Most poll consumers understand that polls are a "snapshot," a reflection of attitudes and preferences at any given time. So a poll in July tells us which candidate the respondents would support "if the election were held today."
However, some pollsters try to effectively "crop" the snapshot , using complex "likely voter" models to try to capture only those who would actually turn out to vote if the election were held today. This second step is harder to do, frequently introduces more poll-to-poll variation and, as such, is often more controversial.
One confusing aspect of this topic is that no two pollsters choose "likely voters" exactly the same way.
Many pollsters, including most now releasing public state-level surveys, use a very simple set of "screen" questions to identify likely voters. At the risk of oversimplifying, most seem to ask respondents whether they are (a) registered and (b) likely to vote, and screen out those who say no to either question. Since most registered voters claim they are likely to vote, the practical difference between looking at self-identified "registered voters" and those who also self-identify as "likely" is small.
Gallup pollsters, on the other hand, and others who follow their lead, use a more complex approach. They ask a series of questions that typically correlate with actual turnout (self-reports on intent to vote, past history of voting and interest in the campaign), create a summary index based on the answers, and use that index to select the 60 percent or so of adults who are most likely to vote.
The Gallup likely voter model has proved to make results more accurate when applied in late October, but as I noted last week, the validity of the model is more questionable in July and August. The underlying question we should ask right now is whether we can use the same tools that work in October to take "snapshots" of turnout in the summer. The answer is not always obvious.
When USA Today and Gallup released their most recent poll (conducted July 25 to 27), it showed leading by 3 percentage points among registered voters (47 percent to 44 percent) but trailing by 4 points among likely voters (45 percent to 49 percent). USA Today emphasized both sets of numbers in its coverage.
Emory University political science professor and Obama supporter Alan Abramowitz was suspicious. So he e-mailed Gallup and requested information on the demographic composition of the likely voters they sampled, and the pollsters graciously complied. The data point that troubled him most was the share of 18- to 29-year-olds among the likely voters.
As the table below shows, just 10 percent of Gallup's likely voters were 18 to 29 years old, fewer than the self-identified registered voters they interviewed (15 percent) and smaller still than three different estimates of the percentage of younger voters in 2004 (16 to 18 percent).
In fairness to Gallup, some caveats are in order: The numbers from 2004 are all survey-based estimates; none amount to a census of voters. While the University of Michigan's American National Election Studies and the Census Bureau's Current Population Survey are high quality, in-person surveys, they both rely on self-reports of turnout that exaggerate the size of the 2004 electorate. And while the exit polls weight by age in an attempt to correct bias from those who refuse to participate, an analysis by George Mason University professor Michael McDonald found that their estimates were still younger in 10 states than data taken from voter files would indicate.
So we cannot be certain what the age composition was four years ago, much less what it will be three months from now. Nevertheless, the age composition of Gallup's likely voter pool seems reasonable only if you assume that the estimates of turnout by younger voters were uniformly too high, or that the share of younger voters will fall despite having increased significantly in almost all of the primaries, or possibly both.
Another caveat: Gallup's Jeff Jones points out that if they had weighted the likely voters so that 18- to 29-year-olds were 17 percent of the sample, McCain's margin would have narrowed, but only slightly (with McCain ahead 48 percent to 46 percent).
Jones also reminds us of the underlying reason for the difference: "18- to 29-year-olds score much lower than those in other age groups on the current voting intention questions, and that has been the case throughout the year."
On Gallup's most recent survey, for example, 59 percent of voters under 30 say they are highly likely to vote in November, compared with 79 percent of those over 30; only 45 percent of younger voters say they have given "quite a lot of thought" to the election, compared with 74 percent of older voters.
Abramowitz is undeterred. "A 7 point difference in [the vote preference] margin between actual voters and all registered voters," he wrote me, "would be much larger than that found in American National Election Studies surveys in any of the 14 presidential elections between 1952 and 2004 -- the largest gap in pre-election margin was 3 points way back in 1952, and the average gap was only one point."
I think he has a point. Gallup's data tells us that Obama's narrow lead among registered voters depends on those who tend to score lower on measures that typically correlate with turnout. So, not surprisingly, the Obama campaign is investing heavily in efforts to register and turn out new voters.
What is less clear is whether news accounts ought to be emphasizing such snapshots in July when the mechanism for those estimates is so inherently hypothetical and potentially shaky.