I agree with Silver that Trump is unlikely to be the nominee, but I really hate the mentality that if you dare to criticize Silver, his website, or any of his proprietary models, then you're "disagreeing with the math." Regression is not some magical tool that can instantly turn data into rock-solid conclusions. Underlying any statistical model is a series of assumptions that can be essentially impossible to independently verify. And the truth is that Silver's models haven't performed all that impressively in the past. Silver would be the first to admit that his feat of "calling" nearly every state correctly in the 2008 and 2012 presidential elections was overrated. There aren't that many competitive states in a given election, so absent some systematic polling error, calling states just isn't that hard. Look at an election where a good model might really add some value over just looking at the polls and his models just don't look like anything special.
I think one big issue is that a lot of Silver's early criticism came from bloviating pundits (Politico), partisan hacks with no incentive to be objective (Rove), and just plain idiots (the Unskewed Polls guy). It's hard to be critical of him without getting lumped in with those people.
I don't want to give the wrong impression here, I actually do like a fair bit of Silver's work. If nothing else, it's good to get people to understand how to think probabilistically. I just wish people would take a more balanced view of the strengths and weaknesses of his work.
