As a follow-up to John’s recent post regarding the Rothschild and Wolfers research suggesting the predictive value of aggregating voters’ beliefs about the likely winner of the election, Pew just recently released its final set of numbers in this regard. The following table from the report shows the most recent numbers compared with results from earlier in the campaign:
Two points are worth noting about these numbers. First, if the Romney campaign is following a strategy of claiming momentum heading into the final days of the polls, these numbers do not suggest much evidence that voters have bought this story. Instead, what seems to have happened is that skepticism about Romney’s ability to win increased in September (the 47% comment?) before returning to baseline levels among Romney supporters in October. This could have been the result of the Denver debate, but could also simply have been a result of the 47% shock dissipating over time. But either way, there clearly has not been a huge uptick in recent weeks in the proportion of voters thinking Romney is going to win the election, either among Obama or Romney supporters.
Second, despite the same Pew report finding that Romney supporters are “more engaged” in the election, Obama supporters clearly have greater confidence that their candidate will win the election than Romney supporters. It will be interesting after the election to try to figure out why this is the case. The baseline argument for using voters’ expectations about election results is that the wisdom of crowds is powerful, but clearly these data show that this wisdom is tremendously colored by partisanship. So why the greater certainty among Obama supporters?
One other point about the Pew Report is worth noting. Buried in the fine print of the report is the statement that
Likely voter estimates are based on a 9-item turnout scale that includes the following questions: thought, campnii, precinct, oftvote, where, plan1/plan3, folgov, pvote08a and scale10. The turnout estimate used in identifying likely voters is 58%.
While I’m not a fan of using variable names in this manner, I do like the fact that Pew tells us what it thinks turnout should be on the basis of its likely voter model. This seems very transparent, and my interpretation is that if turnout is higher (or lower), we can then be one step closer to figuring out why the Pew numbers did (or did not) do a good job of predicting the final outcome. Would be nice if everyone gave out at least this piece of information about their likely voter models.