The Washington PostDemocracy Dies in Darkness

Why were the Iowa polls so wrong?

Placeholder while article actions load

During a splashy event in Des Moines on Saturday night, Bloomberg News invited reporters to witness Ann Selzer unveil her final poll results, an event with all of the fanfare of Oscars nominations. The Selzer poll, sponsored by Bloomberg Politics and the Des Moines Register, has a track record of nailing the results of Iowa elections, including the tricky calculus that goes into figuring out the weird process of the caucuses. Selzer's poll showed that Donald Trump's surge past Sen. Ted Cruz of Texas earlier in the month was real, and it gave him a five-point lead over Cruz.

On Monday night, when the counting was done, Cruz won by three points. What's more, Trump very nearly fell to third place, with Sen. Marco Rubio of Florida nipping at his heels. So what went wrong?

It's hard to say with certainty, but there are a few things that are obviously problematic.

Rubio surged at the end

People tend to take the Selzer poll as a stand-alone augur of what's to come, given its track record. But if you look at an average of the polls that were conducted, such as the one compiled by RealClearPolitics, you can see the late Rubio surge pretty clearly.

He saw a big spike at the end, yes, but it's clear that for the past week or so he'd been climbing quickly.

Entrance poll data shows that, among those deciding on a candidate in the past few days before the caucuses, Rubio took a plurality of support.

Remember, we've seen this before. In 2012, former senator of Pennsylvania Rick Santorum trailed by a lot about two weeks before the Iowa caucuses. But he came on strong in the last few days and eked out a close win.

Primary polls are less accurate than general election polls

Back in 2012, FiveThirtyEight's Nate Silver warned America about the accuracy of primary polling. "On average," he wrote, "a poll conducted on the day just before the election [in 2012] has missed the final margin between the candidates by about four percentage points." That's on average -- meaning that many polls from the day before missed the margin by more.

There are a lot of reasons for this, including the diversity and number of pollsters. But that the results should deviate from a poll conducted a few days out isn't a surprise.

There were plenty of warning signs in the Selzer poll

When we looked at the Selzer poll Saturday, we noticed that the details of it, the numbers under that Trump lead, were not very convincing for him.

The problems for Trump, inside the numbers, included the fact that his net favorability -- those who viewed him positively vs. those who viewed him negatively -- had consistently sunk in Selzer polling to the point that he was barely breaking even. About half of Iowa Republicans viewed him positively, about half negatively. His support was also fervent but shallow. He led as the first choice of voters -- but when you combined first and second choices, he tied for third -- with Rubio.

Iowans also reported growing less comfortable with the thought of Trump as nominee (and president) over time.

While Selzer's prediction for how much of the electorate would be voting for the first time was a little low, Trump was heavily dependent on that vote. On caucus night, he won that group by seven points -- but had led by 16 points in the poll.

Cruz had a stronger field operation

Since it became clear that Trump was running a serious campaign, there was one looming question: Could he actually put together a real campaign, organizationally speaking? Trump hired field staff and sent people across the state. There were rumblings that his operation on the ground was weak, but the proof was in the caucusing.

Cruz, meanwhile, had a robust field operation, including putting up volunteers in a dorm so that they could canvass voters day after day. On the night of the caucus, 36 percent of Republican voters indicated that they'd been contacted by a campaign, according to entrance poll data, and Cruz won that group by eight points. Among those who hadn't been contacted by a campaign, Cruz, Trump and Rubio were essentially tied.

Field programs are often derisively called the "field-goal unit" because the margin they can move voters is generally pegged at about three points. But a strong Cruz ground operation combined with a weak Trump one could see more of a difference.

People -- particularly those whose candidate underperformed -- like to assume that all polling is inherently flawed. In this case, observers may have made too much of one poll and played down a pretty clear late surge from Rubio.

If it's any consolation to dispirited Trump fans, remember that in one week, New Hampshire votes, and Trump's poll numbers there are much, much better.

For now.