Finding the Elusive Voter
September 22, 2004
Pollsters have a choice in election surveys. Do they report the voting intentions of all people age 18 or older, registered voters, or likely voters? The decision can produce sharply different results. The first group includes too many non-voters, so it makes little sense to base results on their political preferences - unless it is extremely early in the election cycle - because many of them are either not eligible or do not register and many who register don’t bother to vote. Starting months before an election it makes sense to report registered voters, even though that group may include many who will not vote, because the decision as to whether or not to vote may depend on whom the candidates are and the nature of their campaigns.
As the election nears, studying poll results based on likely voters should become more useful, as these are the people who should have the highest probability of actually casting a ballot. Of course, if a pollster is going to make a projection of the actual vote on Election Day based on a survey, a pre-election poll of likely voters should have the greatest likelihood of being accurate.
If the pollster’s intention is to report a snapshot in time of where voters stand today then reporting registered voters is a constant and useful base for the reporting.
A pollster should make reports based on registered voters available. Most polls reports based on registered voters will provide a comparable base for results across different surveys. Many pollsters believe that in order to be acceptable for reporting a survey should be based on probability methods. Other pollsters strongly support non-probability and on-line methods. Which ever method is used should be clearly indicated when reporting a poll.
Identifying likely voters is a difficult task, in part because people who participate in surveys are more likely than others to participate in politics and because the social desirability of voting leads some people to give a "good citizen" response when asked about their voting intention. Also, there is variability among organizations in identifying likely voters, Pollsters understand these pitfalls, but nevertheless they search for the elusive voter.
So how do pollsters decide who qualifies as a likely voter? They use a variety of screening questions to try to make a reasonable guess. The exact methods differ from company to company. Some pollsters treat their likely voter methods as proprietary. We urge pollsters to disclose all their methods, if asked, including their likely voter methods.
The simplest approach is to ask a single question such as: "Are you very likely to vote, somewhat likely, or not likely?" or "Do you plan to vote?" However, this is not a sufficient approach. Using a single measure to determine likely voters results in too many people being classified as likely voters. Beware of polls using this simplistic method.
So what should be asked? There is no definitive answer to this question. The more elaborate methods screen on a variety of questions, including likelihood of voting, past voting behavior, strength of candidate choice, level of interest in- or attention to- the election, and knowledge of location of polling place. Responses to whatever questions are asked are usually combined into a likely-voter index to provide an approximation of those who will actually vote. Some pollsters assign a probability of voting to each respondent. This, too, is a useful way to create a likely voter model.
An important determination for likely voter screening is to project the level of voter turnout. This level may be used in combination with other screening questions to determine who is and who is not a likely voter. We believe the turnout level used in screening for likely voters should be disclosed by pollsters because it can be useful in evaluating an important assumption used in their likely voter model.
Knowing the likely voter screening questions that were asked and how the index was created would be useful. It may mean more to some people than others as its evaluation is somewhat technical. Even so, complete disclosure about how the poll was conducted is a goal we commend. Such information should be available, at least on request. Providing information about how a poll was conducted increases the credibility of election polling.
For a further discussion on selecting likely voters see "Screening Likely Voters: An Experiment" by the Pew Research Center for the People and the Press
For more information about this and other polling issues, contact the NCPP Polling Review Board Members.