Most Accurate Polls in the 2012 US Election


Situation 

The conventional wisdom was always that the 2012 Presidential election would be a close race. On one side, Barack Obama was a personally popular incumbent presiding over a sluggish economy that – despite late improvements in the perception of the economy as a whole – was largely not improving for people on a pocketbook level. On the other side was Mitt Romney, seen as largely out-of-touch with ordinary Americans, but who was trusted with the economy and jobs.

In public polling, President Obama consistently led Mitt Romney for much of the campaign – holding a small but stable lead in an election with hardened party bases and relatively few undecided voters. The President’s lead widened after the two parties' conventions, particularly in the battleground states where the Electoral College would be decided. After a strong showing from the challenger in the first Presidential debate, however, public polls began to show Romney taking the lead in the race. From that point on, the media, pundits and countless publicly respected polls showed the race as dead even or trending slightly in Mitt Romney’s favor. Joe Scarborough pronounced, “…Anybody that thinks that this race is anything but a tossup right now… should be kept away from typewriters, computers, laptops and microphones for the next 10 days.” The Wall Street Journal proclaimed “Obama and Romney deadlocked.” The Economist asserted the race was “about as close as it could be.” Most other media and pundits thought so, too. Except us.

Solution

In contrast to other polls, national surveys conducted by Greenberg Quinlan Rosner for Democracy Corps consistently showed Obama with a lead in the last few weeks before the election.

Our final poll (completed two days before Election Day) showed the race 49 to 45 percent – for an unrounded margin of 3.8 points. With other public polls still showing the race tied or Romney ahead, we knew someone had to be wrong.  But we believed in our turnout model and our industry-leading methodology. As of this writing, the margin is 3.6 points and inching up – a remarkable achievement of accuracy.

There are several reasons why we think the national tracking averages likely underrepresented Obama’s vote. The main issue is cell phones and the changing demographics that most other pollsters miscalculated. Our likely voter sample included 30 percent reached on cell phones from a cell-phone sample conducted in parallel with random-digit landline dialing. Pollsters who did not dial cell phones could not reach the new America. Plain and simple.

Indeed, our accuracy in this election reflected years of intense study and a series of careful decisions about key assumptions in our election modeling, including ones regarding demographic and turnout trends among pivotal voting groups, notably Latinos. It reflected our years of attention to the composition and dynamics of the “Rising American Electorate” – young voters, non-whites, and unmarried women – a set of voters who decided this election. And our accuracy reflected our intense focus on the methodological changes necessary to accurately sample the full American electorate – such as insisting on a higher proportion of cell phone interviews, despite the higher costs.

Because of our attention to detail and in-depth awareness and understanding of the contours of the likely 2012 electorate, our polls produced the most accurate results of any national survey conducted in the last three weeks leading up to Election Day. In his assessment of pollster accuracy and bias in the 2012 election, Nate Silver of The New York Times' blog FiveThirtyEight ranked Democracy Corps and Greenberg Quinlan Rosner first among all national polls with less than a 1 percentage point error – an average error that has decreased since his initial ranking as votes continue to be counted. Simply put, Greenberg Quinlan Rosner and Democracy Corps correctly predicted the margin of the 2012 Presidential election, leading the industry’s methods and standards as pollsters are increasingly forced to rapidly adapt in an era of ever-changing demographics, turnout, and technology usage.

Methodology 

Greenberg Quinlan Rosner and Democracy Corps conducted national surveys monthly (bi monthly in the final 2 months of the election) providing in-depth public opinion research and tracking on key political and economic trends and issues. Democracy Corps national surveys include two samples -- one generated via Random Digit Dialing (RDD), and another separate sample for cell phones.  We used an RDD sample frame of cell phones and apportioned the cell frame via census regions according to the latest CDC National Health Survey Interview (NHIS) Wireless substitution to increase cell-phone only (CPO) interview likelihood and representation. We used a dual-frame, full overlap methodology allowing any cell user type to screen into the survey regardless of which frame the respondent was contacted in. Thirty percent of respondents in each Democracy Corps survey are reached on cell phones.