognizes for granting a four-year lease on the White House. Ultimately, Trump received 307 electoral votes, or 57 percent, compared to Clinton’s 227, or 42 percent — a 19 percentage point Electoral College margin sharply at odds with the popular vote.
The problem was the way the complicated data were often reported (misreported?) and the way a public with a shortening attention span didn’t quarrel with the prevailing media narrative that Trump was a sassy flavor of the month, but not a serious candidate with a real chance to win.
“Media polls are a way to check in on the horse race. There’s a big appetite out there among the public for the horse race. It’s like sports,” said Kyle Kondik, the Washington-based communications director for the University of Virginia Center for Politics and managing editor of its nonpartisan newsletter, “Sabato’s Crystal Ball.”
Campaigns conduct polling differently, Kondik said. “They use it as a tool to inform how to run a campaign,” he said.
Virginia-based Republican campaign consultant Christopher J. LaCivita has become nationally prominent doing exactly that. He’s an authority on the use of sophisticated public opinion research to tailor campaign strategy and messaging, and he has no use for media polls.
“I don’t think there’s anything wrong with a college or university doing a survey. But when a news media company pays for polling and puts their name on it, they’re making news. Their job should be reporting the news,” said LaCivita, who advises GOP candidates, political groups and businesses across the country.
Rather than fixate on topline head-to-head totals as most media do, campaigns structure polls to yield specific leading indicators — “favorables vs. unfavorables” — as guideposts on a map, he said. Those polls have a clearer purpose, benefit from better modeling, yield more actionable detail and require financial resources that few media organizations can match.
“Statewide campaigns will spend a quarter of a million dollars in just 10 months on polling data,” LaCivita said. “And they’re getting more expensive all the time.”
“For us to do a quality poll, we’ll make something like 20,000 calls just to get 400 responses. And then some of them will just hang up midway through an interview,” he said.
The disappearance of land-line phones and the rise of caller ID and cell phones is partly to blame. People will not answer calls from numbers they don’t recognize, he said. Those who do answer are more likely to mislead pollsters.
Polls backed by political campaigns or parties also have a higher success rate in reaching likely voters because they compile their sample not from the public at large but from public voter files — data every state maintains from each election on who cast ballots. Costs are generally several thousand dollars per file per state, something few in the financially strained media industry are willing to pay.
Online polling is cheaper than phone polling conducted by human questioners and is becoming more viable, but serious questions remain about whether it can fully supplant telephone polling, Kondik said.
Ultimately, he cautions, getting a full understanding of the snapshot-in-time political insight that polls yield is the duty of voters who care enough to be attentive and inquiring. “People need to be more skeptical and discerning about the information they see,” he said.
Democracy, after all, isn’t a spectator sport but one that requires participation.
Our stories may be republished online or in print under Creative Commons license CC BY-NC-ND 4.0. We ask that you edit only for style or to shorten, provide proper attribution and link to our web site. Please see our republishing guidelines for use of photos and graphics.