I don't think he has a specific "model" for his "predictions". Reading his recent article, he uses statistics and weights informations gathered by polls to determine the probability of each candidate winning per state. You can't really be "wrong" in that sense, but the actual result can vary due to statistical sampling error, polling error, or bias in the polls.

Source: fivethirtyeight.blogs.nytimes.com/2012/11/03/nov-2-for-romney-to-win-state-polls-must-be-statistically-biased