APBRmetrics

deepak · Joined: 26 Apr 2006 Posts: 664

http://www.nytimes.com/2006/06/10/opinion/10berri.html?_r=1&oref=slogin

Thoughts?

Also, could someone who's knowledgable about his methods also explain how/why his results differ from individual Net Wins via Offensive and Defensive Ratings. Are they trying to answer the same question, or are they actually describing two different concepts?

Thanks.

**Dan** **Rosenbaum** · Posted: Sat Jun 10, 2006 11:16 pm Post subject:

[Let me rewrite some of this to add clarity.]

I have not yet picked up the book, but their "simple" formula is the following.

Win Score = PTS + REB + STL - FGA - TO + 0.5*(AST + BLK - FTA - PF)

They also have a simpler formula that they use in several of their academic papers.

Old Win Score = PTS + REB + STL - FGA - TO - 0.5*FTA

As far as I can tell, all of the stats are per-minute (or per-48 minute or per-40 minute) measures. I don't know if they are pace-adjusted.

But Win Score is a simplified version of Wins Produced. And here is the difference between the two.

http://dberri.wordpress.com/2006/05/21/simple-models-of-player-performance

edijorj · Joined: 21 Apr 2006 Posts: 30

Dan Rosenbaum,

I havn't read the book either (and I probably won't), but I think Berri1 is what they call Win Score, not Wins Produced. Win Score is supposedly similar to Wins Produced if player position is accounted for. (Berri apparently realizes that his metric is biased against guards.) So it would make more sense if you added some dummy variables for player position.

It seems like the problem with Berri's analysis is that he believes that his metric is good if it explain team wins. To give an extreme example, the equation Y=1.78*FG-1.49*FGA+1.78*Th-0.57*ThA+0.51*FT-0.41*FTA+1.46*Orb+1.41*Drb+0.06*ast+1.76*stl+0.13*blk-1.52*to+0.13*pf explains team wins very well, but is clearly wrong for evaluating a player.

deepak · Joined: 26 Apr 2006 Posts: 664

Kevin Pelton · Posted: Sun Jun 11, 2006 11:48 am Post subject:

Mark · Joined: 20 Aug 2005 Posts: 807

We can't compute error on individual player level but it seems like there could be higher levels of player error that cancel each other out. Error reported at team level really just measures error for formula for combined team level stats doesnt it? Several other make this same case in this thread http://gladwell.typepad.com/gladwellcom/2006/05/basketball_by_t.html
This would apply to the simple version of Berri, would it to the regression-based version too?

I noticed that 3 of the 10 most underrated were Memphis Grizzlies. http://gladwell.typepad.com/gladwellcom/2006/05/basketball_by_t.html
Only one brought in by Jerry West but the presence of the other two may have been some attraction to him and he did give them 6 year extensions. But they can't get out of the first round even with Gasol . Even win a playoff game over last three years. I thought the bench was supposed to have been a strength. Is it the coaching? Is it the level of "clutchness"? (Dan gave clutch time extra weight and drew an apparent criticism for doing so from edijori earlier viewtopic.php?p=8403&highlight=#8403, I tend to support a clutchtime weight but not sure how much. I mention in case either or others wish to expand of whether clutch weights should or should not be in best practice methods.) Just the toughness of the opponents, not getting a higher playoff seed? But why not higher seeding or better playoff performance? Is it that they are strong on other things that make them appear underrated but not high enough as a group on the "overrated" category of scoring?

I wonder how different the ratings would be if the weights were selected based on team playoff performance instead of regular season.

This year's Knicks had for a time 4 of the bottom 10 "difference makers" on this 2004 list from Dan http://www.82games.com/comm29.htm

Single examples don't validate or demonish a formula, I just noticed these cases. I believe the two differnet systems produce ratings for Kobe Bryant and Ray Allen that are close to each other. But that is surface level result similarity only and folks will write in different explanations of it.

deepak · Joined: 26 Apr 2006 Posts: 664

More explanation on Win Score and Wins Produced

Win Score is a simple, approximate formula that can be used to quickly assess how good a game a player had. I supposed it's analogous to John Hollinger's game scores -- though even simpler to calculate.

Also, I wonder what their rationale is in adjusting for positions, beyond making their list more credible.

Here are the top 20 players, in terms of Win Score per 48 minutes:

Mark · Joined: 20 Aug 2005 Posts: 807

18 big men and 2 perimeter by this method? I'm all for giving full deserved credit for big men but is that the main discovery?

bchaikin · Joined: 27 Jan 2005 Posts: 685 Location: cleveland, ohio

I love Chuck Hayes, but number 5?

what chuck hayes produced statistically in just 535 minutes in 05-06 was excellent. i can easily see why a rating system (not just this one) would have him rated very high....

just looking at his numbers (and not noting that he may have accrued them solely against other teams' reserves and not their starters, which may or may not be true but if true could make a world of difference), if you multiplied his stats by 5 so instead of 535 minutes he played 2675 minutes he'd have scored 18.5 pts/g on a ScFG% of over 56% with 11 rebs/g playing 33 min/g - with 130 steals, 70 bs, and just 60 turnovers (uh... and also 380 fouls committed)....

would he do this if giving the playing time? who knows? but his stats in those 500+ minutes were outrageously good, kind of like anderson varejao in 04-05...

**HoopStudies** · Posted: Sun Jun 11, 2006 4:23 pm Post subject:

Mark · Joined: 20 Aug 2005 Posts: 807

Mark · Joined: 20 Aug 2005 Posts: 807

edijorj, what is the source of the linear equation in your last post? How good is the fit?

**HoopStudies** · Posted: Sun Jun 11, 2006 6:42 pm Post subject:

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

Mark · Joined: 20 Aug 2005 Posts: 807

HoopStudies wrote:
"Apparently, the concept of assigning credit based on difficulty of accomplishing a portion of a task is Marxian."

I'd have to hear more of that explanation to understand what was meant. Maybe you could say that at a level of technical Marxian economic analysis, of value accounting- stressing the importance of the labor input.

But at the simplest philosophical level communism is "from each according to their abilities, to each according to their needs" and that seems contrary to the practice in the quote. I would think a classical political Marxist would likely have a VOP attitude of equal worth of different activities that produce equivalent results to me. And would probably be against moneyball and a skewed freemarket bidding war distribution of salary heavily to the top performers. The quote sounds like it has more in common with a capitalist attitude than a political Marxist one to me.

But I guess economic Marxian economic analysis can be considered somewhat distinct from that and it was initially applied to understanding capitalism, how the system actually works according to the existing rules and market values of different inputs in producing outputs-wins.

Berri is giving all rebounds defensive and offensive equal weight (at least in the simple version)? And you dont view the world that way. Is that part of what makes you Marxian to him? Hmmm... Not sure the label is particularly useful if that is the issue, though this debate is important. I guess if difficult, perhaps especially physical labor (like off. rebounding, defense, blocked shots) was seen as deserving more credit than shooters (and in some systems it is) that might perhaps be called Maoist Gang of 4 peasants vs the elite rhetoric.

Team +/- ratings for players could certainly seem pure communist but for the opposite rationale than the one asserted in the quote. Everybody gets the same score for that time on court together! Does he pull away from that by just using individual boxscore data, not give stress to contextual data, opposed that line of thinking? Extra weight for crunch time is that opposed too?

The general idea of value being added by high level managers, traders, consultants thru their intellectual capital contributions still seems patently capitalist, though I guess you could call it central planning and get away with the same thing as in historical communist state practice. The workers/players continue to be steered to some degree. But compared to football, basketball is a player's game.