The precise details of Nate Silver’s models are not public but I would have thought it safe to say he, and his team, do not just assume straightforward linear relationships. Could be wrong though…
Regardless, I agree with the criticisms you raise of coverage that predicted Clinton as having a 99% chance, etc. but that wasn’t by Silver or anyone worth paying attention to. Indeed, Silver was publicly critical of other polls that displayed such unwarranted certainty. Talking about the 2016 election coverage he said:
Even a 2- or 3-point polling error in Trump’s favor — about as much as polls had missed on average, historically — would likely be enough to tip the Electoral College to him. While many things about the 2016 election were surprising, the fact that Trump narrowly won when polls had him narrowly trailing was an utterly routine and unremarkable occurrence. The outcome was well within the “cone of uncertainty,” so to speak.
That makes sense to me. And it relates to the issue of comparing outcomes with predictions and whether the 538 model was ‘right’. Again it makes more sense to me to approach them as probabilistic predictions than binary outcomes and in all the swing states the margin of error still includes a potential Trump win. This doesn’t mean that the polls, and subsequently, 538’s models did not underestimating support for Trump. They did. But not to the degree that is widely assumed.