View Single Post
Old 12-18-2019, 03:47 PM
NAF1138 is offline
Join Date: Aug 2004
Location: North of Philly
Posts: 10,148
Originally Posted by mhendo View Post
It's not that the model is breakable, per se. Rather it's that any model, however good, is still prone to outliers and statistical fluctuations and runs of luck.

Look at baseball stats like Expected W-L (X W-L), or Batting Average on Balls in Play (BABIP). It's possible for a team (XWL) or a player (BABIP) to have a whole 162-gamee season where they defy the expectations of the model. I've seen teams over- and under-perform their pythagorean run-based X W-L by ten or more games. And I've seen players over- or under-perform their BABIP expectations by 40 or 50 points in a season. But overall, both those models are still pretty good predictors of performance, despite the aberrations.
If anyone is into podcasts? If so this is a pretty good analytics based podcast, called the double pivot, that goes into all this here.

Gonna try to share directly from pocket casts. Let me know if this doesn't link right.

Last edited by NAF1138; 12-18-2019 at 03:48 PM.