beyond polls (or the past is a good model for the future or has nothing changed)?
Over the last week, an anonymous blogger who writes under the pseudonym Poblano did something bold on his blog, FiveThirtyEight.com. He posted predictions for the upcoming primaries based not on polling data, but on a statistical model driven mostly by demographic and past vote data. His model predicted a 17-point victory for Barack Obama in North Carolina and a 2-point edge for Hillary Rodham Clinton in Indiana.
Critics scoffed. Most of the public polls pointed to a close race in North Carolina. Looking back at Poblano's efforts in Pennsylvania, pollster Dick Bennett decried the models as "stepwise regression run amok." Slate's Mickey Kaus predicted failure for "a sophisticated model that ignores... what's been happening in the campaign. Like Rev. [Jeremiah] Wright."
But a funny thing happened. The model got it right. [source]
http://www.fivethirtyeight.com/ (stat city! for the life of me i can't figure out what's going on. also note that his methodology does involve input from polling, to supplement the demographic model)