The New York Times has done an interesting article on political markets, such as Intrade, something that I have been interested in for awhile. I wrote the author an email with respect to my own experience with such markets, which are valuable but still far from completely useful:
I share your interest in this, but one thing you fail to mention is the embarassment that many of us experienced on election day in 2004 when Intrade jerked around from one extreme to another. Intrade was worse than useless on that day. Another occasion on which it was similarly useless was during the Allen-Webb Senate election on election night where Intrade seemed to simply jerk back and forth based upon the reported returns--hardly the kind of predictive vehicle one can have faith in.
For what it is worth, my opinion is that Intrade does a good predictive job under a variety of situations but is basically useless once the polls open.