Originally Posted by cdcox
Silver covers that in his column today. That's really Romeny's only chance. He estimates the odds of the polls being biased enough for Romeny to win at 16% based on historical comparisons of polling results and election outcomes dating back to 1968.
This is why some of the other models (votamatic, princeton, etc) are all 95%-99%. They are basically presuming the polls are not biased, and if they aren't, then Romney's going to lose and maybe lose big.
The 538 model apparently believes in the possibility that all the polls are biased, and thats the only component of the model keeping Romney afloat, the model probably thinks there's almost no chance if the polls are not biased.