Scenario 1: We wake up on Wednesday, November 7th and read about how President-elect Romney seemed to defy the odds to win by a comfortable margin, 52 – 49. He eeked out narrow but clear victories in most of the swing states, and Republicans managed to take slim control of the Senate.
Democrats are shocked and disappointed. All the polls said it was close, Obama’s ground game was supposed to pull it out, and what about Nate Silver’s odds heavily favoring the President? Republicans will feel vindicated that they represent what Americans believe.
Scenario 2: President Obama is re-elected in a narrow but clear victory thanks to how he held on to leads in swing states, buttressed by his ground game. He may or may not win the popular vote, but neither did George W. Bush in 2000. Democrats will also gain in the Senate, something that would have seemed impossible a year before. Democrats will be relieved and feel their vision of the future is winning, while Republicans start soul searching about how they need to transform their party.
How can two very different scenarios each be plausible? Simple: the polling data shows a close election and two different dynamics. Due to inherent uncertainty about which dynamic is actually in play, there is no way to be confident that any conclusion is truly likely.
One reason people discount scenario one is the popularity of New York Times blogger and statistical guru Nate Silver. His models have worked with amazing predictive power in the last two elections cycles. He knows what he’s doing. So when he makes Obama a clear favorite, that has to be taken seriously.
Yet as a social scientist who deals with qualitative and interpretive methods, I warn against reading too much into a quantitative analysis and model. It’s not that such work isn’t good, it’s just that the world is so complex and multi-causal that even good models fail sometime. That’s why Silver’s model gives Romney a 25% chance of winning. In essence a set of assumptions are built into how the polls are treated in Silver’s model (economic factors matter too, though their relative importance dwindles by election day), and if for some reason in this election cycle those assumptions are off, the other guy wins. Silver thinks there’s about a 25% of that happening.
Though Gallup can go off base, it has got a good track record overall. The conventional wisdom (and one suggested by Nate Silver as well) is that Gallup’s methodology is somewhat off. Since the rule of thumb when looking at polls is to distrust the outlier, it has become easy to distrust Gallup.
Yet after releasing a Politico/Battleground poll yesterday showing Obama up 1, pollster Ed Goeas mentioned that their election modeling suggests Romney should win 52-47 — a result eerily similar to Gallup. And, since Gallup stays mum on a lot of how they get their numbers, it could be that they’re integrating some kind of election model in their poll that is similar to that used by the Battleground poll.
That means that a couple big name pollsters with good track records have a model or set of assumptions that yields a clear victory for Romney. Simply, the assumptions built into the methodology of the different models yield different results.
While Silver has developed a model using the universe of polls out there and other data, individual pollsters like Gallup use their own data and then make assumptions about how voters will behave on election day. This leads them to make assumptions about actual turnout by different demographic groups. This could include party identification, intensity, certainty to vote, age and other factors that might not be used for publishing individual poll results.
Gallup has said it expects the 2012 electorate to look much like the 2008 electorate in demographic make up. Yet the trend has been for minority turnout to increase. Gallup apparently believes that lack of voter intensity will keep those voters at 2008 levels.
If these assumptions are right, then Obama’s lead in the swing states is not only soft, but illusory. The dynamics favor voters coming out decisively for Romney. Silver’s model takes the history of polling accuracy into account, the models favoring Romney look at how history guides understanding who is truly likely to vote.
In short, it all comes down to voter turnout. For instance, polls show Latino voter intensity to be very high this year, one poll saying 80% intend to vote.
As this graph shows, however, less than 50% of Hispanics voted in the 2008 election, a number little changed from 2004. If Latino voter turnout actually does increase, assumptions based on the 2008 election may be wrong, and that could swing a number of states and the total vote towards Obama. Moreover, geography matters. The national trend may not change much, but if get out the vote efforts alter them in key swing states, that could make scenario two more likely.
But with swing state leads for Obama very small, the kind of shift that Gallup seems to envision could put states like Wisconsin, Iowa, Colorado, Ohio, Virginia and North Carolina in Romney’s pocket. It doesn’t take much.
One week from the election it comes down to a simple question of who is going to go out and vote. Two very different election results are plausible, respected analysts have models that declare each one to be likely. Who is right will be determined by who votes. That’s democracy!