APBRmetrics

**mtamada** · Joined: 28 Jan 2005 Posts: 377

One topic which was debated at least a couple of times on the old apbr-analysis list was, given a Bill Jamesian "win shares" formula, should there also be "loss shares" (e.g. as with DeanO's player win-loss records).

Bill James, on the SABR StatisticalAnalysis list, recently mentioned that he is developing a system which includes loss shares as well as win shares. He states that measuring defense is still a problem, especially with park effects, but his current version of the system rates Manny Ramirez (223 wins, 58 losses) above Alex Rodriguez (213 - 70).

He gave no details about how the system works however, presumably he's still tinkering with it and will publish another book (he didn't say that either though).

But Bill James, who initially had win shares only and no loss shares, is now putting loss shares into his formula. (He'd previously said in interviews that the reason he didn't have them before was because he couldn't figure out how to do them. But I guess now he does have a theoretical model.)

That is not to suggest that basketball analysts have to do what Bill James does. But it is one person on the pro-loss share side of the discussion.

jkubatko · Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH

I saw that on the SABR SAC list and was intrigued. I threw together a Win Shares system for basketball a while ago, but I have been wanting to tweak it. If you have any suggestions I would welcome them.
_________________
Regards,
Justin Kubatko
Basketball-Reference.com

HoopStudies · Posted: Fri Sep 23, 2005 9:59 am Post subject: Re: loss shares

Kevin Pelton · Posted: Fri Sep 23, 2005 10:00 am Post subject:

Personally, I prefer Hardball Times' Win Shares Above Bench method. I think two numbers for a given player is more than I can have in my brain at one time.

Ben · Joined: 13 Jan 2005 Posts: 266 Location: Iowa City

**mtamada** · Joined: 28 Jan 2005 Posts: 377

**mtamada** · Joined: 28 Jan 2005 Posts: 377

Ben · Joined: 13 Jan 2005 Posts: 266 Location: Iowa City

jkubatko · Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH

jambalaya · Joined: 30 Jan 2005 Posts: 282

I don't know how they do it in baseball and decided to think about it anew rather than direct mimic or mirror any current type of win share method for basketball.

If you wanted loss shares, how about looking at +/- on court only ratings for the season (thus far or the whole season after it is done) for players compared to the same ratings for a lost game and do the same thing for net PER against opponent differential for game compared to season.

There will be four basic groups.

1) Players with negative +/- on court only ratings for the game compared to season average and negative net PER against opponent differential for game compared to season.

2) Players with negative +/- on court only ratings differential and positive net PER against opponent differential.

3) Players with positive +/- on court only ratings differential and negative net PER against opponent differential.

4) Players with positive +/- on court only ratings differential and positive net PER against opponent differential.

You could have a simple scoring system and give two loss points to those in group 1, one to those in groups 2 or 3, and none to those in group 4. Divide personal loss points by the total and you get loss share for that game.

Or you could sum the two ratings differentials together and give loss points to anyone that is negative in the amount of that combined measure divided by 2. Divide personal loss points by the total loss points awarded and you get loss share for that game.

Or if you wanted to give more weight to individual performance and less to the team +/- data maybe cut that team rating by dividing it by (up to) 5 (for 1/5th the team performance blame) and then adding up the two ratings instead of doing a simple and equal weighting. That is probably too far, maybe 3/5ths on net PER differential and two-thirds on overall team +/- performance might be close to fair.

The players are all in it together, but have different levels of importance. Stars are held to their star level performance. Role players are judged by their lesser role and stats. You can be the best player in many ways but if you play below your standard you get part of the loss. Not everybody gets a share of a loss but the standard is a little higher than just do no apparent game specific team or personal statistic harm by these two ratings, you have to separate yourself and play better than usual to avoid a share of the responsibility.

I used differentials because it seemed that if you used just raw game +/- and net PER ratings, it would tend to let stars off the hook a lot of time and maybe give too much of the loss share to scrubs. Using game to season differential seemed more fair but maybe there would be a middle ground whereby your shortfall penalty is not as severe if you are still positive on the raw game rating or if the shortfall is less than a certain amount.

Some type of a loss share evaluation would be worthwhile. I thought I'd step in and throw out one method to get it rolling but hopefully others will do so also. And after the idea is refined, maybe someone will want to calculate and publish them.

HoopStudies · Posted: Sun Sep 25, 2005 9:22 pm Post subject:

I hate doing this, but let me ask the question:

Why do we need another player rating method?

Is it going to revolutionize things? Are people going to throw out all the other methods we have? Is it going to lend new insight? Or are we just following something that someone else put out in another sport?

Maybe there is a valid answer but the question has to be asked.

I know that I'm quite happy with the methods out there. I don't use 95% of them unless I need to replicate someone else's work.
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.

bchaikin · Joined: 27 Jan 2005 Posts: 689 Location: cleveland, ohio

Why do we need another player rating method?

man's quest for knowledge?...

Is it going to lend new insight?

one way to find out is to put it up to peer review in a forum such as this, correct? i think most here are more than happy to lend their opinions...

i like the idea of people always coming up with new methods for statistical analysis of the nba, and that they present them here of all places. you never know when that one goldmine - or our own voros mccraken - will appear, but i think we'll all recognize it when we see it. that's the fun part...

jambalaya · Joined: 30 Jan 2005 Posts: 282

Thanks Bob. I appreciate the support for search and discussion.

I decided to post on this subject because i thought m tamada deserved a basketball response to his question. (But it doesn't stay that simple.) I was a little disappointed with the sound of your comment Dean (at least to my ears), since you made such heavy efforts to expand the number of ratings. The door is not closed, all the good ideas have not been published, and the community can still improve on ideas or add new ones. I assume you are also still relatively open to a true forum as well and just had a cranky or weary moment or maybe I heard something a little different out of your statement than you intended but this response seemed necessary to elucidate your position further.

My proposal is a first draft but I believe it is more useful than commonly used simple player on court W-L records in asssigning responsbility for losses. The field has seen a stream of different ratings, learning along the way and it will see plenty more. I still think that is a good thing. People often say this is the infancy of stats developement and use, so why stand still?

In a 1998 article on your JOBS site "What Do Individual Win-Loss Records Mean?" you looked at three player win-loss methods and said:

"So Which Is Best?...
I frankly wouldn't have presented all three methods above if I didn't think they all were good. "Which is best" is a question with no answer. It's like asking a guy whether a hammer is better than a screwdriver. "

Whereas your Pythagorean Method divided win/loss study into individual offensive and defensive ratings, my method divided it into team +/- and net PER comparison. Different methods, different answers, different origins. But both tools, working with two dimensions instead of just one and have offense and defensive information in them, though mine is sliced differently.

More tools, more information available, more knowledge developing power and torque, more choice... sounds good to me.

Ed Küpfer · Joined: 30 Dec 2004 Posts: 785 Location: Toronto

HoopStudies · Posted: Mon Sep 26, 2005 10:44 am Post subject: Re: More methods?