Wins Produced?!

[jbrocato23]

All right, let's get back from the tangent.

Take two:

http://jbrocato.wordpress.com/2012/09/1 ... er-impact/

[Crow]

I saw your general discussion of weights and related issues but no formula. Are you willing to share that? Without it, I don't have much reaction.

[Mike G]

I take the other view. If the results are implausible, I don't really need or want to see the formulae.
Tyson Chandler is the 6th best player in the NBA?
Lou Williams is right there with Kobe, and ahead of Garnett, Westbrook, Pierce, ... He didn't even start for an average team.
Jose Calderon also ranks above a number of All-Stars.

It would seem you've fallen into the common trap of overstating the importance of certain numbers, while undervaluing others.
Even if your player values give you accurate team totals, that says nothing about whether players on a team are getting undue credit for that team's success.

As I suggested to open the comments: don't start with a fatally flawed argument and try to fix it. You're much better to start from scratch.

[xkonk]

Who gets to decide what's implausible? I've always thought it odd that Nick Collison would be a top-10 player (to pick one example), but people seem to like RAPM well enough.

[DSMok1]

Who gets to decide what's implausible? I've always thought it odd that Nick Collison would be a top-10 player (to pick one example), but people seem to like RAPM well enough.

I don't think anyone who likes RAPM actually thinks Nick is top 10. RAPM is just one of the tools for evaluating players.

[J.E.]

Who gets to decide what's implausible? I've always thought it odd that Nick Collison would be a top-10 player (to pick one example), but people seem to like RAPM well enough.

Whether the ranking of players matches up with personal beliefs is not exactly a good test of a player metric and should be avoided as much as possible

[xkonk]

Sure, you don't need to tell me.

I take the other view. If the results are implausible, I don't really need or want to see the formulae.

[Mike G]

I feel OK with my judgments of what is plausible and what isn't.
An argument which says the Indian Ocean is bigger than the Pacific is weak. You may agree or disagree.

Then there are many other metrics which (I feel) are rather more plausible, and which do not rank Chandler above Garnett, etc.

Tweaking a formula from [Rodman > Jordan] is taking the long way around. One could take a more direct route.

Seriously: Avoid as much as possible believing what you think you know?

[J.E.]

Seriously: Avoid as much as possible believing what you think you know?

The discussion of the analysis of basketball through objective evidence

He does offer some retrodiction results, which is a good and necessary step. Although it seems that he often arbitrarily assigned weights to certain actions, so it's not really fair retrodiction (would he have assigned the same weights 1 or 10 years ago?). He also used total team wins instead of 5on5-unit point differential. Reporting forecast error in terms of (R)MSE would be good, too

[jbrocato23]

The more I think about it, the more I take issue with my own subjective weights. I like the framework, but admit the subjective weights are an issue. Perhaps I could find them through a regression using a multi year APM or RAPM. But with so many terms, I fear an overfitting issue. Maybe I could use a regression on the basic stats and try to apply the values there to the more complex framework. Thoughts?


Also, Mike, I'm not using any of Berri's regression results or assumptions. Not sure why you're stuck on the notion that I'm trying to fix a broken model.

[Mike G]

Oh, heck. Sorry for the assumption. You started the thread, calling it "Wins Produced?".
And:

I like the framework, but admit the subjective weights are an issue.

And you have some wacky results, relative to most other metrics.

Carry on.

[Guy]

There is an interesting post by Arturo over at WOW breaking down Wins Produced into components, http://wagesofwins.com/2012/09/20/break ... /#comments, with a discussion that includes Daniel M. Some of the WOW group are coming close to agreeing with the subversive notion that, because WP sums to point differential at the team level, the true test of its validity is the ability to predict future, out-of-sample results. Sensing the danger, Berri himself has jumped in to try to derail the discussion.

Arturo's results are themselves potentially subversive. He demonstrates that the rebound component of WP plays a rather minor role in explaining wins at the team level, while offense is far more important. I hope Arturo will take the logical next step, and tell us how important each of these factors is in explaining Wins Produced at the player level. If Wins Produced "works," it should be most influenced by the same factors that explain team wins, and so rebounds should also play a modest role in explaining WP for individual players. However, if Arturo puts his ideas to the test and examines this, I think he will learn that player WP is driven very much by rebounds, while team wins -- as he has now shown -- are not.

[EvanZ]

Mutiny!

[Mike G]

Berri senses the danger in a theory that WP can be fixed?
Or that it cannot be fixed?

[DSMok1]

I appreciate Dave & Dre letting me finish what I had to say, there. Just getting a few things clear.