Just one month before the presidential election, Gallup announced it was changing its polling method. The changes were good ones; they included an increase in the number of interviews conducted by cell phones, a reduction in the length of the questionnaire for political questions, adjustments to ensure its sample matches U.S. census targets for the adult population, and modifications to the likely voter screen to account for early voting.
These changes directly address the criticisms put forward by Mark Blumenthal in an expose of Gallup’s methodology, which found that Gallup did not weight its sample to match U.S. census targets for the adult population and under-sampled non-white voters, resulting in a GOP-lean throughout most of the summer. Although Gallup does not acknowledge this explicitly, the effect of the change is to increase the non-white share of the adult population, which will likely improve the president's standing. Already, Obama's approval rating has surged to 53 percent among all adults, up from the mid-forties for most of the last two years.
Changing methodology one month before the election isn’t pretty, and certainly it would have been better if Gallup had made these fixes directly after Blumenthal's piece. If Gallup was going to adjust their methodology today, it would have also been nice if they had provided a retroactive adjustment for their prior polling this cycle. But ultimately, it's a good thing Gallup made adjustments to ensure the most accurate result. They have an obligation to provide the most accurate results they can, even if the timing is imperfect.
And there should be no question that this is a change that will produce a more representative sample. Unfortunately, Gallup’s move is getting caught up in an debate about whether the electorate will be as diverse as it was four years ago. But the demographics of the eventual electorate is an entirely different question than the composition of the adult population, and the latter is not subject to guesswork: there is a right answer and it’s found in the U.S. Census.
So what should we make of new Gallup numbers? Although it leaned toward Romney for most of the year, these changes should probably be expected to produce results that match the national average of other live interview pollsters conducting interviews with cell phone voters. As Blumenthal demonstrated, matching the adult population to the U.S. Census can account for most of the gap between Gallup and similar pollsters. And given Gallup’s huge samples, it promises to be an excellent resource through November.