Early Voting and the Voter File
In a rare occurence, Nate Silver writes something today that is, at a minimum, imprecise. Explaining why he's optimistic about an Obama victory, he says that "McCain is pushing back against the fact that Obama is banking votes every day with a substantial national lead." This is not the only time I've seen this argument, so I don't mean to pick on the 538 crew. But in my view, this really oversimplifies the difference between early and regular voters.
Look at the view expressed by the sentence above. It's implicitly assuming that the early-voting electorate mirrors the general electorate. If Obama is surging nationally, this surge will show up amongst early voters as well. But this might not be a reasonable assumption.
In general, political scientists have found (PDF warning) that early voters are more educated and more interested in politics. Even controlling for this, they are also more likely to be intense partisans, and have more extreme ideologies than their peers. This is at least somewhat intuitive; for example, I personally have known whom I'll vote for since the primaries were settled. If your mind is made up, you are less likely to delay; and you are more likely to seek the immediate gratification of registering your support for your candidate TODAY.
What this means is that support for candidates amongst people voting early is probably much less volatile than support amongst the general population. Which is not to say that the Obama surge has no impact on early voting; to the contrary. But getting into that will require more depth than this post alone can provide; I will be returning to the subject repeatedly as the election gets closer.
Agreed
It's not ENTIRELY irrelevant - there's almost certainly some population segment that both votes early and remains persuadable (essentially, voters who are potentially going to regret the vote they cast later). But as you - and empirical evidence - assert, early voters are likely to be more intensely partisan.
A parallel issue worth considering, though, might be the enthusiasm gap as it relates to polls. Consider, you're highly partisan but not highly motivated to vote. You may not be a safe bet on election day. But you may wind up casting an early ballot - less hassle - for your candidate depending on his pre-election poll standing. This argument could swing both ways: people want to get behind a winner, so voters who might sit out on a close race vote early for the leader when they see who it is; conversely, people who root for the underdog see one candidate clearly behind and come out to try to even things up in the pre-election balloting. Certainly not a proven phenomenon, but a potential one. How might electoral likelihood affect the size of early-voting populations for each candidate?
(Incidentally, I'd suggest different bimodal distributions for each candidate. The key difference between candidate distributions (poll strength vs. early vote size) is probably based on the candidate's GOTV effort and perceived charisma. Within distributions, I suggest bimodality based on two psychological dimensions: high self-monitoring should increase the probability that a voter supports a winning candidate early, low agreeableness should increase the probability that a voter supports a losing candidate early. Just a thought)