A version of this article appears in CNN’s What Matters newsletter. Sign up for free here to receive it in your inbox.
CNN
—
Millions of Americans will tune in to CNN and other networks on election night to hear the results. They will have to wait and see if CNN decides who will win in a series of predictions.
With polls so close this year and several battleground states set to decide 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. be.
We spoke with Jennifer Asiesta, CNN’s director of polling and election analysis, about how CNN approaches the job of predicting winners. It is a complex and thorough process. Our conversation is below.
Ajiesta: We are projecting the race. we don’t call them. We’ll tell you how we expect the race to end based on what we see in front of us. It is a projection from the usual definition of the word.
Ajiesta: There are several layers of people involved in CNN’s predictions.
Initially, there is a decision-making team made up of experts in politics or statistics, who are paired up based on their area of expertise. They all look at individual states, individual races, and make prediction recommendations that are then passed to the oversight team at the decision-making desk. The team consists of two people, one of whom is a statistician and the other a politician.
They review all the predictions that individual teams are making and pass them on to me. If they’re happy with it, I’ll check on it and make sure I’m okay, and as long as all three of those layers are resolved, we’re okay.
We will pass it on to the control room and work with them. I answer questions about why we do or don’t make predictions at this time. It will then be broadcast on air or online.
Ajiesta: A particular projection contains a variety of information. Obviously there is voting data. We want to see what’s counted and what’s not, and our expectations about how big those votes will be and how they’ll be distributed based on past votes. Masu.
We’re trying to figure out what kind of voting there is, absentee voting or Election Day voting, wherever it’s available, and how those balances differ from what we had in mind. We are trying to find out if they match as much as possible. Before election day.
We’re looking at the geographic distribution of votes. Did they come from really Democratic counties or really Republican counties, or did they come from places where the count was really fast or really slow?
Based on what we know about these things, all of that goes into a statistical model that tells us how much confidence we have in the estimates it gives us. We also pay some attention to exit polls in some states and how voters are divided in certain locations.
All of this combines to create the ultimate projection.
Ajiesta: Initially, we are focusing more on exit polls. States that conduct exit polls send reporters to sample precincts where exit polls are being conducted to collect votes from specific precincts. So you can see how the exit poll results are doing in real time.
This is also in many ways a representative sample of Election Day votes, at least within the state, so it gives us another indication of how things are going. These measures are even more important in the early hours of the night.
Then we start to move into the actual voting and investigating the situation. Once we start basing things entirely on information from counties about what is being counted and what types of votes are being cast, models will be important for a very long time. They are.
Then really late at night, those are the states that are very close that we’re predicting at the end – these are the states that we’re looking for the math problem that you and I talked about four years ago, that’s next. It’s like, will there be enough votes out there for a candidate who is lagging behind to be able to catch up? And it becomes more of an algebra problem than a modeling problem.
ADIESTA: I think there are some states that will be unpredictable on election night because of the nature of how votes are counted and what we know about the speed of reporting.
That doesn’t necessarily mean there’s something wrong with those states or that things are messed up or anything like that. But the process of counting those ballots takes time.
Pennsylvania doesn’t allow mail-in ballots to be processed until Election Day, so election officials must go through a verification process that includes opening each envelope, unfolding the ballot, and running it through a scanner. Pass through. And all of that takes time.
So they tend to do the best they can during voting day, stop when voting is over, and return to the remaining ballots later. But in places like Pennsylvania, it tends to be after Election Day.
Similar things are happening in Arizona and Nevada, where many people vote by mail. Many of them came back later in the process. The “slow early days” everyone talked about in Arizona (2020), it just takes time to process them and get them actually counted.
But other states will be able to get predictions very quickly, perhaps on election night. States with large margins will probably be expected on election night, but states with such slow counts will take time to become close.
Ajiesta: What we look for in our models is 99% confidence. And I believe with 99.7% confidence that the numbers from these models are accurate, they have the statistical support to actually make the predictions.
