A version of this story appears in CNN’s What Matters newsletter. To get it in your inbox, sign up for free here.
CNN
—
Millions of Americans will be tuning in to CNN and other networks on election night as returns roll in. They’ll be waiting to see who wins, something CNN will determine with a series of projections.
With polls so close this year and a few hotly contested states set to determine the outcome of the presidential election, everyone should be prepared to go to bed on election night not knowing who will win the White House.
I talked to Jennifer Agiesta, CNN’s director of polling and election analytics, about how CNN approaches the job of projecting a winner. It’s a complicated and exhaustive process. Our conversation is below.
AGIESTA: We project races. We don’t call them. We are telling you where we think the race will wind up based on what we see in front of us. That is a projection by any normal definition of the word.
AGIESTA: There are several layers of people involved in projections for CNN.
At the start of it, we have a decision team that’s made up of people who are experts in either politics or statistics, and they’re paired together based on those areas of expertise. They are all reviewing individual states, individual races, making a recommendation on a projection that gets passed up to the decision desk supervisory team – which is two people, one of whom is a statistician, one of whom is a political person.
They review every projection that those individual teams are making, and then they pass it to me. If they’re comfortable with it, I give it a check, make sure I’m good, and then, as long as all three of those layers have gone, we’re good.
We pass it on to the control room and work with them. I answer any of their questions about why we are or are not making projections at a given moment, and then it will find its way to air or online.
AGIESTA: There are a lot of different pieces of information that go into any given projection. There’s obviously vote data. We want to see what’s been counted and what hasn’t, and our expectations of how large those votes were and how they would be distributed based on past votes.
We’re looking, as much as we can, wherever it’s available, at what types of votes we have, whether they’re absentee votes or Election Day votes, and how the balance of those things matches up with what we thought it would be before Election Day.
We’re looking at the geographic distribution of votes. Did they come from a really Democratic county or a really Republican county, or from a place that counts really fast or counts really slow?
Based on what we know about those things, all of that gets dumped into statistical models that tell us what level of confidence we can have with the estimate that it gives us. We also are looking, to some degree, at exit polls in some states and what those are saying about how voters split in a given place.
All of that kind of comes together to make the ultimate projection.
AGIESTA: It starts off being much more exit poll focused. In states with exit polls, we have reporters out in the sample of precincts where those exit polls have been conducted who are collecting votes from those specific precincts. So we can see how the exit polls are performing in real time.
That also is, in a lot of ways, a representative sample of at least the Election Day vote within a state, and so that gives us another marker of how things are going. Those kinds of measures will be more important earlier in the night.
We’ll then start to transition to looking at the actual votes and where things are, and the models become important for a really long time once we’re basing things entirely on what the counties are telling us about what’s counted and what types of votes those are.
Then really late in the night, that’s where the states that are super close that we’re projecting last – those are the ones where we’re looking for the math problem that you and I talked about four years ago, where it’s like, is there enough vote out there that the candidate who’s trailing could catch up? And that becomes more of an algebra problem than a modeling problem.
AGIESTA: I think there will be some states that we can’t project on election night just because of the nature of how they count their votes and what we know about the speed of the reporting.
It doesn’t necessarily mean that there’s anything wrong in those states, or that things have gone haywire, or anything like that. It just is the process of counting these ballots that takes time.
In Pennsylvania, they’re not allowed to process by-mail ballots until Election Day itself, so those election officials have to go through the process of opening every single envelope and unfolding the ballot and feeding it through the scanner and whatever other verification processes they go through. And all of that takes time.
So they do as much as they can during Election Day, and then when the polls close, they tend to stop and come back to what’s left of those ballots later. But it tends to be after Election Day in a place like Pennsylvania.
The same thing kind of happens in Arizona and in Nevada, where you have a lot of by-mail ballots. A lot of them returned later in the process – the “late earlies” that everybody talked about in Arizona (in 2020), they just take some time to work through and actually get them counted.
But other states are super fast, and we can probably get those projected on election night. States with a wide margin will probably be projected on election night, but close states where they have those sort of slower counts are just going to take some time.
AGIESTA: What we’re looking for with our models is 99% confidence. And I believe the exact figure is 99.7% confidence from those models that we actually have the statistical backing to make any projection.
It really is knowing the full picture of what (votes are) out and what has been counted and how those things are expected to go, making sure that we really know that there’s not some wild thing out there that could totally change the trajectory of what we’ve seen so far.
