11/5: See also this update on a new CNN poll showing independents breaking for Romney by 22 points which would translate into a 7-8 point margin of victory.
There are dozens of presidential opinion polls these days, and even more interpretations what they mean. But predictions keep changing, which calls into question how useful they are. The closer you get to an election the easier it is to call it, at least in theory. But it is not really predicting when you change your take on the race with every new poll. For example Nate Silver at his New York Times blog has consistently “predicted” an Obama victory, yet has altered the odds on a daily basis. This is the equivalent of claiming you can predict the outcome of a baseball game, but then changing the odds inning by inning to reflect the evolving state of play. Yes your “prediction” will be very accurate once you reach the 9th inning, but it also doesn’t take a stat wizard to make the call.
Traditional social science models do more than simply rehash and average out the latest daily surveys. They look at the influence of variables like age, education, income, sex, race, the economy, and other such factors to come up with more durable conclusions. For example University of Colorado Professors Ken Bickers and Michael Berry have developed a model based on state-level economic data that predicts Mitt Romney winning with 330 electoral votes. They have applied this model successfully to every presidential race since 1980. It does not shift around with the polls; in fact it does not use them at all.
The most robust predictor of individual voter behavior is party ID. In these hyper-partisan times, if you know someone’s party affiliation you can say how they will vote with around 95% certainty.
Party ID favors the GOP. 2012 is the first modern presidential election in which the Republicans are going in with a clear edge. Rasmussen’s breakdown is 36.8 Republican, 34.2 Democrat and 29.0 independent. Gallup has the split at 36/35/29. Some have criticized the accuracy of these organizations’ methods – and Gallup has had to deal with the Justice Department, allegedly for coming up with results not to the Obama campaign’s liking. But in the last two elections these pollsters have generated party ID numbers that closely tracked with exit polls of actual voters.
The Republican plurality is not a statistical blip but part of a long term trend. Democrats had about an 8 point advantage in party ID in November 2008, but they quickly lost their lead. By the November 2010 midterm election – the seismic shift in which Republicans took more seats from Democrats than in any midterm election since 1938 – Republicans held a 1.3% party ID edge. Democrats have successfully rewritten the historical narrative to ignore the importance of the 2010 “shellacking.” However the forces that gave Republicans control of Congress then are even stronger now, and since 2010 the GOP advantage has grown to 2.6%.
Seventy percent of the electorate is distributed between the two parties, so the election will be decided by the independents. Throughout the race independents have been leaning towards Romney. So if most voters follow their party, and independents break for Mr. Romney, he will win the popular vote. No doubt about it.
This is easily calculable. Assume the parties hold 95% of their adherents with the rest crossing over, and independents split between Romney and Obama 53/47. In that case Mr. Romney wins by 4 points using the Rasmussen data (52/48) and 2.6 using Gallup (51.3/48.7). If the independents break a full 10%, 55/45, the results are 52.6/47.4 Rasmussen, and 52/48 Gallup. In other words, Romney by 4-5 percent.
This model may understate Mr. Romney’s vote because it does not take into account the enthusiasm gap between the two parties. Just like in 2010, Republicans and anti-Obama independents are much more motivated than Democrats and other Obama supporters. That could be worth another point or two in the popular vote spread.
How these national totals translate into electoral votes is another matter. The same calculations would have to be done state by state, with reliable party ID data, if available. They would also have to factor in the influence of third party candidates who can have an outsized influence in close state races. Case in point Ralph Nader received over 97,000 votes in Florida in 2000, a state which Al Gore lost by 537 votes. Razor-thin margins in swing states in 2012 will likewise amplify the impact of the marginal candidates and could lead to some surprises.
However, facing larger numbers of more enthusiastic Republicans, and with independents turning against him, Mr. Obama can’t win. And on Tuesday we will find out.