Arizona presidential election results from MSNBC are displayed during Democratic Senate candidate Mark Kelly's Nov. 3 election-night event in Tucson, Ariz. Like many other news organizations, NBC held off on calling Arizona for Joe Biden, while The Associated Press called it in the early hours of Nov. 4.
Caption

Arizona presidential election results from MSNBC are displayed during Democratic Senate candidate Mark Kelly's Nov. 3 election-night event in Tucson, Ariz. Like many other news organizations, NBC held off on calling Arizona for Joe Biden, while The Associated Press called it in the early hours of Nov. 4. / Getty Images

Just before midnight on the East Coast on election night, Fox News called Arizona for Democrat Joe Biden.

It was a bold call. It opened up a wider path for Biden to win the presidency after a night that began with a lot of bad news for the former vice president. Florida had been called for President Trump, and other states went for the president by wider margins than expected.

When Fox made its call, Biden was up 9 percentage points in Arizona, a state no Democrat had won in more than 20 years. Almost three hours later, with Biden up 5, The Associated Press followed suit, and because NPR relies on the AP for election results, it meant we too were reporting that Biden had won Arizona.

At that point, an estimated 80% of the vote was in, and Biden was up by a margin of 136,000 votes.

"It was a very comfortable Biden margin," AP political editor David Scott said, noting that Arizona reported about two-thirds of its total vote shortly after 10:30 p.m. ET. "We wanted to see more before moving ahead with a call, so we waited several hours."

But as the hours and then days passed and none of the other networks had called Arizona, nerves were becoming frayed.

"I was trying to sleep Wednesday morning, and let's just say I got woken up," Scott said.

On the other end on the phone was his boss, Sally Buzbee, executive editor at the AP, who wanted to know what was going on.

"It was a tense week for many, many reasons," Buzbee said, "but, yeah, we closely watched Arizona every step of the way."

Biden's margin shrank steadily and drastically. Newsrooms around the U.S., which rely on the AP, were also getting nervous.

It wasn't until Nov. 12, nine days after Election Day, that the other networks with decision desks — NBC, ABC, CBS and CNN — called the state for Biden too.

The margin had shrunk to just over 10,000 votes, but the networks were confident that Trump, at that point, could not catch up with the small amount of remaining vote out in a heavily Democratic county.

In making race calls, the AP has been traditionally known for its caution. It is sometimes deliberately slower than other networks, because it likes to say it isn't making projections but, rather, calls based on solid math. Editors at the AP stress that they make a call only when they have determined that a candidate has no remaining path.

That's partially why the early Arizona call surprised some election watchers.

"It is not unusual in the history of race calling for you to sweat it out a little bit after you call a race to make absolutely sure it's OK," Buzbee said. "Certainly, we watched very closely all week. We were prepared to do the right thing if it had flipped. We're not going to stand by a wrong race call. If it got to a point where the race call would have flipped, we would have changed the call. We believe in intellectual honesty, and we would have changed it."

How polling factored in

One thing Fox News and the AP had in common was the use of AP VoteCast this year in place of the traditional exit polls used by the other networks. VoteCast is not an exit poll taken mostly as people leave their voting center, but a massive preelection survey of roughly 140,000 people across the country and in key states, conducted remotely.

The survey, administered by the reputable National Opinion Research Center (NORC) at the University of Chicago, was conducted starting the Monday before election week through Election Day.

Big portions of the data from VoteCast and exit polls come in before polls close and give a general idea of the direction of the race and the shape of the electorate. NPR published some of the national data in real time. Combined with actual vote results, understanding the kind of vote that's outstanding and voting history and trends, VoteCast (and exit polls at other networks) are part of race callers' special sauce to call or project a contest.

And in Arizona, "those results also generally matched the outcome that was estimated by AP VoteCast," Buzbee said.

VoteCast was put to the test in special elections over the last few years and in the 2018 midterm elections. It performed well, and the AP and Fox News were confident enough to unveil it for the first time in a presidential election. (The cable network refers to it as the Fox News Voter Analysis.)

But like other preelection polls this year, the initial numbers appeared to underestimate Trump's support.

Scott said VoteCast was just one piece of the puzzle that his team used but "wasn't the primary piece."

What's more, he added, "we don't use the poll data to make calls that are closely contested."

He said that VoteCast and exit polls are best used as storytelling tools, and while VoteCast — and exit polls — were off this year, they are weighted to match the actual results and eventually be helpful at understanding who voted and how.

Early vote, the pandemic and lessons learned

Ultimately, the AP's Arizona call wound up being right. Scott said even if he could go back and do it again, he would not change the call.

"I think that we did our best job that we could in the moment to make the call that we thought was right," Scott said, "and we were confident that that call would hold up, and it did."

Scott said his team expected the vote count to tighten, and Buzbee pointed out that the AP was ready to reverse the call if necessary, but it never felt it got to that point.

"Every day, we analyzed the returns as they arrived," Buzbee said. "Trump was winning more votes than Biden in those late returns but not doing so by a large enough margin that it would change the outcome of the race. At no point in our analysis did we think it would change the race call, not enough [for Trump] to win the state. So we stood by the call."

There were a few reasons the margin closed more than was perhaps expected.

The call "was informed at least, in part, by voting patterns from the state in the past," Buzbee said, "and ballots counted after Election Day, in the recent past in Arizona, had trended more Democratic than they did this year. And so we were definitely dealing with pandemic effects that were definitely changing historic patterns."

And then there was all that early vote.

"Early vote factored into every race call this year in ways and in levels that it hadn't before," Scott said, "and we knew that and we talked about that in our conversations ahead of the election. When you go from 42% voting in advance to 70% voting in advance, what took place this year was just a fundamental change and a historic change in how America conducts its elections."

How calls are made by network decision desks has mostly been closed off to the general public. It's a competitive industry, and no one wants to give away something proprietary in their modeling.

Plus, it's confusing, and most people just care about who ultimately wins. The AP, like the TV networks, historically doesn't go into great depth to explain race calls. But this cycle, the AP said it purposefully made an effort to pull back the curtain.

With trust in institutions and the media at precariously low levels, that's going to be even more important going forward, as how people vote and who will be voting changes rapidly.

"We want to make race calling less of a black box and more transparent for the public," Buzbee said. "We deeply believe in demystifying this process. ... We take our responsibilities to democracy and the public enormously seriously."

Copyright 2020 NPR. To see more, visit https://www.npr.org.