AGIESTA: Obviously we want to consider every projection we make carefully, and we don’t really have different standards for how we’re projecting these things.
Every projection that we make, we’re going to look for that same level of confidence, level of certainty before we do it, whether it’s a race for president or a race for dog catcher.
There’s a lot more attention on one of those than the other, so we will have more people involved and a bigger set of fact checks on a more important projection. There will be more people making sure, asking every possible question all the way down the line before we project anything that is of serious consequence.
AGIESTA: We expect to see a lot fewer mail-in ballots generally, just because we’re not in the middle of a pandemic. We saw in 2022 that the share of voters who chose to vote through mail-in ballots has dropped back to where it was before the pandemic. I think we’ll see that continue.
That doesn’t mean there won’t be any mail-in ballots. There will be a lot of them, and they still take time to count. But in a state like Georgia or North Carolina, where you had a huge number of mail-in ballots in 2020, that’s dropped back to a lower share, with most of the preelection voting happening early in person.
In North Carolina specifically, they no longer accept mail-in ballots after Election Day, even if they were postmarked by Election Day, so you won’t have that uncertainty about the number of mail-in ballots that could be out there.
I think in Pennsylvania, it’s all but two of the counties there have taken grant money that requires that they count continuously through the end. So rather than having some counties that might count during the day and then stop overnight, they’re going to be counting around the clock.
Whether they’ll be reporting around the clock is a different question, but it should get them to the end of the count faster, even if it does extend beyond Election Day itself.
AGIESTA: In Georgia, we’ll start to see early in-person votes reported quickly. We’ll see mail-in ballots reported really quickly. So we’ll get a good sense of whether that batch of early votes is as Democratic as it had been in the past, or is there any shifting in the partisan balance of preelection votes versus Election Day votes.
I think we’ll see a similar dynamic in North Carolina.
Within each of those states, we’ll be looking at the margins in specific types of counties. Are the rural counties as strongly (Donald) Trump as they used to be? It’s been a pretty prominent (Kamala) Harris strategy to try to shrink the margins in those rural counties. Is that happening in those counties that count pretty quickly?
At the same time, in the more urban counties in these earlier reporting states, is the turnout looking like it would in an election where a Democrat winds up winning? Or is it looking more like it did in 2016, when urban turnout was a little bit down relative to the norm?
Are the vote divides in those more urban counties similar to what they were in 2020, or do they look different this year?
Those changes year to year in different types of counties are probably where we’d be focusing our attention early on.
Within the states that happen later in the night, in Arizona, in Nevada, Michigan and, to some extent, Wisconsin, there are the key counties that I think everybody who follows elections knows at this point.
In Arizona, it’s all about Maricopa.
In Nevada, it’s Clark and Washoe, because those are where the people live.
In Wisconsin, we’ll be looking at those Milwaukee suburbs. How strong is the Democratic turnout in a place like Dane County?
What are we seeing in an area like Grand Rapids in Kent County in Michigan that’s been kind of swinging over the last few elections?
We have those places identified that are critical to watch, and we’ll see how they turn out over the course of the night.
AGIESTA: I was an actual adult in 2000. That’s probably as close as it’s ever gonna be.
WOLF: But did it seem as close in 2000 heading in? I’m trying to remember.
AGIESTA: I don’t think it did. I don’t know that it was quite as close in so many places. But I mean, it ultimately came down to a few hundred votes in Florida. New Mexico was decided by a few hundred votes. New Hampshire was decided by a very small margin. The places were different, but it was pretty tight.
AGIESTA: It could. I mean, polls are wonderful tools, and obviously I think that they are wonderful measures of how people are feeling, what they’re thinking, how they are making the decisions that they ultimately will make in any given election.
But they’re not precise to the level that they would need to be to tell you that a 1-point shift is happening, or that a margin that looks like it’s going to be a tied race, who actually is up or down in that kind of race. A poll can’t tell you that. We have to wait for the votes to be counted to know.
AGIESTA: (Laughs.) I’m worried about so many things. I think my biggest worry is that it could potentially take a lot longer than we expect, given the tightness of the preelection polls.
In 2020, we were able to make a final projection on the presidency on Saturday. I think the most likely scenario this year is that it’s a little bit earlier than that.
But if it comes down to a situation where we’re waiting on a state that has to go through a recount process, or we’re waiting on a tied Electoral College or something, then it stretches out into possibly December.