Welcome to your first FiveThirtyEight Election Update of the 2020 primary cycle! This is a column in which we’ll talk about the primary race through the lens of our forecast model, which we released earlier this week. Sometimes it will be rather brief, and we’ll quickly run through the latest data — while other times, we’ll go into a deep dive on upcoming states or some aspect of how the model works.
Biden, Sanders neck and neck In Iowa
We don’t necessarily plan to publish an Election Update as a result of each single new poll, but Friday’s Selzer & Co. poll of the Iowa caucuses, published by the Des Moines Register and CNN, warrants an exception and did have a somewhat material effect on the model.
Why is it worth focusing on this one individual poll — something that we’d usually advise against?
- Selzer & Co. is a very good pollster, one of the best in the business.
- There haven’t been a lot of polls of Iowa recently.
- Iowa is pretty darn important, at least in terms of how our model thinks about the race, with the potential to produce fairly large bounces that will affect the rest of the calendar.
The poll showed Bernie Sanders ahead with 20 percent of the vote, followed by Elizabeth Warren at 17 percent, Pete Buttigieg at 16 percent and Joe Biden at 15 percent. This is a reasonably big shift from the previous Selzer & Co. poll, in November, which had shown Buttigieg ahead with 25 percent of the vote. (Although, for reasons I’ll get to in a moment, the model views the latest poll as more neutral than negative for Buttigieg.) Amy Klobuchar was next in the poll at 6 percent, but that was unchanged from November despite a couple of debate performances since November that voters rated strongly in our polling with Ipsos. Andrew Yang was sixth at 5 percent.
So then, how did the new poll affect our model? Here’s what our current national numbers look like:
Biden remains the most likely candidate to get a delegate majority, with a 38 percent chance, followed by Sanders at 24 percent, Warren at 13 percent, and Buttigieg at 10 percent. There’s also a 14 percent chance that no one wins a majority, which could potentially lead to a contested convention.
But those numbers do represent an improvement for Sanders and Warren and a decline for Biden. Here’s a before-and-after comparison:
|Candidate||Last model run before Selzer & Co. poll||Current forecast|
Biden’s majority chances fell by 3 percentage points, from 41 percent to 38 percent, while those of Sanders and Warren each gained 2 percentage points. Buttigieg’s chances were unchanged.
I really like having a model at times like this because it allows for a fairly rigorous and objective answer to the question: How much should I update my priors as a result of this new piece of information? If you’re just winging it, it’s super easy to screw that up in either direction, either dismissing new data as being “an outlier,” etc. — or claiming that the new data has massively inverted the trajectory of the race when it probably hasn’t. (The latter is usually the more common mistake in media coverage of the campaigns since it makes for more dramatic headlines.)
In FiveThirtyEight model terms, swings of this magnitude — Biden falling from 41 percent to 38 percent — are a relatively big deal. They will likely be on the high end of the shifts you see as a result of a single state poll, with the possible exception of final polls conducted on the eve of a primary or caucus. (Let me back up with a caveat: I think that this will be on the high end of poll-induced swings based on what we’ve seen in our past general-election models, but since the primary model is a new product for us, I’m not quite sure.)
At the same time, if this poll has completely upended your view of the race, then — I’m trying to put this constructively — you need to go back and add a little more rigor to your mental model of the primaries. Iowa still has four highly plausible winners; that was true both before and after the poll. Our model has Sanders (with a 29 percent chance) and Biden (also with a 29 percent chance) as being a bit more likely than the others to win, but it’s not really much of an edge. (We have Buttigieg’s chances at 22 percent and Warren’s at 16 percent.) Perhaps the candidate who had the most reason to be disappointed by the new poll was Klobuchar. Making a very late surge to win Iowa is not completely out of the question — Rick Santorum did it in 2012 — but we have her chances down to 2 percent.
Biden remains the most likely overall winner of the delegate race, meanwhile, with Sanders in the next-best position. That’s because Biden, leading in national polls, would be awfully hard to catch if he won Iowa. For the other three candidates, there would be the question of whether the Iowa bounce would be enough to propel them past Biden, with Sanders being in the best position to do so because he’s second in national polls and because his polling is also relatively strong in both New Hampshire and Nevada.
It’s also worth keeping in mind that we do have some other recent information about Iowa apart from this poll. A YouGov poll of Iowa released last weekend showed a three-way tie between Biden, Sanders and Buttigieg. And our model also makes inferences about candidates’ standing in Iowa based on trends in national polls. That’s the reason the model didn’t have Buttigieg’s chances falling as a result of this poll; it had already anticipated that his numbers would decline as a result of his slump in national polls. Conversely, even though the numbers didn’t seem that terrific for Warren on the surface — her 17 percent in the new Selzer & Co. poll is only a 1-point improvement from her 16 percent in November — it comes during a period when she’d been declining in national polls. So it’s a bullish sign for her campaign that she’s still one of the front-runners in Iowa.
By the way, “one of the front-runners” is about as precise as it’s possible to realistically be in Iowa. Our forecast will get a bit more accurate as more polls come in and as the Feb. 3 caucuses approach, but the model assumes that caucuses are awfully hard to poll, which means there are high margins of error.
That’s especially so in Iowa given some of the quirks of the caucus process, the most important of which is that in each precinct, voters for candidates who don’t have at least 15 percent of the vote must “realign” themselves to candidates who do. Iowa will also release three different ways of counting its vote. More about that stuff in future Election Updates. And although I’m not going to get into it today, some of the data from the poll that the model doesn’t use — like favorability ratings and second-choice preferences and how many voters have firmly decided on a candidate (not many, although Sanders supporters are something of an exception) — should contribute to the sense that the race is open-ended.
All of that is a long-winded way of saying there’s a lot of ambiguity about what will happen in Iowa. Through that fog, our model picked up some good news for Sanders and Warren and some bad news for Biden in this poll. But the fog is pretty dense.