Polling is an inexact science, as this year’s election results can attest. After multiple polls gave Democratic nominee Hillary Clinton a wide lead over the GOP’s Donald Trump for much of the presidential race, her numbers shockingly collapsed on Election Day to produce an unexpected Trump victory.
But pollsters in South Carolina say the problem is less in the polls themselves than the media spin that surrounded them.
“The polls showing Clinton ahead nationally were correct. She’s going to end up with close to a million more votes,” said Scott Huffmon, director of the Winthrop University poll. “The error was that the media was ready to interpret that as what the the Electoral College was going to do, because it’s hard to get as excited about 50 different state polls.”
Polls also tightened in several swing states heading into Election Day, which boosted Trump’s chances late in the campaign.
Sign Up and Save
Get six months of free digital access to The State
“They would say, ‘he needs to get an inside straight,’ but if you look at the numbers, he was not far off,” said Luke Byars, a partner in the consulting firm First Tuesday Strategies, which conducted their own poll of South Carolina ahead of the election.
Some have blamed the election result on so-called “shy Trump voters” who didn’t want to admit they were supporting Trump. But Huffmon says it’s more likely those voters just didn’t respond to pollsters at all. Sometimes when a voter’s preferred candidate is suffering from negative headlines – as Trump often did during the campaign – the response rate from those voters tends to go down.
Pollsters can also be hampered by an inability to reach people who only use a cell phone, or how they identify a “likely voter,” a category that became less easy to track in an unusual election.
“Voters are weighted based on whether they voted in past elections,” Huffmon said. “Trump didn’t bring out new voters; what he brought out were irregular voters.”
Byars said Trump confounded polling models because he “redefined what it means to be a Republican” – appealing to voters who demographically would have been expected to vote for a Democrat, at least according to the modelers’ data.
“So when you have a model liberal voter saying they are going to vote for Trump, the pollsters say ‘how did that happen?’” he said. “They model that voter down, changing the data. Sometimes pollsters think they’re smarter than the voters.”
Huffmon said the results of the 2016 election may lead some pollsters to adjust how they conduct polls, but he said a good pollster should do that anyway.
“Legitimate pollsters, academic pollsters never stop researching how to do a more scientific survey,” Huffmon said.