Last week, Gallup reported the anomalous finding that the race was tied. They received quite a bit of attention for that outlying result, including from this blog. But not everyone saw Gallup’s tied race as especially unusual. After all, the recent AP/GfK poll also showed a one-point race, and the RCP average showed Obama leading by 3 or 4 points. Why focus on Gallup’s 3 or 4-point divergence from the average, I was asked on Twitter, when Pew Research diverges to a similar extent in the opposite direction?
This sentiment is understandable, but it’s wrong. Gallup is a poll of registered voters, not likely voters, and that means that Gallup was diverging from similar polls by about 7 points, not just three or four points. The AP-GfK poll that was often cited as an analogue was not, in fact, the slightest bit analogous. In that poll, Obama led by 10 points among registered voters, and the close race among likely voters was only a product of an unusually tight screen. At this point, there’s no question that Obama enjoys a broad lead among registered voters. What's at stake is only whether Obama can get enough of them to the polls on Election Day.
Last week, Gallup stood alone in showing a tied race among registered voters. But over the weekend, Obama gained a couple points in the Gallup tracker; today Gallup is just 5 points less favorable toward Obama than an average of registered voters surveys. This makes Gallup look more consistent with the RCP average, but it's still diverging to a substantial degree from the other registered voter polls. Indeed, a 5 point difference is even larger than Rasmussen's 3 or 4 point Republican-lean, and larger than Gallup's longer term Republican-lean, which averaged about 3 points over the summer.
The 5-7 point gap between Gallup and similar polls of registered voters is also a larger gap than the initial Republican-lean that prompted Mark Blumenthal to take a dive into Gallup's polling. In an expose earlier this year, Blumenthal found that Gallup under-represented non-white voters who typically support Democratic candidates. Impressively, Blumenthal demonstrated that the Gallup poll could be re-weighted to produce similar results to Pew Research surveys, which at the time showed Obama's approval rating a net-3 points stronger than Gallup.
Could the same factors explain Gallup's anomalous results today? Last week, Pew Research and Gallup conducted polls over the same period and differed by nine points among registered voters. Whether the Gallup-Pew difference can still be explained probably depends, at least to some extent, on whether Gallup's post-convention "norm" is a tied race or something further in Obama's direction. If Gallup ultimately settles on a 4 point Obama lead among registered voters, that would sit comfortably within range of Gallup's longer term average, which has generally fallen about 3 points more Republican than the average of registered voter polls. But if Gallup's post-DNC average shows a tied race while other polls continue to show Obama leading by 7 points among registered voters, it would be interesting to learn whether the methodological peculiarities identified by Blumenthal could still cover the gap. They very well could, but it's worth observing that much of the difference between Gallup and Pew seems to be among white voters, not just the non-white share of the electorate.
Ultimately, none of this is a reason to completely ignore the Gallup poll; it will be included in any average published on this blog between now and Election Day. But those who intend to stake their view of the race on its results should recognize that the Gallup numbers are far out on a limb. At the moment, Gallup isn't just leaning slightly toward Romney; it's an outlier, at least over the past 10 days. Outliers can ultimately be right, but, in this instance, there isn't a credible reason to assume that it's more accurate than the other pollsters.