Turnout, turnout, turnout. It is a phrase that is being uttered continuously with concern to the Massachusetts Senatorial Special Election between Republican Scott Brown and Democrat Martha Coakley. The common wisdom is that turnout cannot be predicted. Mark Blumenthal of Pollster.com (my boss in the spring and summer of 2010) has said "I have yet to see any poll or statistical model that can predict voter turnout with precision, especially in an oddly timed special election like the one in Massachusetts." But is that really the case?
While I would agree that pinning down exactly which voters (Democrat vs. Republican, Liberal vs. Conservative, Young vs. Old) will turnout is no sure science, I would argue that we should actually have a very good idea. First, I would point out that we had in 2007 a special election in Massachusetts' 5th district (nearly identical ideologically and demographically to the state as a whole). In that election as in the upcoming Senate election, the Democrats had a female candidate who won a contested primary against a Republican who faced minimal competition in his primary. The election, like the Senate election, also took place at an odd time: October 16th. Although no exit poll data exists for that special election, we have the benefit of two SurveyUSA polls taken a week and a month before the election. These polls correctly estimated the Democrat's share of the vote and slightly underestimated (by 3 and 4 points) the Republican's share of the vote. Thus, the polls were, in my opinion, quite accurate. As it turns out, the partisan breakdown of these polls are nearly identical to the partisan turnout seen during the 2006 midterm elections in Massachusetts.
The question we should be asking ourselves now is "does a 2010 electorate look like a 2006 electorate?" It is tempting to say no. After all, the nation saw an increase in Democratic voters in 2008 (including Massachusetts); however, a look at the 2009 off-year elections suggests that 2008 was an aberration.
Exit polls in the New Jersey (see graph) and Virginia gubernatorial elections show that the partisan makeup of the 2009 electorate was very close to the partisan makeup in 2006 in those states (actually slightly more Republican). Nationally, we see something very similar when comparing November 2006 and 2009 partisan breakdowns among likely voters. Considering Republican and Democratic party identification nationwide among likely voters is at the same levels as they were two months ago, it would make sense that if, in some mythic land, general elections took place in January, this general electorate in Massachusetts should be similar to 2006. But remember, the 2010 election is not a general election. It is a special election.
So where does all of this information leave us? We know that the partisan turnout for a special election (2007) with similar characteristics to the 2010 election in Massachusetts was very much like the electorate for the general election preceding it (2006). Thus, even a special election in 2010 should look very similar to the general election preceding it (2009), which in turn looks like the 2006 electorate. Ergo, the electorate for the 2010 special election should have a very similar to the electorate as the 2006 general election (and perhaps slightly more Republican).
Actually applying this information to the polls for the Massachusetts Special Senate election is a little difficult due to differences in question wording for party identification. The most recent polls (Public Policy Polling and Rasmussen) ask voters "if they are Republican, Democrat, or Independent", while exit polls (and SurveyUSA similarly) ask "do you usually think of yourself as a Republican, Democrat, or Independent". Luckily, this difference in question does not seem to impact the partisan breakdown (see difference between the final national tracking poll of Rasmussen and the national exit poll in 2008). By this logic, the Public Policy Polling poll, whose partisan makeup looks very close to 2006, is closest to what the partisan makeup of the electorate would look like for this special election (although it potentially overestimates the amount of Democrats and underestimates the amount of Independents and Republicans), while Rasmussen clearly overestimates the amount of Republicans and underestimates the amount of Independents.
Therefore, I believe the Public Policy Polling poll, which has Republican Scott Brown in the lead, is closest to the truth. The Rasmussen poll, which shows Brown down 2, is probably shortchanging his percentage of the vote.
Does that mean Scott Brown will win? No, because the election is still very close and within the margin of error. But I dare say, he should actually be considered a favorite.