There has certainly been a great deal of discussion about election polling after last November’s presidential election, more than I recall seeing over the last five presidential elections. Some of that focused on our Gallup Daily tracking program. From our perspective, the goal is to be as accurate as possible in estimating election outcomes. President Barack Obama won the election over Mitt Romney by nearly four percentage points in the popular vote, 51% to 47%, while Gallup’s final estimate showed a statistical tie: 49% for Obama, 50% for Romney. Our final estimate was quite similar to a number of others, and the average estimate for polls using methodology similar to Gallup’s was about +1 for Obama -- suggesting some fundamental issues with this election that affected many polls. Still, our goal is to have our estimates come as close as possible to the final result.
To that end, we launched a comprehensive review of our election polling methods shortly after the election. Dr. Michael Traugott of the University of Michigan joined that effort in December, and is playing an important role in our examination of a number of aspects of the election process. Dr. Traugott not only is an expert in the field of survey practice and methodology, but he led AAPOR’s review of polling in the New Hampshire Democratic primary in 2008 in which virtually all polls inaccurately predicted Obama would win.
Among the elements that have been our methodological focus over the years that we are reviewing in the context of last November are:
- cellphone sampling and phone status weighting (Gallup used 50% cellphone interviews during the final month before the election)
- the registered voter screening process
- the likely voter screening process
- measurement of early voting
- the impact of campaign contact and get-out-the-vote efforts on the final voting electorate
- gender and cellphone quotas within region
- the impact of within-household selection procedures
- Gallup interviewing house effects
- basic representation of demographic categories in our weighted and unweighted national adult samples, including racial and ethnic groups
- methodological issues involved in daily tracking as opposed to “stand alone” polling
Also, it’s worth noting that our generic, 2012 congressional vote estimate -- included in the same survey as our presidential polling -- was in fact highly accurate. Gallup’s final estimate of the aggregated congressional vote showed a one-point Democratic advantage among likely voters, 49% to 48%, with 3% undecided. The latest update on the national House vote, as calculated by David Wasserman of the Cook Political Report, also shows a one-point Democratic advantage: 50.5% Democratic, 49.5% Republican. This difference in accuracy between the presidential estimate and the generic House vote estimate is something we are examining.
We invest in election tracking and report it publicly because we think it provides significant value in understanding the dynamics and nature of the presidential race as it unfolds. We hope our review process adds to that understanding. Of course, we are never far from the next national election, and polling will no doubt continue to evolve by the time those elections -- particularly the 2016 presidential election -- take place.