Exit Polls Can Be Misleading — Especially This YearNovember 2, 2020
Exit polls are usually a core component of election-night reporting — especially before real results come in. They can provide an early sense of which candidate is leading and how certain key demographics voted. But here at FiveThirtyEight, we won’t be covering exit polls much at all, at least not on election night. That’s because pandemic-related changes to how the exit polls are being conducted will make them even less reliable this year.
Even under ideal conditions, exit polls can be misleading. People often use them to suss out the horse race while waiting for results to get insight into who’s winning, but exit polls aren’t intended to be used this way. Really, what they’re designed to do is help explain how a candidate won after everything is said and done, breaking down a candidate’s support using various demographic variables like gender, race and education. (For example, they made it really easy to tease out how the gender gap between Democratic and Republican support has grown with time.)
But exit polls should not be used as a way to gauge a candidate’s support in real time. For starters, exit poll numbers can — and do — change over the course of election night, so you should be especially skeptical of the earliest exit-poll numbers, typically released around 5 p.m. Eastern. By definition, this first wave of exit polls are incomplete: Polling places are still open, and who’s represented in the electorate could change. For example, people who vote early on Election Day tend to be older than the electorate at large, which can skew the numbers some.
As the night wears on, though, more interviews get added to the sample, and the exit polls become much more representative of voters as a whole. In fact, exit polls are eventually weighted to match the final results, but that only happens after the election is over. On election night itself, the numbers are reweighted as more data becomes available, but what you’re seeing is still subject to a lot of uncertainty — especially this year given that the pandemic has radically transformed the way we run elections.
The pandemic has undermined the one major advantage exit polls have over other kinds of polls: their ability to survey only actual voters, since exit polls catch respondents as they exit polling locations. But in 2020, far fewer people will be voting in person on Election Day than in previous years; according to data from political scientist Michael McDonald, more than 93 million people nationwide have already cast an absentee or in-person early vote. That means that interviewing voters as they leave polling places on Election Day won’t give a reliable estimate of how people will vote, especially as the people waiting until Nov. 3 are likely to skew more Republican than the electorate as a whole.
To account for this, Edison Research — the polling firm that produces the exit polls used by ABC News, CBS News, CNN and NBC News — has changed its methodology so that the “exit polls” you see on Nov. 3 will actually be a combination of traditional exit polls of Election Day voters and phone polls of mail voters. In certain states, there will also be exit polls of early in-person voters, conducted at early-voting locations in essentially the same way as traditional exit polls (just conducted earlier).
But exit polls will still be less reliable this year because the phone poll and the early exit poll are much tougher to get right than the traditional exit poll.
The phone poll essentially puts pollsters back in the uncomfortable territory of having to guess whether respondents who claim to have voted actually did. (In a world in which trying to model likely-voter turnout is half the battle pollsters face, having that certainty in a traditional, non-pandemic exit poll is a big leg up.)
And while the early exit polls are better able to identify actual voters than pre-election polls, exit pollsters don’t have as much experience with early exit polling. That means they’re making less informed decisions about how to weight their samples, which can affect the accuracy of their polls. It’s just really hard to know in advance what the proportion of in-person Election Day voters and absentee voters will be in 2020. So how heavily Edison decides to weight the traditional-exit-poll respondents versus the phone respondents and early voters will add to the exit polls’ uncertainty this year.
What this boils down to is that, this year, there’s no reason to believe that exit polls will be any more accurate than traditional pre-election polls. And so it’s hard to make a case for giving them more weight than the many pre-election polls we’ve been looking at already. We will be using exit polls in the weeks to come to understand how age, gender, race and education shaped voting behavior, once we know the outcome. But if your main question is who will win the election, well, we’d never have recommended exit polls for that anyway. This election night especially, it’s best just to wait for the final results — even if it takes a while.