Updated with correction – the model also predicted Indiana incorrectly.

I’m going to do a more thorough analysis in the next couple of days, but I think I can say that the model worked very well yesterday. It predicted the times that Pennsylvania and Virginia would be called accurately (it was 33 minutes late for PA and just 5 minutes or so early for Virginia), and also that Obama would be elected president with the calling of the Western states at 8pm Pacific.

The only two states it predicted incorrectly was were Florida and Indiana, which the model predicted was were going to go to McCain, based on George Bush’s over-performance there in 2004.  As far as predicting when states would be called, it correctly predicted that Missouri and North Carolina wouldn’t be called until today (although it thought John McCain would have been called as the winner in Missouri in the early hours of this morning).  Overall, I believe it predicted the call time correctly (or within a few minutes) in about 40 states, although I’m still trying to find the exact time CNN called all the states (if someone knows where to find the info, perhaps they could post it in the comments?  Their Political Ticker blog seems to have most, but not all of the call times).

The model’s one real miss was Ohio, which was projected to be still too close to call.  That, and Florida and Indiana, look to be the two states where Obama significantly out-performed compared to 2004.

Not bad, given that I constructed the projection a week before the election.

One other thing… USC has posted the video from last week’s panel here.  It seems to work for me on Windows, but not on my Mac.  Your mileage may vary.