There is a tendency to lionize Nate Silver on these boards because his model picked our guy this time -- but it's just a predictive model, it's not a "liberal" predictive model. I was skeptical about all of the praise Silver was getting because his predictions for non-presidential races just seemed wrong. Especially because he gave Tester and Heitkamp no shot in their senate races. Heitkamp had a "now cast" of 8% on election day -- less than Romney -- and she WON.
If I had to guess at the problem with his model, it's that his "secret sauce" overweights the historic lean of a state. I think you do better focusign on the current data -- the polls. So even though the polls were showing a win for Tester and a nail biter for Heitkamp, the 538 model discounted those polls. But the polls, not the "secret sauce", were right.
So why do I bring this up now. Because all this love for the 538 model is based on it predicting an Obama win this time. Next time it may predict the Republican. If it does, we shouldn't care. Go out and vote, and get people to the polls, and raise money for candidates, because only the election day results, not the 538 model, matters.