President Obama's campaign is renewing criticism of the Gallup Organization in the wake of a USA Today/Gallup poll showing the president trailing in a survey of likely voters in 12 swing states. This marks the second time this year that the campaign has leveled complaints against the ubiquitous polling firm after its polls showed results for Obama less favorable than other survey houses.
The latest poll was actually a selection of 1,023 interviews from Oct. 5-11 with voters for the Gallup Daily tracking poll, for which Gallup has already released national results: Mitt Romney led among likely voters nationally, 49 percent to 47 percent. In the 12 states (Colorado, Florida, Iowa, Michigan, Nevada, New Hampshire, New Mexico, North Carolina, Ohio, Pennsylvania, Virginia, and Wisconsin), Gallup interviewed as few as 14 registered voters (in Nevada) and as many as 215 (in Florida). The sample sizes were then weighted to account for the relative populations of the 12 battleground states.
Among all registered voters, Obama led Romney, 49 percent to 47 percent. That is relatively unchanged from mid-September, when Obama led, 48 percent to 46 percent.
But among a subset of 869 likely voters, Romney vaulted into the lead, 51 percent to 46 percent, according to USA Today; Gallup had not yet posted the data on its website as of Monday evening. USA Today also reported that Romney's 5-point lead includes a 12-point advantage among men, and a tied race among women.
Within hours of the story's publication on USA Today's website, the Obama campaign blasted out a memorandum from Joel Benenson, its lead pollster, which said the latest result "underscores deep flaws" in the way Gallup pares down its overall sample of registered voters to project which of these voters will actually cast ballots.
Benenson wrote that Gallup's likely-voter model -- a battery of seven questions about past voting behavior, current vote intention, and attention paid to the campaign -- projected an electorate that overstated Republicans' gains in the 2010 midterm elections.
On Oct. 31 of that year, Gallup projected that among likely voters in a low-turnout model, Republicans led by 15 points, 55 percent to 40 percent, while in a high-turnout scenario, the GOP's advantage was 10 points, 52 percent to 42 percent. In nationwide House elections that year, Republicans won 51 percent of the vote, to 45 percent for Democrats, Benenson wrote.
Benenson then points to 14 individual battleground-state polls conducted since the first presidential debate, writing that Obama leads among female voters in each poll, with an average margin of more than 10 percentage points.
"Gallup's data is once again far out of line with other public pollsters," Benenson wrote.
Obama's pollster then took direct aim at Gallup's methodology.
"We believe the problem with Gallup's outlying data is rooted in their 7-question likely voter screen, which distorts the composition of likely voters, leading to erratic and inaccurate results," wrote Benenson.
Specifically, Benenson pointed to questions Gallup asks about whether voters have voted in their precincts before, and if they know where people in their neighborhood go to vote, writing that these questions "create a bias against groups inclined to support Obama." If these voters score low on those questions, they may be excluded from Gallup's subsample of likely voters.
"This creates a bias against registered voters who are more likely to move from time to time, such as young voters, renters, minorities and urban dwellers, all of whom tend to lean toward the president."
For its part, Gallup denied in a statement that its likely-voter model, rolled out last week, is improperly biased against either side.
"There is no evidence ... from the last two presidential elections that the likely voter model was disproportionately Republican in its effect compared to the outcome," said the statement, conveyed over e-mail by a senior Gallup official.
The statement continues: "Gallup's likely voter model predicted a (slightly) more Democratic outcome than the actual result in the last two (2008 and 2004) presidential elections (i.e., predicted Obama winning by a slightly larger margin than he won, and predicted Bush winning by a slightly lower margin than he won). In both presidential elections the [likely-voter] model was more predictive than the [registered-voter] model."
In April, Obama senior adviser David Axelrod tweeted a link to a piece written by National Journal's Ronald Brownstein. That article suggested that Gallup's registered-voter pool included fewer nonwhite respondents than the overall population of registered voters and the probable November electorate.
Two months later, in an exhaustive story, the Huffington Post's Mark Blumenthal wrote that certain methodological quirks were leading to Gallup surveying fewer nonwhites than other pollsters, which was leading to poorer ratings for Obama on its presidential approval measure.
Last week, however, Gallup made some changes to its methodology. It began making fully half of its phone calls via cell phone and made slight alterations to demographic targets for weighting, which had the initial result of surveying more nonwhite respondents for the president's approval rating.
But little is known about the composition of Gallup's likely-voter sample. Gallup editor-in-chief Frank Newport restated that the firm only weights for the overall adult population (from which the sample is constructed for presidential approval), but it does not adjust for the voters who make it through their screen for the presidential ballot test.
Gallup does not provide detailed information about the demographic nature of its sample, so it is difficult to determine the veracity of Benenson's charge that young voters and minorities are being underrepresented, or to evaluate how closely Gallup's likely-voter sample matches projections of the electorate, past surveys, or exit polls from previous elections. Newport told National Journal in an e-mail sent prior to the Obama campaign's criticism that Gallup plans on releasing some of this information on Tuesday afternoon.