Im breaking summertime bloggins silence and my book-writing discipline to comment on Peters post from below about the rising number of votes cast weeks before the election by absentee. This is a perfect phenomenon for quantitative political scientists to study. My own hunch is that most early voters are partisans who have their mind made up, and are not likely to be swayed by late news, a last debate, etc. So it may not really make any difference to the outcome, though it will mean an even more maniacal focus on the "undecided" voters, who are typically a confused lot.
Of course we dont really know. Ive asked around a few DC political scientists (i.e., Norm Ornstein) if we have any idea about the cohort that voted early in 2000, and the answer is, we dont.