What’s really important is knowing the full picture of what’s on the ballot, what’s being counted, and how they’re expected to go, which could completely change the trajectory of the election. It’s about making sure that we really know there’s nothing wild about it. we’ve seen this before.
AJIESTA: Of course, we want to carefully consider every projection we make, and we don’t really have different standards for how we project these things.
Every time we make a prediction, whether it’s a presidential election or a dogfight, we look for the same level of confidence and certainty before implementing it.
One of these has received more attention than the other, so the more important predictions will involve more people and undergo more extensive fact-checking. In the future, more and more people will ask every possible question and check before predicting anything with serious consequences.
AJIESTA: We generally expect mail-in voting to be significantly reduced simply because we’re not in the midst of a pandemic. In 2022, we found that the percentage of voters who chose to vote by mail fell to pre-pandemic levels. I think that will continue.
That doesn’t mean voting by mail will disappear. There will be a lot of them, and it will still take some time to count them. But in states like Georgia and North Carolina, which had huge numbers of mail-in votes in 2020, most of the advance voting was done early and in person, and the share of mail-in votes fell again.
North Carolina, in particular, no longer accepts mail-in ballots after Election Day, even if they are postmarked by Election Day, so there is no need to feel as much uncertainty about how many mail-in ballots may be cast. .
I think all but two counties in Pennsylvania received grants that needed to be counted continuously until the end. So instead of counting during the day and stopping at night, some counties will be counting around the clock.
Whether they will report 24 hours a day is another matter, but the count should be completed sooner, even beyond Election Day itself.
ADIESTA: Early in-person voting will soon be reported in Georgia. Mail-in votes will soon be reported. So we’ll have a good idea whether the early voting population is as Democratic-leaning as it has been in the past, or whether there’s been a shift in the partisan balance between early voting and Election Day voting.
I think we’ll see a similar movement in North Carolina.
Within each of these states, we look at the margins for specific types of counties. Are rural counties as strongly supportive of (Donald) Trump as they were before? It was a very prominent strategy of (Kamala) Harris to try to shrink margins in these rural counties. Is something similar happening in counties where the counts are very fast?
At the same time, in the more urban counties of these states previously reported, will turnout be like elections that Democrats ultimately won, or will turnout in urban areas be a little lower than the norm? Is it similar to the situation in 2016?
Will the vote disparity in these urban counties be similar to 2020, or will it be different this year?
These year-to-year changes in different types of counties are probably what we’ll be looking at early on.
In states where elections are held late at night, such as Arizona, Nevada, Michigan, and to some extent Wisconsin, there are important counties that are likely known to anyone following elections at this time. be.
In Arizona, it’s all about Maricopa.
Clark and Washoe in Nevada. Because people live there.
In Wisconsin, we look at the Milwaukee suburbs. How strong is Democratic turnout in a place like Dane County?
What will we see in areas like Grand Rapids, Kent County, Michigan, which have been swinging elections in recent years?
We have identified areas where monitoring is important and we will see how they turn out overnight.
Agiesta: I was a real adult in 2000. You’re probably closer to adulthood than you’ve ever been.
Wolf: But as the year 2000 approached, did it seem like it was coming? I’m trying to remember.
Ajiesta: I don’t think so. I had no idea it was so close to so many places. But 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 narrow margin. The location was different, but it was much smaller.
Agiesta: That might be possible. So it’s clear that polls are a great tool and a great measure of how people are feeling, what they’re thinking, and how they’re making the decisions that they ultimately make in a particular election. I think so.
But they aren’t accurate to the level needed to tell you that a one-point change is happening, the margins that are likely to result in a tied race, and who is actually up or down in it. A type of race. Polls don’t tell you that. We’ll have to wait for the votes to be counted to find out.
Agiesta: (lol) I’m worried about a lot of things. I think my biggest concern is that it could take much longer than expected given the intensity of pre-election polling.
In 2020, we were able to make our final presidential inauguration predictions on Saturday. I think the most likely scenario this year is that it will be a little earlier than that.
But if you’re in a situation where you’re waiting for states that have to go through a recount process, or you’re waiting for tied electors or something, it’s probably going to stretch into December.