More than a week after the 2016 presidential election, voters have had a variety of reactions: happiness, anger, fear, shock and — most of all — confusion.
The University’s Public Policy Research Lab operates under the Manship School of Mass Communication’s Reilly Center for Media and Public Affairs and the College of Humanities and Social Sciences. PPRL Director and mass communication assistant professor Michael Henderson researches public opinion and polling.
Henderson said the main problems with election polling originated from the interpretation of the polls and discrepancies within individual state polls, especially within battleground states.
The average polling error is about two points. National polls estimated Democratic presidential nominee Hillary Clinton winning the popular vote by 3.5 to 4 points. She actually won by two points.
Polling error is the difference in what the polls indicate and what actually happens. Henderson said this error was exacerbated by the structure of how the Electoral College works.
“The reason I think people were so surprised actually has to do with extrapolating away with the national polls themselves,” he said. “To put it in perspective, the national poll error of about two points is not that different from the national poll error four years ago.”
In 2012, polls overestimated support for Republican presidential nominee Mitt Romney by two points. The overvalue wasn’t a deal breaker though, because the polls still had Romney losing. This year, however, the winner was someone different than predicted.
Henderson said there has been historically large error in state polls — more than two times the amount of national polls. When the margin of difference is expected to be extremely close in swing states at less than two points, bigger problems arose when those polls were off by five or six points.
In terms of a mathematical explanation, pollsters still don’t know why the estimates were so off.
Henderson said the notion of the shy Trump voter could have contributed to polling misconceptions. The concept is based on social desirability bias, a theory positing people don’t outwardly admitting their support of Trump because they see it as being socially undesirable. This theory also includes the idea that more undecided voters were actually pro-Trump than were originally thought.
In addition, polls overestimated Democratic voter turnout while underestimating turnout among Trump voters. In the week and a half leading up to the election, polls missed a late trend toward Trump which could be responsible for tipping the election in his favor in battleground states.
“I think there’s also an inadequate respect for uncertainty,” Henderson said. “Polls have error … and we sometimes ignore those other potential sources for error, which is a little ridiculous because we know from past elections it’s not unusual for polls to miss by two or three points.”
Henderson said the election results shouldn’t have taken voters by surprise because — in a close election — missing by two to four points is not unheard of.
The tighter the race, the more difficult it is to make an accurate prediction. 
“I think it would be a mistake to say that ‘Well, the polls were ridiculously wrong because they were saying it was going to be a Clinton landslide — not the last couple of weeks they weren’t,” Henderson said. 
There’s also the theory that the public didn’t pay enough attention to polls because of their own emotional investment in the election results, also known as motivated reasoning. 
“A good share of voters, parties aside, don’t like to hear information that doesn’t confirm what they want to see happen,” he said. 
As the polling trend was moving toward a close election, voters should have been more realistic about the possibility of error and Clinton’s chances. 
“In those last few days, we all should have been like, ‘This thing’s narrowing, and now we’re looking at a race where we could be within a not unusual polling error, and then it could go the other way,’” Henderson said.