I've been thinking quite a bit about polls lately, about what can go wrong with them, and how they might react and even how some of Nate Silver's critics have almost stumbled onto a couple of valid points (albeit ones that Silver has acknowledged).
I'm working on a full post on the subject but in the meantime, here's some background complete with sound track. (hit play now)
One of the things I've been coming up against is what can happen when something moves a poll.
Let's say a candidate gives a rousing speech that gets base fired up. It makes supporters more likely to talk to pollsters but it doesn't really change anyone's mind. After a short period the excitement fades and the polls return to their previous position.
Another speech, but this this time with enough substance to win over some undecideds. The polls move to another position and stay there.
This starts with an event that would normally cause an elastic shift, but the deformation of the polls is drawn out, either because it's followed by other trivial events or because the timing of the polls makes it seem drawn out or because the media simply decides to dwell on it or (most likely) some combination. Under these circumstances we can easily get a feedback loop where bad polls create the impression of eroding support which causes supporters to become discouraged and money to dry up which causes support to actually erode.
I suspect Romney was experiencing creep in the last week of September.
I'll try to flesh this out more in the next 24. In the meantime, does anyone have any thoughts on the subject?