Part 2: Nate Silver is not the Data Messiah

Okay, so I'd originally planned to write a Part 2 that was about something else, and to do it on Monday, but I've had people ask me to address 538, or more specifically, how everything on 538 is tanking and it's down to nearly a coin flip and oh God we're all going to die in the Trump Camps.

Okay, to begin with: 538 isn't the only poll aggregator/modeler out there. Princeton has one. I've heard Princeton is kind of a reputable joint. The New York Times, the 'grey lady of journalism', has one. Perhaps less reliable journalists have some. Now, knowing this: watch this great video.

There is no way I should be in a position to comment
on someone else's hairline, and yet.

Note that, from about the first week of September forward, 538 has consistently given Clinton the lowest chance of success. Nate's model is always more bullish for Trump. To the point where just over a week ago he wrote an article specifically addressing that point, using the word "bullish" because I'm not the only one who uses that term, god dammit.

Anyways.

I want to address one of the specific points he brings up, because to me it's pretty obvious where things are wrong. His second point in that article is that he includes polling data going back to 1972 in order to make his models, and he inherently implies that by including more data, he is naturally more correct. But that's a fallacy. Let me put it this way: if a doctor told you that he was using all of the tools at his disposal, including ones they stopped using in the '70's, would you trust him more or less?

Back when people who despised each other and everything
they stood for respected the process enough to feign civility.

In 1980, polling organizations consistently called the presidential race "too close to call", and admitted complete surprise when Reagan shellacked Carter 51 - 41. The polling data in here is in Nate Silver's model, and not in other people's models. Silver admits that if you take the pre-2000 data out, his model immediately jumps to be much more in line with everyone else.

538 gets a lot of press, and Nate Silver somehow got stuck in the public consciousness as "the guy who got it all right"  even though Sam Wang did it earlier and with more consistently correct results since 2004. Wang and Silver have even had public back-and-forths over how the other guy's model sucks.

If you want to take it in a much larger picture, consider this: Nate Silver aggregates polls, adding them all together and averaging it out* on the assumption that if one polling agency is off in one direction, standard statistical deviation makes it likely that another polling agency is off in the other, and once you have enough in the pile, you can get a true average. So throw 538's chances for Hillary on the pile, average it with the other polls trackers, and she's not at 538's dire sub-70, but at a much more respectable near 90s.




*Yes, yes, it's more complex than that, but being specific that he applied weights and corrections means I have to explain more before I get to the analogy which is really unaffected by the additional facts.