I just wanted to share an observation that I had about polls, which relates directly to the 'Keith Number' often talked about post-New Hampshire. Basically, Olbermann soundly pointed out that many of the polls we looked at were profoundly incomplete: in addition to their ~5 point margin of error, they also had large portions of the electorate responding 'uncommitted'.
As I see Kos posting piles of polls, I'm as excited as anyone else to see Obama doing well - but I'm only REALLY relieved when I see him breaking 50%. Until that point, take it all with a grain of salt - not only will the numbers change, but we are certain that the polls are hiding something right now.
As an aside, does anybody more versed in stats know why 'undecided' is an option on a poll? It's obviously a compromise, because throwing out anybody who refused to pick between the two introduces its own set of errors - but my intuition is that it would be a much more useful number.
How many ballots actually have an 'uncommitted' on them, and how often is it actually picked? I see no reason (but am willing to be proven wrong) why the poll condition should allow that easy out when the real deal wouldn't.