The final days of Iowa polling had a cohesive set of predictions. Donald Trump was supposed to win. Marco Rubio was supposed to come in a distant third. And Bernie Sanders, despite a last-minute challenge to Hillary Clinton, was expected to narrowly lose.
Now we know the polls got it wrong, particularly on the Republican side. Trump lost to Ted Cruz and barely eked out a second-place showing over Rubio. Sanders and Clinton are still locked in a virtual tie for first.
The Iowa caucuses are notoriously difficult to predict. But even the final Des Moines Register poll, which has a good record, missed Rubio's rise and Trump's fall. Here are three factors that help explain why the polls got it wrong this time around.
A bigger turnout was supposed to help Trump, but it didn't
Republicans turned out to vote in huge numbers: 185,000 people went to the caucuses, up 54 percent from 2012, according to the Washington Post. According to conventional wisdom going into the vote, a big turnout should have helped Trump by proving he could motivate the infrequent voters who were among his strongest supporters.
But it wasn't enough. While about 45 percent of Republican caucus-goers were caucusing for the first time, only about 30 percent of those new voters supported Trump, according to exit polls. So the new voter turnout wasn't a groundswell in his favor.
Observers had been warning that Trump's ground game — the network of volunteers and organizers doing the crucial, down-to-earth work of making sure supporters show up to the polls — was disorganized or nonexistent. On January 13, the New York Times's Trip Gabriel called the campaign's Iowa operation "amateurish and halting":
As temperatures plunged to single digits over the weekend, canvassers for Hillary Clinton posted photographs of themselves on social media going door to door in the snow. Meanwhile, Mr. Trump’s volunteers in Davenport, a city where the campaign appears to be better organized than elsewhere, decided it was too cold to go out.
Seven volunteers worked the phones at the Iowa headquarters of Senator Marco Rubio of Florida in a Des Moines suburb one night last week. At the state headquarters of Mr. Cruz, there were 24 volunteers in a room beneath a sign proclaiming a daily goal of making 6,000 calls. The Trump state headquarters in West Des Moines were largely deserted.
Trump ended up losing Davenport, too, to Rubio.
Evangelical Christians turned out to vote for Cruz
The Des Moines Register poll estimated that 47 percent of caucus attendees would be evangelical Christians. When the pollster changed the model for bigger evangelical attendance, it found Cruz would pull even with Trump if 60 percent of voters were evangelical.
Even that turned out to be an underestimate: 62 percent of caucus-goers described themselves as evangelical or born-again, according to exit polls. And the plurality of their votes went to Cruz. Less religious voters supported Trump.
In the end, according to exit polls, it was most important to caucus-goers that their candidate share their values. And among those voters, Cruz won. Trump won voters who wanted a candidate who "tells it like it is" and "can bring needed change." Rubio won voters who thought electability was most important.
The best poll suggested Cruz had hidden support
The Des Moines Register poll has a better track record than most. It predicted Howard Dean's loss in 2004 and Obama's victory in 2008. But this year, pollster Ann Selzer still found a Trump victory was likely. And the poll got it wrong in other ways: It found Cruz's support was falling and that Trump's was solid.
Still, buried in the details of the poll was enthusiasm for Cruz and Rubio.
Iowa Republicans said they were more enthusiastic about both candidates than they were about Trump, and that Cruz had more knowledge and experience than Trump. A last-minute switch would benefit Cruz, the poll found. And Republicans polled said they'd pick Trump over Cruz in a head-to-head matchup.
To some observers, that boded well for Cruz. They turned out to be right.