Wins Produced?!

[mystic]

It seems to be a prevalent opinion that NBA coaches know what they're doing (although that may or may not be true). So if the coaches asked Dantley to shoot at the rim instead of Laimbeer, we would assume that Dantley is better at it.

Or, that Dantley isn't a good outside shooter, thus he wouldn't provide the necessary spacing for the team. I think, what you really don't understand is how the game works. Somehow your answer makes no sense, because you are trying to imply that two players could just occupy the same spot without any kind of consequences. If Laimbeer gets the space inside, because the defense is respecting Dantley's outside shot, he gets easier opportunities inside. Now, gets Dantley punished for his lack of outside shooting? No, he even gets a bonus, because usually SF don't have such a high fg%.

It wouldn't make sense to mark Dantley down for lack of interior defense unless he played without a center and was thus asked to guard the paint.

So, Dantley wasn't able to guard inside players, but occupied the room inside on offense, wouldn't you say that this is a huge problem, because we can't really assign a position to him? We get a similar problem with Magic Johnson, who actually guarded small forwards, played a lot of times in the post, and yet, he is considered a point guard. In reality he is more a point forward than LeBron James.
And in the end Dantley gets the credit for the team defense as well, even though he had nothing really to do with that. WP is shifting credit around, taking it away from the teammates and putting it on to Dantley. And while all boxscore metrics missing a part, WP is making it even worse with their positional adjustment.

But in the eyes of WP, not shooting as well as an average center is bad for centers. So if Laimbeer shoots from all over the court and lowers his efficiency, that would be a knock on him. If you think that being a proficient shooter from any location has value, then obviously no boxscore measure is going to make you happy because none that I know of make any distinctions between 2 point shots.

Honestly, you are either completely not understanding the point, or you are intellectual dishonest. No, no other boxscore metric is punishing Laimbeer for not shooting as well as other centers, only WP is doing that, because only WP is actually so flawed from the beginning that huge discrepancies between centers and guards are there before the positional adjustment. Somehow, no other metric needs such adjustment.

Speaking of Dantley, every boxscore measure I can find including Daniel's ASPM agrees that he was far above average as a player. Seems like such a consensus should mean something.

It means that the boxscore is missing something. If Dantley would have been so great, why did the Pistons trade him for a player who wanted out AND the Pistons actually needed to attach a first round pick in order to get a deal done? The Pistons after going to the finals are trading their leading scorer, who is seen as such great efficient player, for a rather unefficient volume scorer like Mark Aguirre. So, there was obviously something wrong here, either the Pistons were total morons or the boxscore metrics are wrong in their assessment that Dantley is a great player.
Now, we can test that by looking at the performance level of the Pistons prior and after the trade. We have 45 games before the trade and 37 games after the trade plus 17 playoff games. If we look at the scoring margin, the average opponents SRS and HCA, we see a team which played +4.6 in 45 games before the trade, and +9.1 in the 54 after. We replace a great efficient scorer, Dantley was making 19 ppg on 61 TS%, with a player who then scores 15.5 ppg on 55 TS% and the team is getting better? WP would have predicted a debacle for the Pistons, but yet, they went on not only to win the title in 1989, but also in 1990. The explanation by Dave Berri? Well, Rodman got more minutes, that's why they improved. Oh well, Rodman got indeed more minutes after the trade, he went from in 26.1 minutes per game in the first 45 games to 26.7 minutes per game in the last 54 games. That's the kind of explanations someone needs who has no clue what he is even talking about.

The explanation is rather simple. Dantley's skillset didn't put the other Pistons into their best positions, they were forced to be in worse position in order to feature Dantley. Once Dantley was removed and a better fitting Aguirre, who wasn't stopping the ball, was able to hit some outside shots, was in, ALL other Piston players were shifted into a better position on the court, making the overall team play better basketball. And, the ball was not as much occupied by a single player during a possession, leading to more ball movement and consequently to better team play.

