In the aftermath of Sen. Ted Cruz's (R-Texas) surprise Iowa win, pollsters have been trying to figure out why he and Sen. Marco Rubio (R-Fla.) surged beyond expectations, even as businessman Donald Trump flopped.
The reasons why pollsters got it wrong could be key to understanding how much trouble Trump faces in New Hampshire and beyond, and how accurate surveys of future primaries will be.
There are two main theories:
-- Polls screwed up in predicting which people would actually show up to the caucuses. Most analysts had predicted that high turnout would benefit Trump, but the record turnout instead brought out many of his detractors, including "traditional," evangelical voters, while many of his backers stayed home.
-- Voters changed their minds after most polls were finished. Undecided voters broke last-minute for Cruz and Rubio, and those backing Trump or other candidates decided at the last minute to switch sides.
There is good evidence for both theories thanks to polling from SurveyMonkey that found significant momentum building toward Cruz and Rubio in the final week of polling, as well as an increasing share of evangelicals in the electorate.
Now there's more hard data on exactly how big a role each of those factors played.
Monmouth University -- which, like most other pollsters, found Trump ahead -- went back and recontacted 263 of the Republican voters it had originally talked to, finding evidence that both theories were true.
Nearly a quarter of voters who'd supported Trump in Monmouth's final poll never ended up voting, compared to just 13 percent of Cruz and Rubio supporters.
But the more important factor was the 21 percent of those who did make it to the polls but ended up switching their vote. Those switches largely worked in favor of Cruz and Rubio.
Much of the last-minute shift is probably owed to Cruz's superior ground game. SurveyMonkey found that Cruz supporters were significantly more likely to get letters, emails and phone calls reminding them to vote. (It's also possible that some voters didn't want to admit to pollsters that they'd backed a losing candidate.)
But Monmouth also found that half of the voters who jilted Trump were "unhappy with his candidacy, including both his personality and ... his decision to skip the final debate," which came after most Iowa polls had left the field.
“Donald Trump didn't go to the last debate. That made me think he didn't deserve it," one voter told pollsters.
If Trump's numbers naturally crumble as voters start paying more attention, it "would be the worst-case possibility for Mr. Trump," the New York Times' Nate Cohn wrote. "It would mean his support really might evaporate ahead of future contests, as voters focus on other candidates and as he faces even more attacks. ... If skipping the last debate did it, then he can simply avoid making the same mistake again."
Ironically, despite his well-publicized fondness for touting polls conducted by other people, Trump hasn't done any of the internal polling that could help him answer those questions.
“He doesn’t understand it as a predictive or testing tool,” a Republican strategist told NBC. “He’s only interested in ‘Am I ahead or behind’ -- which is the least important thing if you’re using polling for predictive purposes.”
The reasons for the misfire also highlight a major limitation of primary election polling: While surveys remain pretty accurate on the whole, they're not particularly well-equipped to predict highly volatile races -- like, say, the New Hampshire primary.
"The only thing that would have improved our model enough to reflect the final outcome would have been a time machine," Monmouth's Patrick Murray said in a statement. "The polling industry is tackling a number of methodological challenges in our industry right now. Unfortunately, the ability to manipulate the time-space vortex is not one of the technological innovations we are currently exploring.”
Until pollsters figure that one out, expect the occasional election night surprise.
Monmouth's post-Iowa survey surveyed 263 of the 500 likely Republican caucusgoers polled in their final pre-Iowa survey.