Posted: Sat Apr 26, 2008 1:50 pm Post subject: Winshares
I know there are approximately as many ways to measure player value and performance as there are people doing the measuring, but I thought I would submit mine, and ask for some recommendations. Basically, I multiply contribution to team success by team success to arrive at some sort of wins created estimation which I call Winshares:
value = pts + as * 2 + tr + st + bk - to
value[value<0]=0
winshares=team wins * value / team total value
Now, I've already made it subjective by multiplying the assists time two. Without that scalar, point guards drop off remarkably. Now, I am very happy with the idea of multiplying value percent (val/team val) by team success. What I would be interested in, however, is ideas on non-arbitrary ways to compute the value part. I've looked at Win Score, and find it somewhat arbitrary... I've run every regression I can think of, and I'm not satisfied that any of the results reflect reality. I'd really appreciate some ideas, and maybe we can arrive at some sort of consensus.
Posted: Sat Apr 26, 2008 2:25 pm Post subject: Re: Winshares
dsparks wrote:
What I would be interested in, however, is ideas on non-arbitrary ways to compute the value part. I've looked at Win Score, and find it somewhat arbitrary... I've run every regression I can think of, and I'm not satisfied that any of the results reflect reality.
As David Berri has shown, regressions won't work at the player level, although they're probably a good start. The best work on player value -- certainly as the starting point for further work -- is Dean Oliver's stuff. An old version can be found on his site, but if you're really interested you should pick up his book. The starting point is a team possession, which is broken down into it's components (makes/misses, turnover, assists, etc), and variable credit given based on the relative scarcity of each element within the team context.
I am also fond of arbitrary models. They are useless for single answers in terms of ranking players (which may be what you want) but are great for answering "what if" questions. For example, what if credit for an assisted basket was one-tenth of the points scored? Nine tenths? What if a defensive rebound was worth the entire value of a possession? None of the value? It's a great way to include subjectiveness in a semi-objective context. I've suggested this approach a few times, but nobody has done anything with it (including me). This is a little surprising -- it means less work in justifying a model, since you end up with a bunch of models instead of one, each using a particular arbitrary premise as the starting point.
Anyway. I like arbitrary, is what I'm trying to say. _________________ ed
Joined: 05 Jan 2005 Posts: 609 Location: Columbus, OH
Posted: Sat Apr 26, 2008 3:03 pm Post subject:
I came up with a method that I call Win Shares over three years ago. (I don't care what you call it, I'm just saying that you might want to change the name to avoid confusion.) I make extensive use of Dean Oliver's metrics. The Win Shares method is described in detail on my site. _________________ Regards,
Justin Kubatko
Basketball-Reference.com
Joined: 05 Jan 2005 Posts: 609 Location: Columbus, OH
Posted: Sat Apr 26, 2008 3:35 pm Post subject: Re: Winshares
dsparks wrote:
value = pts + as * 2 + tr + st + bk - to
Another thing I wanted to mention was that I think you need to show that there is a strong relationship between the above and team wins. My guess is that "value" would be a relatively poor predictor of team wins. _________________ Regards,
Justin Kubatko
Basketball-Reference.com
Joined: 13 Oct 2005 Posts: 524 Location: Atlanta, GA
Posted: Sat Apr 26, 2008 4:49 pm Post subject: Re: Winshares
This is essentially the same as Martin Manley's old "credits", or NBA Efficiency, except with the 2* multiplier for assists and no penalty for missed shots.
jkubatko wrote:
dsparks wrote:
value = pts + as * 2 + tr + st + bk - to
Another thing I wanted to mention was that I think you need to show that there is a strong relationship between the above and team wins. My guess is that "value" would be a relatively poor predictor of team wins.
The R-squared between team Value/Min and Pyth% since 1973-74 is .153.
Posted: Mon Apr 28, 2008 12:11 am Post subject: Re: Winshares
dsparks wrote:
I know there are approximately as many ways to measure player value and performance as there are people doing the measuring, but I thought I would submit mine, and ask for some recommendations. Basically, I multiply contribution to team success by team success to arrive at some sort of wins created estimation which I call Winshares:
value = pts + as * 2 + tr + st + bk - to
value[value<0]=0
winshares=team wins * value / team total value
Now, I've already made it subjective by multiplying the assists time two. Without that scalar, point guards drop off remarkably. Now, I am very happy with the idea of multiplying value percent (val/team val) by team success. What I would be interested in, however, is ideas on non-arbitrary ways to compute the value part. I've looked at Win Score, and find it somewhat arbitrary... I've run every regression I can think of, and I'm not satisfied that any of the results reflect reality. I'd really appreciate some ideas, and maybe we can arrive at some sort of consensus.
And here's the money part (at the end of the post):
I look forward to hearing everyone's comments and criticisms. Also, any comments on the graphic would be appreciated.
This is pretty much how I do win shares on a seasonal level - although I use different weights that what you are using.
Another way that I like to do as comparison (and has already been mentioned) is minutes played - divvying up wins proportionate to minutes played of each individual on a team. I think this works well as a career thing - and eliminates the possible bias of individual weights. _________________ Statman
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum