The July 25-27 USA Today/Gallup poll shows Barack Obama leading John McCain by a 47% to 44% margin among all registered voters but McCain leading Obama among likely voters by a 49% to 45% margin. This difference between registered and likely voters indicates that now McCain voters are disproportionately represented among the estimate of those most likely to vote if the election were held today. This difference (in which Republicans gain among likely voters compared to registered voters) appears for the first time in USA Today/Gallup polls this year. In earlier 2008 polls, more Democrats than Republicans were engaged in the campaign and considered likely voters. This is generally a rare occurrence given that Republicans have historically been more likely to qualify as likely voters under Gallup's model (a fact that has been borne out in the real world as Republicans are able to win elections despite facing deficits in party identification or pre-election standing among all national adults).
The similarity between the likely voter and registered voter numbers in previous polls this year may be because there has been atypical interest in the election and enthusiasm among Democrats -- likely due to the exciting nomination campaign between Obama and Hillary Clinton.
The current shift in likely voters could be a result of a short-term energizing of the GOP base as a reaction to the Obama foreign trip or some other cause. (Data from the current poll do suggest that Republicans are overwhelmingly likely to feel that the news media are too positive about Obama and too negative about McCain.) The degree to which this current shift toward the GOP candidate among likely voters remains in place remains to be seen. In general, most poll consumers agree that the likely voter model is most predictive in the final poll before an election; analysis based on likely voter models this early in the campaign should be considered to be a snapshot in time and suggestive of possible turnout scenarios and their implications.
Understanding Gallup's Likely Voter and Registered Voter Models
Who are likely voters, and how do they differ from all registered voters?
Several groups of voters can be isolated in pre-election surveys: First is the group of all national adults, which is the basic sample designed to represent all adults 18 years of age and older. Second is the group of voters who say they are registered and able to vote in their local area, which is the group most commonly used for ballot and other questions directly related to the election. And third is the group of voters considered most likely to vote -- "likely voters," which is the group designed to be most predictive of those voters who actually turn out and vote on Election Day.
The relative importance of emphasizing one of these groups over the other can (and has been) debated. Clearly, in the final poll before an election, it's important to execute an estimate of the vote intentions of those most likely to vote. Many previous elections have shown that differential turnout can be a factor in a candidate's final vote tally beyond that candidate's relative position among all registered voters.
The value of looking at likely voters this far out ahead of the election is generally in the category of interesting, advisory information. The vote intentions of the pool of likely voters months before an election shows the potential that voter turnout can have on the popular vote outcome, but is not a predictor of what that turnout will look like on Election Day.
So, as a rule of thumb, registered voters are the relevant group to trend to establish basic patterns of change in voter support for presidential candidates. Likely voter modeling at this point is an additional analytic tool.
There is no agreed upon way of determining likely voters, and generally every polling organization that reports on this group uses distinct procedures. Over the years, there has been debate -- sometimes vigorous -- on the relative merits of one procedure compared to another.
Gallup has a general set of ways of determining likely voters, which have proved to be quite accurate in past congressional and presidential elections. Voters are asked a series of questions about their interest in the election, their plans to vote, and their previous history in voting. The total sample is then divided into two groups: One group is made up of the 60% of voters who score the highest on these measures and are therefore considered most likely to vote, and the other group is composed of voters who are less likely to vote. (The 60% estimate can vary depending on information about likely turnout.)
At this point, the USA Today/Gallup poll is using a set of three questions to determine likely voters:
1. How much thought have you given to the upcoming election for president -- quite a lot, or only a little?
1 Quite a lot
2 SOME (vol.)
3 Only a little
4 NONE (vol.)
5 DON'T KNOW
2. How often would you say you vote -- always, nearly always, part of the time, or seldom?
2 Nearly always
3 Part of the time
6 DON'T KNOW
3. Do you, yourself, plan to vote in the presidential election this November, or not?
3 DON'T KNOW
Closer to Election Day, Gallup will expand this list of questions to the traditional seven-question likely voter model.
The analysis of the results of a registered and likely voter model shows how voter preferences might change given certain assumptions. The registered voter number, which would be equivalent to all eligible voters turning out, generally produces more stable estimates. This is the number that Gallup's Daily tracking poll has reported to date.
The likely voter model gives a sense of what voter preferences might look like if those most likely to vote -- based on their current interest in the election along with their past voting behavior -- voted. The likely voter group measures hypothetical turnout "if the election were held today" with the understanding that conditions will change between now and the election, and knowing that well-designed and executed opinion polls can only measure the current state of affairs and not project many months into the future.