today i've been reading mydd. i don't read it as often as dkos, atrios, or tpm because it gets pretty technical & wonkish. but that wonkishness makes it a fantastic place to go for detailed poll analysis, like this post explaining how undecideds historically break overwhelmingly for the challenger. chris bowers calls this the "incumbent rule" but you might see it referred to as the 50% rule or the 48% rule (i mentioned this briefly in my last "kerry wins" post).
it goes like this: any election with an incumbent is fundamentally a referendum on the incumbent. the electorate is intimately familiar with the incumbent, having lived with him for a full term by now. their decision is whether they like him enough to re-elect. if they don't, & the other guy isn't a total doofus, the challenger wins.
in this context, an "undecided" voter already has misgivings about the incumbent & needs to be convinced that the incumbent deserves their vote. if they aren't convinced when they get to the ballot box, they won't vote for the incumbent. historically, most undecideds aren't convinced come election day.
so the incumbent rule says that the real measure of who's winning is whether the incumbent has around 50% of the vote or more. if a race is bush 51, kerry 45, then (assuming the poll is even accurate, a big if these days) bush wins that state. but if the poll is bush 47, kerry 45, that really means the election would be more like kerry 52, bush 47, because the undecideds will mostly vote kerry. some adjust the magic number to 48%, saying that's the minimum the incumbent needs to even have a chance of winning.
this isn't bs: the mydd post has bunches of empirical evidence dating back decades, as well as links to other in-depth discussions of the incumbent rule and quotes from professional pollsters.
unless bush does something serious soon to turn his numbers around, he's toast.