Forecasting Ends Tonight
Most forecasters will lock their models at midnight tonight. God willing, the Times Needle will remain entombed.
Networks, newspapers, and other outlets will no doubt provide “live coverage” of Election Day. They will be looking for signs of “momentum” for either party; they will be looking for signs of “trouble” in turnout or ballot access; they will be checking the vibes.
And then, as the polling places begin to close, and precincts and counties begin to report their vote tallies, they will start to report on the state of things.
Some states will report quickly, thoroughly, and with a very wide margin for one candidate or another. The news outlets’ decision desks will call these early, sometimes with only a small percentage of precincts reporting - but the total number of available votes in that state will indicate that it’s not even close.
Some states will report more slowly than others, and with more tension and drama. Some will report their mail-in ballots first because they can start counting them on or before Election Day itself. Some will report in-person voting first because they have to wait for polls to close (or even later) to start counting mail-in ballots. There will be red and blue mirages in all sorts of directions.
We will hear updates with tags like “too early to call” and “too close to call”. Those that are too close for hours may be called on Election Day if the math shows there is no viable path for the underdog to pull ahead. But they may not be called on Election Day, or the next day. Some congressional races may not be fully decided for weeks.
So as you brace yourself for whatever comes next, whenever it comes, let’s talk about exit polls.
A brief (and incomplete) history of exit polls
The first scientific exit poll was conducted by CBS News’ Warren Mitofsky in 1967 in a Kentucky governor’s race. Mitofsky was a statistician hired out of the US Census Bureau. He designed a random sampling method for intercepting voters as they exited the polling place – an approach inspired by market research for movies (grab a moviegoer on their way out of the theater and ask them what they thought). They would be invited to complete a survey about who they voted for and why. The 1967 experiment was a success, and Mitofsky expanded his operation to 21 states in the 1968 election, and nationally in 1972 – this inspired other networks to mount their own exit polling operations.By 1980, exit polling was common - every major outlet did it, and there were exit polling had gone national.
But there is such a thing as too much polling. Imagine competing news services setting up interviewers with clipboards and survey forms at major polling stations, competing for voters to interview. And it was expensive. So in 1990, Mitofsky left CBS to run the Voter Research Service, a consortium of the networks came together to conduct one national exit poll they could all use. Now, the networks pretended that they all had their own proprietary exit polling operation; and for a few cycles the exit polls were accurate enough that most of the Election Night reporting began to feel like a pantomime of suspense. Jeff Greenfield of ABC news talked about how the anchors and reporters would know the outcome of the election before they could officially announce it, but they had to pretend they didn't. He said the experience was like live reporting the World Series after learning how every part of the game would turn out.
In 1994, the VRS, was reconstituted as the VNS, with 5 news networks and the AP involved.
And then it was the 21st century, and things started to change. In 2000, Florida exit polls led to two miscalls, and possibly contributed to the chaos surrounding the Florida recounts and the ultimate intervention (or interference, if you like) of the Supreme Court in Bush v. Gore. In other places, the results were wrong, sometimes by as much as double-digits.
In 2002, the VNS’ database couldn’t handle the data load. They couldn’t vouch for the integrity of the data they were crunching, and they were running later than the AP’s reporting of official vote tallies. If the whole point of exit polling was first to provide Election Night commentary, context and color for the audience as the official votes were counted, and then to give audiences the impression that the networks were prescient or had inside knowledge with which to “scoop” the official vote, then this type of failure would prove fatal.
VNS was replaced by the National Election Pool in 2003.
In 2004, response rates to exit pollsters plummeted. People didn’t want to fill out a form or talk to a pollster after voting. Not enough interviewers were deployed; many were untrained. They did not follow the random sampling method laid out in the methodology. The printed questionnaires included the logos of the networks involved in the consortium – by 2004 we were already living in a highly partisan media environment, so some respondents may have declined to talk to media brands they didn’t like or trust. Some election boards placed constraints on where exit pollsters could set up their stations – the farther from the polling place, the lower the response rate.
But other changes in the electorate were also unfolding. By 2004, Oregon had gone to exclusive vote-by-mail; in Florida, voters used absentee ballots in large numbers. You can’t poll a piece of mail - so this made it impossible to conduct in-person exit polling in 100% vote-by-mail states; and skewed results in states or precincts with large numbers voting absentee.
