Another thought on Nate Silver

If poll errors are randomly distributed across states, then Obama has it won.  But if they are not, Romney still had a chance.   If how voters break is one state relative to polls is correlated with how they break in others, then if they break toward Obama, it makes no difference to the ultimate win-lose outcome; but if they break toward Romney, it does make a difference.

So statistical independence in errors across polls is good for Obama; correlation is bad for him.