The nerd wars among different tribes of election prognosticators have been surprisingly vicious, given that they’re all basing their forecasts on the same thing: polls asking people whom they intend to vote for. But what if this is the wrong question?
I think it is.
In a recent academic research paper, David Rothschild and I examined the results of a different kind of poll, one that asks instead: “Who do you think will win?” We find convincing evidence that this alternative approach yields more accurate forecasts. In the 345 races we studied, determining a state's Electoral College vote, the consensus of the “who do you think will win?” poll picked the winner 81 percent of the time, compared with a hit rate of 69 percent for the standard polls of voter intentions. Roughly two-thirds of the time, we find that forecasts based on voter expectations are more accurate than those based on the usual polls of intentions.
So, what does this alternative -- we think, better and more accurate -- approach say about today’s race?
The latest Ipsos/Reuters Poll says that 51 percent of respondents expect President Barack Obama to win, versus 32 percent who expect a Mitt Romney victory (the remaining folks don’t have an opinion). The latest CNN/ORC poll scores it 57-36 for Obama; likewise it’s 55-35 in the ABC News/Washington Post poll; Pew scores it 52-30; and Politico/GWU/Battleground sees it 53-38.
There’s a clear consensus: A robust majority of the American electorate expects Obama to win re-election tonight. So that’s my tip, too.
Among the five polls I cited, an average of 54 percent of respondents expect Obama to win, while only 34 percent anticipate a Romney victory, and the rest don't have an opinion. Narrowing to those who have a view about who will win, 61 percent expect Obama to win, versus 39 percent for Romney.
What does this say about Obama’s likely winning margin? Here we need to be careful: Just because Obama has a 20-point lead in a voter-expectation poll doesn’t mean he’ll win the election by 20 points. Instead, we have to use a nifty formula we’ve developed for translating Obama’s lead in these polls of voter expectations into an expected winning margin.
Bottom line: Surveys of voter expectations suggest that Obama will win by about five points tonight.
This forecast is more aggressively pro-Obama than the conventional wisdom. Indeed, each of the five polls suggests a much closer race if one focuses on voting intentions. If you look at pre-Hurricane Sandy data from a sixth pollster, Gallup, they suggested that Romney had a five-point lead, while Gallup's contemporaneous voter-expectation polls implied that Obama was the one holding the five-point lead. Tonight we’ll see which question Gallup should have been relying on.
There's another advantage of our approach: Although the consensus of standard voter intentions polls has started to tip toward Obama over the past few days, polls asking voters who they think will win have consistently pointed to a robust Obama victory for the past few months. We’re not suddenly hopping aboard the “Obama will win” meme. It has been evident in our data for a while.
(Justin Wolfers is a Bloomberg View columnist. Follow him on Twitter.)
Read more breaking commentary from Bloomberg View at the Ticker.