Part and parcel of predicting the probable outcome of a presidential (or any other) election is determining who is likely to show up to vote on Election Day. Only a subset of the eligible citizens who reside in a geographic area will actually cast a ballot, and the identity of these individuals is not known in advance. Citizens may be willing to express a preference for a political candidate to a pollster, but ultimately, the utility of that preference is a function of the probability that the citizen will vote.
Pollsters have used a variety of strategies over the years for determining whether an individual is a “likely” voter. In some instances, being registered to vote is presumed to make a person a likely voter. However, a dozen states now permit citizens to register to vote on Election Day itself, and Oregon is now automatically registering all eligible citizens at age eighteen. Thus, not all citizens need to be aware of their registration status in advance of the election in order to cast a ballot. Moreover, many registered voters fail to show up on Election Day; registration is, at best, an imperfect measure, particularly if it is not asked proximate to Election Day.
Often, questions about whether or not a citizen is registered are combined with questions about a person’s past voting history and whether they know where to vote. Like voter registration, this information is less likely to be helpful if it is being used at a time not proximate to Election Day, particularly for young or first-time voters, who are less likely to have such information at hand before they need to use it.
A citizen’s self-reported level of interest in (or enthusiasm for) the campaign is often used as an indication of his or her likelihood of voting, and this may be used in conjunction with measures of vote intent, vote history, voter registration, or knowledge about voting locations. In this scenario, only a citizen who expresses a very high level of interest in (or enthusiasm for) the campaign is treated as likely to vote. Measures of campaign interest and enthusiasm are quite volatile and may respond to the ebb and flow of campaign events, but they don’t exhibit much ability to predict future events. Moreover, some cohorts (e.g., young people) exhibit lower levels of interest in campaigns, generally, but may still turn out to vote. In the Campaign of 2012, Republican nominee Mitt Romney’s campaign team was convinced that he had won the election, based largely on its internal polls of likely voters in battleground states. Romney’s team constructed likely voter estimates based on levels of voter enthusiasm and found that when looking only at voters who reported the highest levels of enthusiasm, Romney seemed to be performing better than Obama in several of the battleground states. In fact, Romney lost, and the campaign’s internal polls were wildly off the mark in some cases. Aside from other sampling problems, it seems likely that Obama’s supporters showed up even if they did not exhibit extremely high levels of enthusiasm for the election (and thus perhaps had different motives for voting), a shortcoming of the Romney model.
See also Poll of Polls
Burns, Alexander. “The GOP Polling Debacle.” Politico, November 11, 2012.
Erikson, Robert S., Costas Panagopoulos, and Christopher Wlezien. “Likely (and Unlikely) Voters and the Assessment of Campaign Dynamics.” Public Opinion Quarterly 68, no. 4 (Winter 2004): 588–601.