I think the "probabilistic-based predictions" are best, and here's another one. At UMN, to boot.
With 3 weeks to go it seems that get out the vote will be absolutely critical in close states.
