The Iowa Caucuses Are In 4 Days. Almost Anything Could Still Happen.

As much as we try to remind you all about how uncertain elections can be — pleas that sometimes fall on deaf ears — it’s important to keep in mind in advance of the Iowa caucuses. To begin with, primaries are much harder to poll than general elections, and caucuses are even harder to poll than primaries, as they introduce a number of complications. Caucuses require a long time commitment, which can make turnout harder to predict. They aren’t a secret ballot, so voters can literally try to persuade their neighbors to change sides. And Iowa Democrats employ a viability check — we’ll talk about that more before the caucuses on Monday — that asks voters to switch candidates if their first choice doesn’t clear a certain threshold, usually 15 percent of the vote at that caucus site. That makes second choices important, and even opens up the possibility of strategic alliances between the candidates.

Could Sanders Sweep Iowa and New Hampshire?

So even though Bernie Sanders and Joe Biden are doing a bit better than other Democrats in Iowa polls, everyone would need a bit of luck to win the caucuses. Per our model, as of 8 p.m. Wednesday, Sanders had a 37 percent chance of winning the most votes¹ — which is the best in the field. But it also means there’s a 63 percent chance he won’t win. Our model forecasts Biden to win 35 percent of the time, meanwhile, followed by Pete Buttigieg at 16 percent, Elizabeth Warren at 9 percent and Amy Klobuchar at 3 percent. Even Klobuchar isn’t that much of a long shot. Her chances are about the same as — let’s go with a football analogy since the Super Bowl is this weekend — Brett Favre’s chance of throwing an interception on any given pass attempt.

[Our Latest Forecast: Who Will Win The 2020 Democratic Primary?]

Moreover, the plausible range of vote shares for the candidates varies widely. For Sanders, for instance, the 80th percentile range — meaning that in 80 percent of simulations, his numbers wind up somewhere in this range — runs from 11 percent to 44 percent of the vote. That also implies there’s a 10 percent chance he finishes with less than 11 percent of the vote and a 10 percent chance he finishes with more than 44 percent.

These ranges will narrow slightly by the time we actually get to caucus day on Monday. And our model will be more confident once we get to states that hold primaries, because those are easier to poll.

But the reason for these wide ranges is not to cover our asses. Instead, it’s strictly empirical. Take a look at what polls said a few days before previous Iowa caucuses and you’ll find they were sometimes much less correlated with the results than you might assume. Instead, there was often late movement, election night surprises, or both.

More precisely, I ran retroactive versions of our polling averages for all Iowa caucuses dating back to 1988, excluding the 1992 Democratic nomination, where the other candidates essentially ceded the caucuses to Iowa U.S. Senator Tom Harkin. (Before 1988, polling of the caucuses was fairly sparse, so we can’t really calculate a polling average in the same way that we do today — although there were a number of upsets, like George H.W. Bush coming from way behind to beat Ronald Reagan there in 1980.) I’ll compare what the polls looked like after allocating undecided voters² as of 12:01 a.m. four days before the caucuses — so as of just after midnight on Thursday in years where the caucuses were held on a Monday, for example — against the actual caucus results.³

You can find all of this data in a table at the end of this article. But just to show how frequent surprises are — really, it’s a surprise when there isn’t a surprise or two! — let me briefly run through each race in reverse chronological order.

• Polls at this point in the 2016 Republican race had Donald Trump at 31 percent, Ted Cruz at 26 percent and Marco Rubio at 14 percent. The actual results were Cruz 28, Trump 24, Rubio 23 — so while the polls nailed Cruz, they were 7 percentage points too high on Trump and almost 10 points too low on Rubio.
• Polls in the 2016 Democratic caucuses did much better. Hillary Clinton led Sanders 49-46 at polls at this point, and she wound up winning the caucuses 49.9 percent to 49.6 percent, although it was close enough that you could essentially call it a tie.
• The 2012 Republican caucuses brought another surprise, however. Rick Santorum was at just 11 percent in polls at this point — in fifth place — but he wound up winning the caucuses (after a vote-counting dispute) with 25 percent of the vote, narrowly ahead of Mitt Romney.
• Polls in the 2008 GOP race were quite accurate, predicting Mike Huckabee’s victory there and Romney’s second-place showing.
• But the 2008 Democratic race saw a late surge for Barack Obama. He was trailing Clinton 31-27 in the polls four days before the caucuses, but the final Des Moines Register poll showed Obama moving ahead. That proved prescient, as he won the caucuses with 38 percent of the vote to 30 percent each for Clinton and John Edwards.
• The 2004 Democratic caucuses saw even bigger last-minute swings. At this point, the polls showed the race as: Howard Dean 28 percent, Richard Gephart 25, John Kerry 20, Edwards 15. The actual finish bore almost no relation to the polls; it was Kerry 37 percent, Edwards 33, Dean 17 and Gephardt 11.
• In the 2000 Republican race, polls showed George W. Bush leading Steve Forbes 50-23. The actual result was much closer: Bush 41 percent, Forbes 31 percent.
• Polls in the 2000 Democric race were spot on, showing Al Gore leading Bill Bradley 62-38; he actually won 63-35.
• In the 1996 Republican race, our average four days before the race would have shown Bob Dole ahead with 39 percent, followed by Forbes at 24 percent and Phil Gramm at 11 percent. Although Dole did win, he did so with only 26 percent of the vote. The second- and third-place candidates were surprises, meanwhile: not Forbes and Gramm, but Pat Buchanan (23 percent) and Lamar Alexander (18 percent), who had each been polling at just 9 percent at this point.
• In 1988, GOP polls again correctly predicted Dole to win. But Pat Robertson, who was in third place at 13 percent (trailing vice president George H.W. Bush in second with 30 percent), wound up beating Bush 25-19 for second instead, behind Dole’s 37 percent.
• On the Democratic side in 1988, Gephardt was correctly identified as the winner, receiving 31 percent of the actual vote as compared to 29 percent in polls. But Paul Simon, at 18 percent and in third place in the polls, finished in a surprising second place with 27 percent, ahead of Michael Dukakis at 22 percent.

So out of 11 races, we have only three cases (the 2016 and 2000 Democratic caucuses and the 2008 Republican caucuses) where the polls at this point were more or less spot on. And two of those three cases were in races where there were essentially just two candidates, which are sometimes easier to predict. Every other race featured some kind of late polling movement or election night surprise involving the top three candidates.

The trick is that … there doesn’t seem to be much of a pattern in which candidates surge and which ones don’t. In 2004, the establishment candidate, Kerry, overtook the insurgent liberal in Dean. But in several other races (say, Bush in 1988), the more establishment-type candidates underperformed. Sometimes, the late movement accelerated an existing trend, and sometimes it reversed one. Trump had been gaining ground in polls right up until a few days before the 2016 caucuses, for instance, before the last round of polls showed his lead slipping — and he further underperformed on caucus night. You can find several examples of Midwestern candidates overperforming on caucus night — but one of the biggest underachievers, Gephardt in 2004, was from Missouri, a neighboring state.

[Our Latest Forecast: Who will win the Iowa caucuses?]

So you should be prepared for surprises on caucus night — and they may be genuine surprises, not necessarily the surprises that you’re hoping for or the ones that are easiest to conceive of right at this moment.