That year, the exit polls leaked early and called the election for Kerry. Bloggers - now a dominant source of information in the news ecosystem - got hold of the leaked numbers and reported them; and newspapers and networks did as well. Obviously, in the final tally, Bush won by about 3 million votes, or about 2.5%. But this failure of polling built on the failures of 2000 and 2002 – and began to solidify conspiracy theories about vote fraud on both the left and the right, built on a belief that there was some nefarious discrepancy between the exit polls - which had been pretty accurate for a couple of decades - and the official vote tallies. Were votes stolen? Or were exit polls rigged?
The Modern Era
After the 2017 elections, the National Election Pool split - Fox News and the Associated Press left to form their own exit polling organization. Known as AP Votecast, it is designed by NORC at the University of Chicago with the AP, and is underwritten by the AP and Fox News. Their method now is much more at a distance – less setting up tables at polling places, and more surveying voters online. They use a random, probability-based sample of registered voters to, in their words, “calibrate” an online opt-in panel of likely voters who take an online survey. They are able to survey tens of thousands of likely voters this way.
Their chief competitor is Edison Research’s National Election Pool. Since 2018, the NEP is made up of ABC, CBS, CNN and NBC, along with Reuters. Given the complexity of modern voting, Edison polls different kinds of voters differently.
By-mail registered voters are reached via phone, email or text. Interviewers station themselves at early voting centers to talk to voters right after their vote early in-person. They do the same on election day, conducting in-person surveys with voters after they cast their ballots.
The two organizations tend to get similar results. But in 2020, AP Votecast and the Fox News polling team called Arizona for Biden on Election Night; it took 9 more days for Edison/NEP to make the same call, because their data indicated later mail-in ballots were breaking for Trump.
What to Expect From Exit Polls Tomorrow
Tomorrow is to AP Votecast and Edison researchers what April 15 is to accountants. They will do their absolute best to provide the networks and news outlets with data to help provide context as the vote counts come in, and to enable earlier calls in easier places. Exit polling information will not be reported before 5pm Eastern on Election Day.
But battleground states usually earn their names: elections in those states are close, and campaigns fight it out down to the last hours of voting.
Already we have news outlets telling us that some of the swing states may be called before midnight Eastern tomorrow, and some may take days or even weeks to be confirmed.
Can you trust the exit polls? Sure. The same way you can trust any other kind of poll. They’ll be right – within a margin of error. But they’ll also be wrong. All that color and context, mining the results for narratives about which demographic groups voted for which party, or which issues were most salient to a group of voters? They’ll harden into established narratives quickly, even if they subsequently turn out to be complete bunk.
You might have heard that 53% of white women voted for Trump in 2016. The actual number was 47%. But the “more than half” story stuck with us.
This is one of those times when living with an 18th-century constitution is challenging for modern ways. We believe we are entitled to know things quickly. We believe that if we can know who won the game last night, then we can know who won the election. But we can’t. There are 50 states with dozens of counties and municipalities, each with its own ways of conducting elections. Their constituents vote at different times, using different methods; their election officials count ballots at different times, using different methods. Nothing is consistent, and the country is large. So, you can not expect to know things with certainty or accuracy on election night. By midnight Eastern time, I’m not even sure Alaska and Hawaii are done voting.
The way we fill that void — the collective gaping maw of need for certainty — is with “data”. That data is mostly used to give the commentators and news anchors something to chew on while they entertain you with soundbites and motion graphics. But even though it’s data, it might not be information. And I think the distinction matters.
This is the period in which the saying, “polls don’t vote” is at its truest. The votes will be the votes. The only thing left for the professionals to do is count them.
The only thing left for you to do is decide what you want to do with the time it will take for them to be counted. I’m one of those psychos who will be watching the returns all day and well into the night. But that doesn’t mean I think you should.
I think you should vote if you haven’t already, then go home, pop some popcorn, watch an old movie, and go to bed early. Whatever our democracy will look like in a week, or a month, or by January 6… I’m pretty sure you’ll need your rest.
Sources:
“When to Expect Results in Each State on Election Night, and Beyond” https://www.nytimes.com/interactive/2024/us/elections/results-timing-presidential-race-calls.html
“How we survey the electorate with AP VoteCast” https://www.ap.org/elections/our-role/ap-votecast/
“The problem with exit poll takes, explained”: https://www.vox.com/21552679/exit-poll-accuracy
“The big split persists in how national news outlets call presidential elections” https://www.poynter.org/commentary/2024/how-news-organizations-call-election-results/
“How exit polls work and how NBC News uses them on election night” https://www.nbcnews.com/politics/2024-election/exit-polls-how-work-voters-data-demographics-rcna175474