Silver claimed that his model was better because it predicted a high correlation between PA/MI/WI.

A model that predicts a 30% chance of winning the election will be wrong 1 out of 3 times, which is not quite a coin flip but close enough.