Come Tuesday night, millions of Americans will be glued to their TV screens or refreshing their browser windows to see the latest election returns, all in anticipation of a final race call. (Though we might won’t know the next president until days later.)
Counting ballots can take a while, but news organizations don’t necessarily need to wait for every cast ballot to be counted before announcing a winner. They’re often able to declare who won without the full returns, thanks to the work of teams colloquially known as “decision desks” — groups of political scientists, statisticians, pollsters, and reporters who use mountains of data, statistical models, and on-the-ground reporting to understand which candidate is up where, and how likely a candidate is to win a given precinct, county, or state.
Given the doubt that former President Donald Trump has sown over the past eight years, both about the election process and the media, it’s worth understanding in detail how the processes of projecting and calling election results work, and why consumers of news should trust those results.
“Remember that we don’t elect anybody,” Anthony Salvanto, who as CBS News’s executive director of elections and surveys, oversees the network’s decision desk, told Vox. “The voters do that. Elections officials are reporting the vote, and what you’re getting from us and the networks is our analysis of what they’ve reported, as well as our first-hand reports from talking to voters.”
To figure out who won an election, news organizations like Fox News, CNN, the Associated Press (AP), and others use a combination of data from election officials, statistical modeling, and polling and surveys of voters.
Raw vote counts come in at the precinct, county, and state levels, and these help decision desks both ensure voting is in line with their expectations and to make decisions on tight contests. Those expectations are shaped by statistical models based on history and other voter information, like geographical location, gender, age, and more.
This year, there are two main systems that news media will rely on for their projections.
The AP and Fox News use a system called AP VoteCast, which debuted in 2018 and has been used in every national election since. In a shift from past practice, VoteCast doesn’t rely on exit polling, and instead uses large-scale online surveys of registered voters who are chosen randomly from a probability-based sample, in an attempt to get the most accurate information from the most representative sample.
A different method is used by the National Election Pool (NEP), which provides data to ABC News, CBS News, CNN, and NBC News. The NEP relies on Edison Research to conduct three types of surveys: Election Day exit polls, in-person early voting exit polls, and polls of likely voters to capture data from those likely to vote by mail, Rob Farbman, executive vice president at Edison Research, told Vox. (AP and Fox News used to be part of this group, but left after the 2016 election.)
(Decision Desk HQ, a private company that contracts with news organizations including the Economist and The Hill — and Vox.com in 2020 — doesn’t use voter surveys, and instead it relies on a proprietary statistical methodology to project winners.)
Each outlet and agency creates their own criteria for interpreting these results.
Sometimes, that can lead to one decision desk getting ahead of the others, like in 2020 when Fox News’s decision desk head Arnon Mishkin called Arizona for President Joe Biden much earlier than any other news source, including the AP, or when Decision Desk HQ called the race far ahead of other experts.
Overall though, when it comes time to make a call, “Our decision team will examine all of the models we are running, consult with the networks’ decision teams, and consider any possible data issues to ensure that the possibility of our call being incorrect is sufficiently small,” Farbman said. “We generally will not make a call unless we are 99.5 percent confident in the call.”
Similarly, the AP doesn’t call an election until “we are confident that there’s no chance the trailing candidate can catch up,” according to David Scott, the AP’s vice president and head of news strategy and operations.
The combination of inputs allows the services to accurately understand who has won each of the around 5,000 elections taking place this year, from the presidential race to local contests and ballot measures. And they can do it quickly, without having to wait on election officials to count each vote. That’s true even in the case of a tight race (like the presidential race is expected to be), though calling those is a bit more complicated.
“If you get a very close race, then you’re looking at where the outstanding vote is, the vote that hasn’t yet been reported, and you’re looking at the kind of places that the outstanding vote is from,” Salvanto, of CBS News, said. “You’re looking at whether it is a mail vote or Election Day vote, if there are any differences in the patterns that you’ve seen by ballot type.”
Along the way, news organizations keep viewers up to date as the polls close and votes come in, showing the public the data that’s being used to make the calls is accurate.
“We will tell you if our models show that it’s a toss up or that it’s leading one way or the other,” Salvanto said. “We will show you, in real time, where the counted vote is coming in — from which counties, which areas of the state, and where it’s still outstanding, where we know there are registered voters, and we know there are still reports to come, so that the viewer can see the whole picture, the way that we see it.”
Of course, these methods aren’t perfect. Very occasionally, news organizations call a race wrong. The most dramatic instance was in 2000, when news networks initially called Florida for Al Gore. Errors do happen — decision desks are made up human beings, after all — but when they do, organizations work to correct them as quickly as possible. Still, mistakes are incredibly rare, so come Election Day (and the days after) you can be confident you’re seeing the real results.
