I was going to post this in response to Desidero on a different thread, but figured I could just share w/ the whole class!

There's a freaky blog up by Nate Silver (for baseball nuts he's the one with the PECOTA algorithm that predicts player and team performance with scary accuracy).  Silver's algorithm also predicted primary results down to the percentage split.  Anyway, it seems that polls are good predictors when you pull vast quantities of data from the right responses, combine them with demographic data and voting patterns from past elections and monitor them over the long term - very cool!