Gallup's flashy Romney+7 result is deservedly getting quite a bit of attention. It's a much stronger result for Romney than any of the other national polls and it also seems inconsistent with the state polls.
After their recent adjustments to interview more cell phone voters and adjust its adult sample to match U.S. Census demographic targets, the demographics of Gallup's registered voters should be about the same as other pollsters. Gallup now has Romney ahead by one point among registered voters, so at least part of Romney's lead is due to sampling. My suspicion is that this won't last for very long (unless the other pollsters move into agreement with Gallup), but we'll see—predicting the movement of polls is a pretty perilous exercise.
But Gallup's difference might also rest in its likely voter screen, not just sampling of registered voters. A few days ago, Gallup released a demographic breakdown of its likely voter poll and simple algebra demonstrates that white voters represented 80 percent of Gallup’s likely voter universe, up from 77.5 percent in 2008. It's not hard to imagine how the electorate could be less diverse than it was four years ago. Obama’s candidacy generated historic—and potentially difficult to repeat—turnout from African Americans and the young voters who allow Democrats to capitalize on demographic changes. Outside of the battleground states, the decline in enthusiasm could be especially large.
On the other hand, the idea that the electorate would be less white than the 2010 midterm elections seems harder to imagine, but Gallup’s likely voter universe is actually even whiter than their likely voter surveys prior to the 2010 midterm elections, which was 79 percent white. This observation is likely to produce one of two responses, with Democrats all but assured to assert that the poll is demonstrably wrong and Republicans taking it as confirmation that Democratic enthusiasm is down, particularly among the non-white voters who brought Obama to victory four years ago.
Both are possible. My presumption is to defer to the balance of pollsters, so I wouldn't discount the polls if they collectively pointed toward a whiter electorate, much in the same way that it's a mistake to ignore the Democrats' persistent edge in party-ID. It's also important to make comparisons within polls, rather than with the exit polls. As a general rule, surveys have shown whites as a larger share of the electorate than the exit polls, which differ over simple but potentially important wording issues. For instance, the exit polls ask voters if they are "Latino," other telephone polls ask if people are "Hispanic," and other still don't provide the option of "Hispanic" or "Latino," but then ask a follow up about whether they consider themselves Hispanic or Latino after categorizing them within more standard racial categories. The multi-cycle consistency of demographic house effects provides added cause to focus on comparisons within polling firms.
Unfortunately, there aren't many post-debate national polls with racial/ethnic breakdowns for 2012, 2010, and 2008. But those that do seem to suggest an electorate more reminiscent of 2008 than 2010. The most recent Battleground/Lake/Tarrance poll showed whites as 79 percent of the electorate in 2008, 78 percent in 2010, and 77 percent in 2012. The NBC/WSJ poll showed whites as 76 percent of the electorate in 2008, 78 percent in 2010, but 74 percent in their most recent survey. And although Monmouth/SurveyUSA doesn't have a corresponding poll for 2008, they show whites as just 72 percent of the electorate. PPP's recent tracking poll puts whites at 71 percent of the electorate. On the other hand, the ABC/Washington Post poll seems to be whiter than it was in 2008, although an exact number can't be calculated because they now provide a single result for non-white voters, even though they disaggregated African Americans and Hispanics in 2008.
This probably doesn't represent enough data to assert that there is a "consensus" that the electorate will be as diverse as it was four years ago, so it's not clear that Gallup is off on a limb. But if more data depicts such a consensus with Gallup's standing alone, then this might go some way towards explaining how Obama holds such a large lead in their poll of likely voters. If Gallup showed whites as 77 percent of the electorate instead of 80 percent, Obama would gain a net-3.7 points among likely voters (based on their internals from earlier this week).
It's important to remember that Gallup appears to weight its sample of adults to census targets, which suggests that major differences in changes in the racial composition of the electorate should be the result of their likely voter screen, not the composition of registered voters (although this is not assured). Likely voter screens are much more of an art than a science and they're calibrated to predict the behavior of the electorate on Election Day. That makes them potentially less useful a few weeks out, when many voters might not yet be paying full attention, haven't researched where they'll vote, and before the effects of GOTV efforts. Gallup has a long history of showing big swings in its likely voter model and Gallup showed a large gap between likely and registered voters that eventually closed by Election Day in 2008. Something similar could occur again.
But maybe not. Perhaps non-white turnout will disappoint Obama supporters and prove the Gallup poll right.