A crash course on exit polling
If it’s after 5 p.m. Eastern on Election Day, it’s exit poll time! If you’re watching the news or scrolling social media tonight ,you’re likely to encounter findings from this mega-poll of voters throughout the night (and coming weeks). But there are some reasons to take what you see, especially early, with a grain of salt.
The exit poll is a survey of voters nationally and in key states that asks people questions like who they voted for, when they made up their mind and why they voted the way they did. The exit poll is conducted by Edison Research (a survey research, voter data and election returns company) and sponsored by a consortium of news networks in the U.S. (ABC News is one of them.) ABC News will be reporting exit poll results nationally and in 10 states: the seven battleground states (Arizona, Georgia, Michigan, Nevada, North Carolina, Pennsylvania and Wisconsin) plus Florida, Ohio and Texas.
Long ago, the exit poll only interviewed people in person as they left their polling places after casting ballots in person on Election Day. With the rise of mail-in and in-person early voting, the pollsters and statisticians behind the exit poll added telephone interviews to account for early and mail-in voters.
The exit poll is an incredible feat of engineering and a rich source of data on voter behavior in America, and you can expect it to be cited frequently after the election as a measure of “ground truth” (or at least one of the highest-quality estimates available) about how and why people voted. However, at the end of the day, the exit poll is just a very detailed survey, so it is not infallible. In particular, the way it is conducted warrants two main caveats:
First, the results of the exit poll can and will change over time, especially at the subgroup level. The earliest results of the exit poll are essentially toplines and crosstabs from a very large poll of voters, but which contain a lot of guesses about the eventual demographic composition of the country. As the votes roll in over the course of the election night (or week), the pollsters behind the scenes adjust the results of the poll to match the results of the election.
As a result, the findings of the exit poll can and do change as more votes are counted, especially for subgroups. In Wisconsin in 2020, for example, preliminary exit poll results had Trump +1 among men. In final data (weighted Wednesday morning), Trump was +10 among men. Overinterpreting the poll — say, if you had hitched your horse to the wagon of low or no gender gap in the electorate — risks setting up narratives that are swiftly knocked down with actual data.
The second word of caution is to remember the exit poll is just one imperfect estimate. Yes, it is generated with a lot of data and care, but misspecified weights or sampling noise can impact results just like with any other poll. Interpretations of the final exit poll results should be used with caution. As a result of all these caveats, 538 rarely cites exit polls, and we prefer to wait for actual results before wedding ourselves to narratives. If you are going to cite the exit poll, we recommend waiting until it has been at least somewhat weighted to be representative of actual election results.