Nate Silver is crowing that his model predicted the actual outcome as “most likely” but that’s a bit like saying that exactly 50 heads is the most likely outcome in a series of 100 coin tosses (which it is) despite its still being extremely unlikely. To actually see results so close to the predicted means in all states is an indication that the variances were grossly over-estimated and so that the likelihood of a Trump win should have been estimated as much higher.
According to commenter ‘Gabe’ on another of Silver’s posts:
Model held pretty well (using NYT website for result numbers):
PA: Model R +.1, Result R +1, Difference .9
MI: Model D +1.2, Result R +1.4, Difference 2.6
WI Model D +1, Result R +.9, Difference 1.9
NC Model R +1.1, Result R +3, Difference 1.9
GA Model R +1 Result R+2, Difference 1
NV Model R +.6, Result R+5, Difference 4.4
AZ Model R +2.4, Result R+5, Difference 2.6
Uuh! NO!! Although these are (almost) all within $#\pm#$3%, the differences are ALL in the same direction – which has almost no chance of happening if they were truly random.
So Silver’s “gut” feeling was a much better predicter than his actual model.
Source: Comments – The story of Trump’s win was foretold in New York City