Nate Silver is the one garnering all the headlines, but the Princeton Election Consortium is another blog I follow that does meta-analysis of state polls. In 2008, their prediction came within 1 electoral vote of the final outcome, beating out FiveThirtyEight, among others.
Currently, Princeton Election Consortium forecasts Obama winning 303 electoral votes, giving Obama a 99.8% chance of winning re-election. FiveThirtyEight currently has Obama winning 307 electoral votes, with an 86.3% chance of winning re-election.
Silver has been the target of much recent criticism from pundits, and I've seen many people argue that stating Obama's chances at re-election as a percentage is meaningless. That is, if Silver says Obama has an 86.3% chance of re-election and he loses tomorrow, Silver loses all credibility.
No doubt Silver would take a fall from grace in the eyes of many if Obama loses tomorrow, but the percentage is meaningful. It's not the exact same analogy, but if I tell you a fair coin will come up heads 50% of the time and you flip it once and it comes up tails, that doesn't mean my percentage was wrong. Silver and others who are analyzing these state polls suffer from only having one trial. If you could run the election multiple times instead of once, the sampling error would likely decrease and the % of the time Obama wins should converge on Silver's prognostication, according to his math. This is why high level poker players will often ask a hand to be run out multiple times, or why if you count cards you have to play blackjack over an extended period of time to squeeze out a profit from your edge.
If Silver retires from the political space after this election (and from past things he's said it sure sounds like he's ready to move on to a new field; who can blame him), I'd love to see him release his methodology publicly, like open sourcing software. It will ensure his legacy lives on in this field even as he goes on to point his methodology at another problem space.