This article is a few months old, but still relevant. It's written by a political journalist over here who is one of the main election-night pundit/psephologist and is one of the most accurate at calling elections in the Australian media (probably not the pre-eminent journalist for behind-the-scenes deals or day-to-day governance, but in terms of predicting elections and commenting on elections, he's one of the best).
http://www.abc.net.au/elections/qld/2009/guide/predictingelections.htmHe's updated it to discuss the impact of the redistribution (redistricting) prior to the Queensland election, but since presidential elections don't incorporate redistricting, that's not overly important. Still makes for an interesting read, though.
Basically, he says that the swing is unrelated to the size of the booth/precinct, so he works off that rather than the actual vote. Of course, we don't have exit polls in Australia (well, actually, I think we did recently - but given that our federal electorates are a quarter of the size of your congressional districts... the inaccuracy makes it pretty much pointless).
