Sam Wang from the Princeton Election Consortium has released his meta-analysis of all available state polling. It predicts, with 99 percent certainty, an Obama win were the election held today:
Wang explains his approach here. He's using an extremely different analysis than Nate Silver of fivethirtyeight.com, and frankly, Silver's approach, which takes into account historical trends in electoral trajectories (namely, that races tighten, and states correlate) strikes me as a bit more realistic. I'm not near a good enough statistician to argue with Wennington's model, but even if McCain had only a one percent chance of victory looking at state polling data from a week or two, my sense is he'd have more than a one percent chance of victory if the election were to be held a week from Tuesday. All of which is to say, take this with the proverbial grain of salt: Elections aren't won until they're won.
Wang explains his approach here. He's using an extremely different analysis than Nate Silver of fivethirtyeight.com, and frankly, Silver's approach, which takes into account historical trends in electoral trajectories (namely, that races tighten, and states correlate) strikes me as a bit more realistic. I'm not near a good enough statistician to argue with Wennington's model, but even if McCain had only a one percent chance of victory looking at state polling data from a week or two, my sense is he'd have more than a one percent chance of victory if the election were to be held a week from Tuesday. All of which is to say, take this with the proverbial grain of salt: Elections aren't won until they're won. 
