WEST HAVEN, Conn. (WTNH) — Call it public opinion research or data journalism, but it’s all based on predicting unpredictable human behavior, which can help explain why polls were so wrong on Tuesday.

From U.S. Senate, to the presidential race, pollsters were way off in many predictions of how election day results would go.

“You want to get a random sample of population you’re trying to describe,” said University of New Haven Political Science Professor Chris Haynes. “The population we’re trying to describe for elections is voter who turns out on election day.”

Haynes said election day turnout by white voters with no college degree was much higher in old industrial states that were once reliably democratic.

“There just weren’t enough polls done in those state like Michigan or Wisconsin or Pennsylvania because they didn’t think they’d be competitive states,” Haynes said. “For the longest period of time Hillary was up 10-15 points in those states. So public opinion pollsters ignored them.”

There’s also the so-called hidden Trump vote theory. Trump supporters who tell pollsters otherwise.

“Voters maybe not so willing to admit to a live pollster on the phone that they wanted to vote for Trump,” Haynes said.

Another factor could also be the FBI director letter to Congress about reviewing more Clinton emails just days before election.