Editor-in-Chief, @FiveThirtyEight. Author, The Signal and the Noise (amzn.to/QdyFYV). Sports/politics/food geek.

Search Tweets

Twitter Feed

@NateSilver538 — 2,757,771 followers, 17,238 tweets

Nov 18, 2017

@PhilBirnbaum I guess my overall view is that Elo does "surprisingly well" despite somewhat obvious flaws, i.e. as well as more complex methods like Sagarin/Massey. I think allowing for intra-season variation in talent/form is a part of that—it's sort of both a bug and a feature of Elo.

Nov 18, 2017

@PhilBirnbaum In theory, you could vary K based on the point in the season. Also, the optimal K for predicting game n may be different than for game n+10. But in general, when I've experimented with adding wrinkles to Elo, it doesn't help w/prediction too much & detracts from its simplicity.

Nov 18, 2017

@tangotiger @PhilBirnbaum In Elo math, that's basically the k-factor. And the k-factor varies a lot from sport to sport. The value we chose for the NBA optimizes predictive power for the team's next NBA *game* (not necessarily the next *segment*, which may be different).

Nov 18, 2017

@tangotiger @PhilBirnbaum I don't think you guys are recognizing how much more deterministic the NBA is than MLB. Intuitions that seem right for MLB can be way off for the NBA, and vice versa.

Nov 17, 2017

@julia_azari You haven't lived until you've typed a Stata command into Slack.

Show More