In terms of polling analysis, I feel as though I have morphed into one of the bi-partisan pundit concern trolls that dominate the big media commentariat whenever Democrats have gained the upper hand in our political discourse (calls for bipartisanship were far less common, of course, during the Republican trifecta). Pretty much no matter what the polling situation is, I stick to my constant refrain "the truth is in between."
Well, I am about to start that refrain again, this time when it comes to the dueling likely voter models floating around the polling universe. Nowhere is the discrepancy between likely voter models more pronounced than in the Gallup tracking poll, which publishes two separate likely voter models every day. One likely voter model, the "traditional" model, includes questions about past voting behavior and assumes 60% turnout of the voting age population. The other, "expanded" model does not ask about past voting behavior, and makes no assumptions about national turnout.
Most Democratic-leaning election websites have decided to use the "expanded" model as the daily Gallup tracking poll number, rather than the "traditional" model. This is the case at TPM, Pollster.com, and fivethirtyeight and, it would appear, among most of the commenters I read on Open Left. I haven't taken sides in this argument before, but I actually think it is a mistake to use only the "expanded" likely voter model and discard the traditional one entirely. As I always say, the truth is in between.
In 2004, according to Dave Leip, 72.9% of registered voters had their votes recorded for President. Given ballot spoiling and invalid ballots, probably about 75% of registered voters attempted to cast a vote for President in 2004. It seems reasonable to assume that number will be either the same, or a bit higher, in 2008. However, both of the likely voter models in the current edition of the Gallup tracking poll shoot far on either side of that number. With 2,800 registered voters in their survey, Gallup uses 1,825 for their "traditional" likely voter model (65.2%) and 2,437 for their "expanded" likely voter model (87.0%). Both numbers strike me as highly unlikely, with one far too low and one far too high.
Yet again, I think the truth is in between. With the "expanded" model showing Obama +7%, and the "traditional" model showing Obama +5%, my best guess would be Obama +6%. This difference also probably goes a long way toward explaining the widely varying national polls of late, with current numbers coming in between Obama +3% and Obama +11%. The different polling methods and various forms of polling error appear to be balancing each other out, as Obama currently leads by 5.9% in Pollster.com's national regression line.
The Obama campaign will bring a lot of new voters to the polls. However, we are simply not going to reach 87% attempted turnout among registered voters. 80% might just be within reach, but 87% simply isn't going to happen. As someone who argued for months that Kerry would win 80% of the undecided vote, I believe we are all doing a disservice to ourselves by assuming such an implausible turnout wave. The best expectations are the most realistic expectations, and right now I am not at all convinced that we are being entirely realistic.
