Nate Silver's “Polls Only” model proves to be the most accurate of 538's three models on election day.

Created by NathanMcKnight on 2016-08-11; known on 2016-12-16; judged wrong by Bruno Parga on 2016-11-14.

• NathanMcKnight estimated 60% on 2016-08-11
• Bruno Parga said “Aren’t all three models supposed to converge by election day? Or at least the polls-only and nowcast?” on 2016-08-12
• two2thehead said “points up What brunoparga said.” on 2016-08-12
• sflicht said “measured how?” on 2016-08-16
• NathanMcKnight said “brunoparga: I think they’re supposed to converge in general, but I wasn’t under the impression they would converge precisely. I imagine there will be some variation. ” on 2016-08-19
• NathanMcKnight said “sflicht: Good question. I think the way to do this is to determine which of the three final probability estimates most closely approximates the candidates’ vote shares.” on 2016-08-19
• NathanMcKnight said “…I realize that’s kind of apples-to-oranges, but shouldn’t the one be a good proxy of the other? ” on 2016-08-19
• HonoreDB said “It’d be work that I hope 538 does for us, but we could take the average predicted popular vote share across the lifetimes of the models. I don’t think comparing probability estimates to vote shares gets you very far—80% of the vote=99+% probability.” on 2016-08-20
• Bruno Parga said “All three models had Clinton with 48.5%. Polls-only and now-cast gave Trump 44.9%; polls-plus gave him 45.0. Given that this broader in scope than looking at specific state predictions, I suggest we use this to consider this prediction wrong.” on 2016-11-09
• Bruno Parga   judged this prediction wrong on 2016-11-14.