October 14th, 2008

term

Election statistics

A couple of years ago some faculty and grad students in Computer Science here and at two other universities did some interesting simulation work involving Bayesian statistics and state election polls, and have recently applied their methodology to the current presidential race. Their result is up on one quad of the giant video wall at the Siebel Center where it's attracting a LOT of attention, and is on the departmental web site here.

At the moment it appears to be giving a 1.00 probability of Obama winning the election even in the "strong Republican swing" scenario. I wonder how good their prediction really is and how likely it is to drift around in the last weeks before the election. I read their paper and it makes sense to me although it's been an awfully long time since I had probability and statistics and have only seen the Dirichlet distribution once, in CS 441, so I'm not the best critical reviewer.