Win Shares versus Wins Produced?

Moderator: Doctor MJ

User avatar
floppymoose
Senior Mod - Warriors
Senior Mod - Warriors
Posts: 57,263
And1: 15,721
Joined: Jun 22, 2003
Location: Trust your election workers

Re: Win Shares versus Wins Produced? 

Post#21 » by floppymoose » Sat May 1, 2010 10:06 am

Give me a big ole pile of +- and I'll be happier than with either of these stats.
lorak
Head Coach
Posts: 6,317
And1: 2,231
Joined: Nov 23, 2009

Re: Win Shares versus Wins Produced? 

Post#22 » by lorak » Tue May 11, 2010 6:56 pm

http://dberri.wordpress.com/2010/05/10/ ... des-again/

I especially recommend last part, named “Once Again, jbrett’s Coded Responses”

For example:


K. I saw that someone regressed their model (i.e. adjusted plus-minus or APM) on your model and this devastating and damning test (it is important you use the words “devastating” and “damning”) proves that your model is wrong. I hold this belief because 1) I am able to ignore the fact it is not clear which model is being tested, 2) I am able to ignore the obvious shortcomings APM (very inconsistent over time, most player evaluations are statistically insignificant), 3) I really have no idea how one would actually evaluate an empirical model.

L. All you’re doing is arbitrarily dividing up the team’s wins amongst the players, then making a team adjustment to hide any discrepancies. I can say this because I never read the books or the underlying empirical articles. I have read many “experts” on the Internet (and you know they are “experts” because they said so).
mysticbb
Banned User
Posts: 8,205
And1: 713
Joined: May 28, 2007
Contact:
   

Re: Win Shares versus Wins Produced? 

Post#23 » by mysticbb » Tue May 11, 2010 7:10 pm

DavidStern wrote:http://dberri.wordpress.com/2010/05/10/the-wow-all-nba-team-and-the-jbrett-codes-again/

I especially recommend last part, named “Once Again, jbrett’s Coded Responses”


Lol, how insecure. Shows more what kind of a douche bag Berri is than anything else.
jicama
Freshman
Posts: 67
And1: 1
Joined: Mar 25, 2010

Re: Win Shares versus Wins Produced? 

Post#24 » by jicama » Tue May 11, 2010 9:21 pm

You'll get no satisfaction with WP, more with WinShares; and you could get eWins daily playoff updates, series by series, here:
http://sonicscentral.com/apbrmetrics/vi ... php?t=2570
jambalaya
Sixth Man
Posts: 1,668
And1: 286
Joined: Feb 01, 2005

Re: Win Shares versus Wins Produced? 

Post#25 » by jambalaya » Sun Oct 3, 2010 9:28 pm

Hoopdata is the rare site to include both PER and the simple "Win Score" side by side and it even goes the extra mile- for those interested- by showing "Alternate Win Score" (AWS) which is Win Score with weighting of rebounds the way Rosenbaum, PER and the results of some other regressions suggest they should be weighted.

Hoopdata also adjusts PER for some of its perceived flaws. It would be great if Hoopdata also listed WinShares and some version of Adjusted +/- and Statistical +/- and perhaps the the Floor Impact Counter measure created on this site and others too. But maybe that is hoping for too much. Probably will have to do much of the comparison manually, if so inclined.

One of the biggest differences between AWS and PER is the break-even point for shooting efficiency. PER's is considered too low or way too low by Wins Produced advocates. Berri's considered too high by PER advocates and others. I saw some regression analysis, done perhaps in support of Statistical +/-, where it found the break=even point was in-between PER's and WP's. I don't recall the exact figure but I think it was somewhere between 45-47% eFG%. If you adjusted AWS even further to that break-even point I'd use it over simple PER or simple WinScore or the first version of AWS.

At first look, there doesn't seem to be much difference to me with how Wins Produced and Win Shares (based off Dean Oliver's Offensive and Defensive Ratings) handle shooting efficiency.

I can't immediately tell exactly how Win Shares handles rebounding compared to Wins Produced and PER. I assume it is close to PER but am not 100% sure and not sure it is exactly the same for all players. Oliver’s formula for the rebound parts of the Offensive and Defensive Ratings used by WinShares are not simply comparable to the weights for defensive rebounding in PER and Win Score. If any wanted to sort that out in detail it might be a useful step forward in the discussion.

