During the weeks leading up to this most recent election, news sources came out using poll results to show that Democrat Hillary Clinton was likely to win by a large margin. Even as polling sites were starting to close Tuesday night, the New York Times projected that Clinton was still highly favored to win.
Throughout the evening, that scale gradually moved in Republican Donald Trump’s favor. At about 3 a.m., Trump reached 276 electoral votes and was declared President-elect.
“The media became too reliant on the pollsters,” Frederick Pearce, a USF mass communications research methods professor, said. “The pollsters are really only giving a snapshot of what people have the potential to do. It is not a predicted method.
“For instance, I think probably in most polls, millennials are under-represented because of landlines. I also think rural Americans are underrepresented. If you know what you’re looking for, you will wind up getting it. It’s the myopia of pollsters.”
While Trump won the electoral vote, Clinton led the popular vote by 206,770 votes at the time of print with three states still counting votes.
“In my opinion, polls didn’t get it wrong and missed the story, although I understand that that is a popular narrative today,” Al Tompkins, senior faculty for broadcast and online at Poynter Institution of Journalism, said. “I think that what we’re missing is this, ever since the primaries … there has not been a day since the conventions that the national polls have shown the race as greater than the margin of error.”
Tompkins said the margin of error for such races is 3.2 percent, which means that to be actually winning a candidate would need a lead of more than 6.4 percent as the margin of error is applied to both candidates.
“I think what we saw there was something no one can predict, even with the best polling technique,” Edwin Benton, a USF political science professor, said. “The chance of error in the polls, which everyone knows, what happened fell within that range and there is nothing you can do about it.”
Tompkins also explained two aspects of collecting data that may have influenced before voting polls: the social desirability index and weighting.
“We can’t prove that it happened, but anecdotal information indicates that it did,” he said. “The social desirability index means that … it seems to be true that a fair number of Trump supporters told pollsters that they either hadn’t decided or were not going to vote for Trump.
“In fact they were going to vote for Trump, but avoided saying so as to avoid ridicule.”
Tompkins said he heard one estimate that this effect could have been up to five percent of Trump supporters. This has been seen in past elections.
“It makes sense to me that that could have been a factor here,” he said. “But I have a little bit of doubt because it seems to me that Trump supporters were really proud that they were Trump supporters … We just don’t know.”
The weighting effect that Tompkins referred to as a pollster’s struggle to get responses from a specific demographic, which leads the responses that they do receive to count more.
“(For example), Pollsters had a heck of a time getting young, black males to respond this time around,” Tompkins said. “What happened was that in some polls they ended up taking those young black males when they could get a response and let that person represent many others that they needed in order to have a representative sample.”
The problem with that comes from putting too much value on one statement. There’s usually many different ways of looking at a situation and it becomes dangerous to assume the position of a demographic based on too small of a sample size.
“It certainly is a quandary as to how much stock you can place into big data and it might be beyond the ability of people to sort through and pick out what appears to be accurate and what is not,” Benton said. “It’s almost like your groping in the dark; do your best to try and interpret but it might not be good enough.”