Probable Electorate

The probable electorate is defined as those citizens who are registered to vote and who very likely will [Page 624]vote in an upcoming election. In election polling and surveying, this concept is operationalized as a sample of pre-election survey respondents whose candidate preferences have been weighted by the respondents' estimated likelihood of voting.

One ongoing challenge in election polling lies in ascertaining which respondents will actually turn out to vote. The distribution of candidate support among the sample in the aggregate, while revealing, is essentially irrelevant when it comes to making predictions about election outcomes. What matters are the choices of those respondents who will actually vote on (or in some cases before) Election Day. If survey researchers had a crystal ball, they could determine with precision which respondents would actually be voters (but then, of course, they could dispense with the preelection survey altogether and simply foretell the outcome of the election). In the real world, techniques are necessary to carve out a probable electorate upon which to base pre-election predictions.

Estimating a probable electorate requires either the creation of a binary turnout screener, by which survey respondents are assigned voting probabilities of either 0 or 1, or the use of a weighting method in which respondents are assigned estimated probabilities on a continuum that can range anywhere between 0 and 1. However, few pollsters provide detailed information on how, exactly, they construct their likely voter estimates, treating their procedures like a secret (proprietary) formula. What is known about how probable electorates are constructed in practice is based on a relatively slim store of publicly available information.

Turnout screening (i.e. a series of survey questions that helps the researcher estimate the likelihood a given respondent will vote) aims to differentiate voters and nonvoters, taking into account the candidate preferences of the former, but not the latter, when making election predictions. Turnout screens can take multiple forms, some simple, some exceedingly complex. On the simple side, a screen might be based on a self-reported turnout intention, asking respondents how likely they are to vote in an upcoming election, then counting only the preferences of those who say they will “definitely” or “probably” vote.

Of course, not everyone who says they plan to vote will actually do so (although most of those who say they will not, do not). More sophisticated screeners take into consideration registration status, past voting behavior, campaign interest, knowledge about the location of polling places, or some combination. Some turnout screeners are based on as many as a dozen questions or more, although research suggests that a screener that includes 15 questions is not much more accurate than screeners based on eight or fewer items.

Using multi-item indices to construct a turnout screening method produces estimated levels of voting likelihood (a “turnout score”) that require the application of some threshold or cut-point: for example, should respondents scoring a 6 or above on a 10-point likelihood scale be counted as probable voters, or only those scoring 7 or higher? The answer depends in part on a priori estimates of what turnout will be on Election Day. Higher anticipated turnout entails including more respondents in one's probable electorate.

...