Don't miss Andrew Foland's 5-part tour de force on how to incorporate pre-game estimates of win probability with the in-game estimates. Awesome job, and I really appreciate the time Andrew has put in. This is something I've been asked to do for a while. I think it's a great idea, and I developed a very similar method to do it.
However, I would be reluctant to make it the 'official' WP or WPA model. If the Patriots have a game against a weaker opponent pegged as a 70/30 match-up, and Tom Brady plays lights-out, should he only get 0.30 WPA instead of 0.50 WPA? Should he be penalized for being favored pre-game?
Bryan Davies expounds on the go-for-the-2-point-conversion-twice strategy teams should be employing when down by 2 TDs.
Jim Glass on the eight stats he hates most.
Thanks to everyone who has made the Community Site such a success this season, and to editor Ed Anthony in particular. Keep the posts coming!
New Posts at the Community Site
By
Brian Burke
Subscribe to:
Post Comments (Atom)
"If the Patriots have a game against a weaker opponent pegged as a 70/30 match-up, and Tom Brady plays lights-out, should he only get 0.30 WPA instead of 0.50 WPA? Should he be penalized for being favored pre-game?"
The idea would be to hand out .20 extra points of WPA to Patriots' players before the game and dock the weaker opponent .20 points. No matter who wins, that team will end with .50 totals points of WPA and the loser will have -.50. Apportioning out pre-game credit is tough, though. Back to the Community!
Tango's take.
Thanks for the kind words.
I think Brian, Tango, and I are all in agreement that for tracking player quality / play-by-play importance in general, WP should start from 0.5. As I suggested, if something were implemented to include the prior matchup probability, it should probably be in the form of an option / checkbox. (Because the servers aren't already overloaded!)
As I mentioned to Ed, I stil have the academic habit of writing the "minimum publishable unit" so my CV shows five articles instead of one :)
The main reason I went into all the detail is because I think the conclusion ends up very surprising to many people--namely that team quality has a pretty minor impact on win probability towards the end of games. And I think that is going to be the generic conclusion no matter what form you use to incorporate the matchup probabilities. So if you implement your own system one day Brian, the articles will stand as an independent attestation to the conclusions you reach. (As long, of course, as they are the same.)
I'm most curious what you came up with, Brian! Especially if you didn't cut so many approximation corners as I did. The normal-gaussian is the one that sticks out in my mind as the most likely to make a non-negligible difference (and also the one tricksiest to remedy.)
The reason that team quality has a minor impact on win probability towards the end of games is that the better team should already be winning by the end of the game. If they aren't, then they aren't the better team. The remaining advantage that they have is because perhaps they are better but have just been unlucky up until that point in the game.
It's been my impression that the system of stats used by this site boil down to the idea that a better team or better player has a small advantage every play that adds up over the course of a game or a season. By the time you reach the end of a game, those small advantages should have added up to the better team having a lead. If the better team is tied at the end of the game, there aren't enough plays left for that small advantage to amount to anything substantial.