Wins Produced has a very powerful and helpful team adjustment. Win Shares also has one for player defense but unlike Wins Produced it is not exactly the same for every player. The Defensive rating used by WinShares adjusts the amount of defensive credit among team players to reflect the relative contributions of players to total team defense results from their outright defensive “stops” – blocks, steals and defensive rebounds. The only part that is assumed to be equal is shot defense for lack of reliable and agreed upon data. Wins Produced is weighting the blocks, steals and defensive rebounds directly in the Win Score for the individual player and then indirectly and imprecisely to all players in the team adjustment. Doing the Offense and Defensive ratings separately allows WinShares to avoid this double counting of the impact of stops and even if one of the counts by Wins Produced is spread evenly and doesn’t affect the relative ranking of players on a team it still appears to be a double-counting that would affect comparisons of players between teams. Wins Produced could avoid this by making the team adjustment just for the impact of shot defense.

PER never had a team adjustment and at this point it doesn't appear Hollinger will officially change PER in any way. 82 games has long offered counterpart PER though to try to cover counterpart defense and that is useful to some. But these are rough estimates and defense is certainly not all one on one, though to some it might be at least 50% that and checking that detail may be better and more helpful than not. Roland Ratings were a rough way to integrate both one on one and team level player impacts. They were a decent for a quick check of impact to some but they got no media buzz and probably light use by fans.

The position adjusted PAWS that Berri uses is useful to better compare the level of achievement of players across positions. I'd like to see something similar for PER or WinShares published though it is easy enough to get position or height sorted lists for these.

Adjusted +/- can help some at painting the picture of a player on top other metrics. Statistical +/- is useful too though it misses shot defense and applies its impact to other stats. No one metric by itself trumps all others to me. Using several, or blending them in some fashion, or building something new from what each does best and avoiding what they do worst seem like a better approach to me than just going by one. Using several helps me learn more about players and the metrics. Totally dismissing any metric is a choice but there can be some useful guidance with any if you are careful to recognize limits and bias.


(The Floor Impact Counter treats shooting efficiency in a way that is pretty close in impact to what I just suggested above. The rebounding weight for defensive rebounding is in-between the way PER and Wins Produced weights them. FIC gives more weight to offensive rebounding than either but in doing so is closer to what Statistical +/- does. Assists and blocks are more highly weighted than in other systems and than I would give them. Fouls are tricky but using a .5 weight is probably better than a full weighting. Even less weight or excluding them might be appropriate as guys who foul more frequently, often are producing better opponent efficiency results.)
jambalaya
Sixth Man
Posts: 1,668
And1: 286
Joined: Feb 01, 2005

Re: Win Shares versus Wins Produced? 

Post#26 » by jambalaya » Mon Oct 4, 2010 5:05 am

quote from "DavidStern":

"Berri ran regression and results showed how valuable rebounds are. No place to discussion here unless someone will provide other result done on bigger sample of data."


There are a number of regressions that have been done by various parties supporting a lower weight for defensive rebounds in an individual rating system . It is different, in their view, in an individual rating system than at team level.

Hollinger's credit in PER for defensive rebounds was based on his layman's run of a regression, not just his subjective view. Rosenbaum found the same thing and he has the formal training / PhD. Others with and without PhD's have agreed with this finding or "view".

If you are undecided and open-minded about the defensive rebounding weight, follow some of the links previously mentioned and read related articles and threads on other main NBA statistical forums and you can try to judge the merits of each side for yourself. But the research has been done and presented. It would be too long to post the full account here. The best trained have the best basis to judge the debate but everyone can try as they wish.

The debate is about defensive rebounding elasticity- how much you think team rebounding will vary with the addition or subtraction of one individual. That is what the regressions are for, but how you specify them matters. I am a little surprised that there is this fundamental disagreement among the well-trained, but I will also say that I do not recall seeing any other model (besides Berri's) where offensive and defensive rebounds were clearly treated as separate variables where the values came out nearly identical.

Part of it may come down to where you philosophically think a possession ends. Clearly there is a difference of opinion about what assumptions or understanding about the nature of team rebounding each want to use in their individual rating system. There will probably never be universal agreement on this or several other factors. Each metric can be somewhat helpful. Neither is perfect on everything or the end of advancement.

Return to Statistical Analysis