While working through this year’s “what to expect on election night” posts, it occurred to me that it would be a good idea to do a brief refresher on exit polls and projections.
To start with, projections have no legal meaning. Legally, there is no winner until several weeks after the election when canvassing boards certify the official results. Instead, a projection is a media company’s prediction of who will be the winner once all votes are counted.
In practice, all of the media companies isolate their projection team from those reporting the election results. Thus, the number crunchers do not know what the talking heads are saying. On the other hand, the number crunchers aren’t idiots. They know what the expectations are and what races are probably going to determine who wins.
Projections are based on two things — exit polls and actual vote counts tempered by experience. I will discuss the count issue later in this post, but let’s start with exit polls.
For the most part, exit polls are conducted by randomly choosing voters (say every tenth or every fifteenth) voter at “representative” precincts. Unlike other polls which are weighted primarily on demography and expectations of turnout in different groups, the choice of representative precincts is designed to already do that type of weighing. Instead, the weighing is based on how many precincts each representative precinct represents. At least in the initially reported exit polls, that weighing is based on historical turnout. Of course, until results come in, we do not know actual turnout. The final version of exit polls adjust the weighting for actual turnout in similar precincts.
The fact that exit polls as initially reported are based on expected turnout is one of several problems with the accuracy of exit polls (although they are relatively accurate). The second problem is that a significant percent of voters do not vote in person on election day. Recognizing this problem, most polling companies use some equivalent polling of people who vote before election day but that polling has all of the issues with phone and on-line surveys along with the possibility that those answering may have “buyer’s remorse” and report who they wished they would have voted for rather than who they actually voted for. Of course, there are issues with polls conducted in person too. In short, all of the things that lead to margins of error in pre-election polls apply to exit polls.
The third problem is that the exit polls have an effective “cut-off” time. It takes time to crunch the data. Thus, the initial exit polls only include responses through mid-afternoon. In theory, the evening voters should not be that different from the morning and afternoon voters, but they could be.
In doing projections, the media companies are aware of this issue. Thus, they have started to be more conservative about projecting based solely on election polls. For an “instant” projection, the exit polls have to point toward a rather significant margin. If the winner is showing something like 55% in the exit polls, the media company is going to want to wait for some returns to show that the exit polls are getting it right.
Turning to the returns, as discussed last week, not every local election authority is reporting votes at the same time and early voting is not always reported at the same time as election day voting. Past elections give some clue as to the expected total vote from the various local election authorities. And the exit polls give you some idea if turnout is higher or lower than normal in certain areas. And past election returns give you some idea of what you expect the breakdown to be within each local election authority.
To use my own state as an example, in 2020, there were about 10,300 votes in Adair County which President Trump received about 62% of the vote. On the other hand, Boone County had about 91,100 votes of which President Trump received about 33% of the vote. Knowing what to expect then allows you to compare the current returns to past returns and the exit polls.
Turning to how this plays into election night, the media companies have an idea of what each candidate needs to happen in terms of the vote breakdown and turnout to win. If the exit polls are showing that one candidate should win but are not clear enough to permit an instant projection, you wait for the first set of results. Turning back to Missouri, we would expect to see the votes from Adair County relatively early in the evening. Is turnout in Adair County up from 2020? Are we seeing a swing in Adair County that differs from what the exit polls project. If the vote total from Adair County are close to what the exit polls project, that is confirmation that the exit polls are accurate. If the exit polls are outside the margin of error, having a couple of counties release results that confirm the exit polls can lead to a projection. But if we have a real close race in Missouri, I know to expect there to be 500,000 votes from St. Louis County. If, from the first 50,000 votes, the Democrats are getting around 70% of the votes, then I know that for the rest of the evening, we can expect Democrats to gain 150,000 to 200,000 votes as the rest of the precincts in St. Louis County report their votes. (If you know which precincts in St. Louis County have reported, it is possible to be even more precise in the estimates.) This history is what allows people like Steve Kornacki to do the math on the election boards on election night.
If the exit polls however essentially show a tie (1-2% margin), the exit polls get tossed to the side, and you need to wait to get substantially more results. Again experience tells whether to expect the early results to be more favorable to one side than the final returns will be. (The initial returns are almost guaranteed to historically be more favorable for one party, but, as discussed in last week’s post on vote counting, which party that is varies by state.) In close states, you are waiting to have enough returns from enough areas of the state (including at least partial returns from the largest county) to be able to extrapolate from the early returns to the final returns. And it depends upon how close the final returns appear. If one candidate is up by 200,000 votes, and your extrapolation suggests that the trailing candidate will only close the margin by 100,000 votes, the state is ready to project. Likewise, if a candidate is ahead slightly, but the overwhelming majority of the vote is coming from areas that tend to heavily favor that party and candidate, the state is ready to project. But if the margin is only 100,000, and you expect the trailing candidate to close the margin by 90,000 votes, the state is not ready to project and you may need to wait until almost all of the votes are in.
The media have come up with terminology to describe this situation. “Too early to project” tends to mean that the exit polls suggest a likely winner but it is close enough to want actual votes to confirm the election result. It may take an hour or two to get enough actual votes to project a winner, but the projection room thinks they know who will win. “Too close to call” means that the projection room thinks that both candidates have a legitimate chance to win — either the exit polls show a very close race or the early returns call into question the exit polls (in which case the state moves from too early to project to too close to call). Given the slow rate of results in more Democratic areas, a too close to call state might take three or four hours to get a projection.
One thing that impacts projection is the state laws on recounts (more on that after the election if they come into play). While recounts very rarely change the results, media companies tend not to project if it looks like the final margin will trigger a recount.
Which gets us back to where we were at the start of this post, exit polls and projections are not the official result. They merely tell us where the race appears to stand. The media companies have gotten better over the years (especially after Florida in 2000) in not projecting prematurely. As we continue with the election night posts, the key thing to remember is that there are certain states that we expect to project pretty quickly, some that we expect to be too early to project races, and some that we expect to be too close to call states. Knowing what too expect in terms of projection will help (somewhat) to ease the anxiety as we wait into the early morning hours for a projection of the key states.
