Just What is a "Likely Voter"?

The prospect of a May 29 primary, with any necessary runoffs on July 31, has spawned much "conventional wisdom" about who will turn out for this election. Most of what you’ll hear comes from one campaign or the other, usually in a blatant — if understandable — attempt to paint their chances in the best light as a means of attracting donor dollars in what is already an extremely long campaign season. All this posturing raises the question: Given a survey of potential voters, how should we go about determining what the electorate might look like on Election Day?

If you follow campaigns closely, you have certainly heard the term "likely voter" thrown around on television and in articles. Rarely do you hear who, or what, a likely voter actually is in these discussions.

Likely voters are generated by the person conducting the survey. But although they are invented, they are essential if one wants to glean some sort of truth from a public opinion poll. Whereas a good survey will garner responses from a representative sample of the population of interest (for example, registered voters in Texas), even the casual political observer knows that most people don’t vote — not in presidential elections, and certainly not in lower-level elections or primaries.

A likely voter screen, as it’s called, allows for a better approximation of what the electorate might look like on Election Day. This approximation is necessary because we know that most eligible voters are not going to turn out in any given election. It is also the case that if we simply ask people whether they intend to vote, they overwhelmingly say yes. This is the result of what is called a social desirability bias, a tendency for respondents to over-report good behavior and under-report bad behavior.

Respondents, having offered a number of opinions on political questions, don’t want to then turn around and tell the truth — that most of them won’t be voting. Also, a historical conception of citizenship in America is based on the notion of voting, and to say that one doesn’t intend to vote violates a basic premise of being American. For both of these reasons, among others, simply asking people whether they intend to vote, and then making those respondents the "likely voters," just won’t do.

In order to construct the likely voter screen, survey researchers often include questions that might lead to a better guess at what the electorate might look like on Election Day. Three basic factors commonly used are interest in politics, reported tendency to vote in elections, and (to a lesser extent) political knowledge.

Teasing out a distinction between those who are interested in politics and those who are not seems simple, but honing in on this distinction can be complicated. The standard survey question offers four answers: extremely interested, somewhat interested, not very interested and not at all interested. Should we just consider those who are extremely interested in politics to be the likely voters, or should we also consider those somewhat interested? If we only consider those who are extremely interested, we can be certain that our group of likely voters is one that has a high probability of showing up on Election Day, but the purpose of the screen is to provide an accurate assessment of the race.

If vote certainty were our only goal, we could create a screen resulting in a group of highly educated, knowledgeable and interested voters who are wealthier and whiter than the population of interest with a reported tendency to vote in every election. Of course, this sample of likely voters would now consist of a tiny fraction of what we started with, and it would tell us nothing about the race that we’re interested in. This is why we talk about a "likely" voter screen, not just a voter screen.

The standard approach to past voting history is similar to that of political interest, and often strives to address desirability bias by offering some relief to those who might feel awkward about voting less often than ideals about citizenship might suggest. For example, in the last University of Texas/Texas Tribune poll, the item looked like this:

There are many elections in the state of Texas. Furthermore, many people intend to vote in a given election, but sometimes personal and professional circumstances keep them from the polls. Thinking back over the past two or three years, would you say that you voted in all elections, almost all, about half, one or two, or none at all?

1. Every election
2. Almost every election
3. About half
4. One or two
5. None
6. Don’t know/Refused/NA

This item provides flexibility similar to that provided from the interest item — you can filter the result with varying degrees of selectiveness while doing your best to limit the tendency of voters to over-report their voting behavior.

One other criterion we might consider is the respondent’s level of political knowledge. Oftentimes we ask respondents basic and relatively simple political questions at the end of surveys, and a significant portion of the electorate gets these questions wrong. While interest may be a subjective criterion, some level of basic knowledge might provide a more objective account of the potential to vote because, for example, if you don’t know which party is in control in the House of Representatives, you are less likely to know when or where the Republican primary is being held.

The notion here is that knowledge of basic political facts is necessary if one is to hold more complex political knowledge. In a low-turnout or low-interest election, showing up may be a function of feeling or knowing that the election is important, and attaching importance to what might appear to be an unimportant election in many cases requires having basic and complex knowledge about politics. For example, understanding the role that a judicial election might have on the balance of the court, and how that balance might then affect some particular issue of importance to that voter.

The overall decision of what likely voter screen to use then comes down to two major factors: first, what screen does the pollster think will produce the most accurate results under the circumstances; and second, how much statistical uncertainty does the pollster add to the results by creating a tighter screen. Adding additional criteria leaves fewer voters able to satisfy those criteria; with that smaller sample comes a higher margin of error. The end product might be an accurate assessment of who is likely to vote, but it will contain so much uncertainty about actual preferences that we can no longer say anything useful about the state of the race.

Sometimes the choice of likely voter screen is significant and sometimes it is not. As can be seen in the chart below, based on selected data from the last UT/TT poll, if the election result looks to be rather lopsided, as is the case with the GOP presidential primary, then various voter screens might change the outcome of the survey results slightly, but without much consequence.

However, when considering a closer election or one in which turnout might be low, for example, the Republican Senate primary, different likely voter screens can lead to different expectations about the state of the race. In situations in which the electorate is "less predictable," we should expect the likely voter screen to provide some guidance. Again, the current GOP Senate primary is a choice example. With an uncertain election date (at the time of the poll), low visibility with so much of the focus on the presidential nomination contest, and the possibility of a summer runoff — generally considered the dead season of politics — examining the opinions of the full electorate would be very misleading. But, of course, there is still an enormous amount of speculation involved in deciding just how stringent a screen should be, particularly in primary elections.

As a savvy consumer of this type of political information, it is especially important to be wary of election polls that do not use a likely voter screen, because we know that at the very least, the electorate on Election Day is not going to include everyone. It is also important not to ask more of a likely voter screen than it can provide. If the Republican Senate primary in May turns out exactly, or very close to, the results provided by our most recent UT/TT survey, in which likely voters were considered those who are extremely or somewhat interested in politics and those who vote in every or almost every election, it might be tempting to say that we got it "right." However, this would be going too far. The goal of the screen is to provide an accurate assessment of the electorate right now, a snapshot in time. Opinions change, campaigns campaign and circumstances shift. So even the most accurate of voter screens may deviate substantially from the eventual outcome if it is applied to data two, three or four months before the actual election. All polls become more accurate the closer they are to Election Day, but what one has to ask, given the nature of that specific election with respect to interest and turnout, is whether a likely voter screen provides additional accuracy, and how it should be constructed.

Joshua Blank is a research assistant for the University of Texas/Texas Tribune poll. He is a Ph.D. student in the Department of Government at UT-Austin.