Everyone got it wrong.
From Nate Silver's vaunted fivethirtyeight.com predicting a 72 percent chance of victory for Hillary Clinton to University of Virginia's Larry Sabato's Crystal Ball group predicting she would win 322 electoral votes in a landslide, to the polling aggregator site Real Clear Politics that had Clinton with a 3 percent national lead.
But when the real votes were counted Tuesday night, Donald Trump racked up enough states to claim an Electoral College victory. Clinton was slightly ahead in the popular vote as of 9 am Wednesday, and was preparing to give a concession speech in New York.
Some political commentators believe that the polls were wrong because people lied to pollsters about their intentions to vote for Clinton. The thinking is Trump's divisive campaign rhetoric and offensive comments about women, minorities and immigrants made it difficult for people to admit their support for him to strangers.
RELATED: Move to Canada? It's Not That Easy
There's no way to gauge people's truthfulness on the phone, but there may be something else going on, according to David Lublin, professor of government at American University who has written extensively on electoral systems in the United States.
Lublin said polling is as much an art as it is a science. Pollsters have to make decisions about how they put together a representative sample of national or statewide voters, sometimes these decisions are based on judgments with incomplete information.
"The key to the science of polling is if you take a random sample of the population and get their views, and that it is the best predictor of what people are going to do," Lublin said Wednesday.
"The problem has been that getting a random sample is no longer possible because the usefulness of this invention has led to huge mountains of polling for marketing purposes. People don't answer the phone."
It's even worse for two important groups for the Democratic candidate: young people and Latinos.
In 2013, 41 percent of U.S. households had no landline, according to a study by the Pew Research Center.
Consumer protection laws prohibit telemarketers from autodialing cellphones, so it becomes more expensive for pollsters to contact young people.
Lublin also noted that polling Spanish-speaking voters in English gives less accurate results, according to studies.
"If you poll Latinos in English you get radically different answers than using bilingual callers," he said. "The latter is more accurate."
In Pennsylvania, for example, many polls had Clinton leading by 3 to 6 percentage points. The final poll of Pennsylvanians by Muhlenberg College and the Allentown Morning Call had Clinton up by 4 points before the election.
But by the morning, Trump won Pennsylvania by 1 percent, the first Republican to win the Keystone State since 1988.
"We simply didn't accurately project turnout among certain groups," said Christopher Borick, director of the college's Polling Institute. "Those groups are unlikely voters, folks that don't usually vote and that we often in our models will not include. This time they did."
RELATED: Climate Diplomats Call Trump Presidency a 'Disaster'
Borkick says pollsters uses results from online focus groups as well as telephone surveys.
"We have to look at the crowdsourcing models and online models versus probability models and how the various phone methodologies work," Borick said. "This a field that's in a very turbulent moment right now, there are lots of disruptions. This is part of the autopsy that's going to go on.
Borick said the Muhlenberg poll accurately predicted the outcome of Pennsylvania's Senate race that he is optimistic that the next election's polls will be more accurate.
"Trying to deduce what the factors are will take a bit of time and some experimentation," he said. "This is destined for deep analysis."
Photo: Trump supporters in Scranton, Pa. Credit: Reuters WATCH: What's the Science Behind Polls?