Dantley was traded, because Chuck Daly wanted him out, the Pistons were even more desperate to get rid of Dantley than the Mavericks to get rid of Aguirre (that's why the Pistons send the 1st round pick and not the Mavericks). At the time the trade occured, the Mavericks were 25-21, they went 13-23 after the trade for Dantley.

[DSMok1]

Brilliant post, Mystic.

For what it's worth: ASPM had Dantley in 1989 at +1.57 for DET and then -1.46 for DAL; Aguirre at -0.47 for DAL and +0.00 for Detroit.

As a frame of reference, in terms of explaining what happened on the court:

ASPM explains ~60% of what happened, ~70% of offense and ~45% of defense. ( http://godismyjudgeok.com/DStats/aspm-and-vorp/ )

So there's obviously a lot left over that the boxscore can't pick up.

Wins produced explains ~38% of what happened, 21% of offense and ~14% of defense.

[Mike G]

ASPM explains ~60% of what happened, ~70% of offense and ~45% of defense. ...
So there's obviously a lot left over that the boxscore can't pick up.

Wins produced explains ~38% of what happened, 21% of offense and ~14% of defense.

Should that be "~18% of what happened"?

Brilliant post,..

... what you really don't understand is how the game works. Somehow your answer makes no sense, ...
... you are either completely not understanding the point, or you are intellectual dishonest....

A "brilliant post" generally doesn't attack others' interpretations. One could write, for example -- "what we don't fully understand...", "to make better sense of this...", "an alternative point to be made...", "to be intellectually most honest..." -- in place of these starkly belittling comments.

For me, it's hard to be objective about the rest of the post, when these grenades are tossed into the discussion.

[DSMok1]

Good point, Mike. I had sort of skimmed over the questionable comments.

[mtamada]

Yup good points, but the Dantley observations are right on. There's no question that Dantley was a well above average player, a legitimate all-star. But his box score stats make him look like not just an all-star, but an all-pro and a potential Hall of Famer. He might indeed deserve his Hall of Fame status when you include his college career, but he was not as good an NBA player as his stats, all-pro votes, Hall of Fame plaque, etc. imply.

In addition to the Detroit example (where Dantley's attitude was at least as big a factor in Detroit dumping him as his lack of team fit), there's the Laker example: despite having Kareem, the Lakers front line got manhandled two years in a row by the Sonics. Which was not surprising given that the Lakers were trying to advance in the playoffs with Dantley and Jamaal Wilkes at forward, which meant that one of those players had to try to handle the Sonic power forwards Sikma (he was the Sonics starting PF in 1978), Shelton, and Silas. To put it mildly, that did not go well for the Lakers.

It's no accident that the forward the Lakers chose to get rid of was Dantley, whose game did not mesh with Kareem's. Whereas Wilkes provided better all-around play including an outside game. He could even contribute to Showtime better than Dantley could; though not a speedburner nor finisher in the way that say Worthy and Byron Scott were, he would fill the wing on the break and if the fastbreak didn't result in a dunk or easy layin, he'd be open for what Chick Hearn liked to call the "17-foot layin".

Dantley was a very good player who could help take a crummy team such as the Jazz and make them a decent team. But he wasn't a great player, not as good as his gaudy stats imply. If you wanted to win a championship, it was very hard to create a roster that would work well with Dantley. Easier to do what the Lakers and Pistons did, and simply get rid of him in order to win a championship, while using seemingly inferior players such as Mark Aguirre and Kurt Rambis instead. Individually, those two WERE inferior to Dantley. But teamwise, they were easier to create championship rosters with than Dantley was.

[xkonk]

Let me try going back to the original comments to see if it makes more sense that way.

Italian Stallion's concern was that WP punishes small forwards/players with the ability to shoot from anywhere compared to centers/players who only shoot from near the basket because shots from further away must be taken but come with lower efficiency.

I simply stated that with WP's position adjustment, it's possible that both of those players would be equally rated because small forwards are expected to shoot more poorly (because, on average, they are asked to shoot from more places on the court).

Mtamada countered that some centers may be able to shoot from more places, and are thus underestimated by WP.

I then made a misguided attempt to argue that centers who don't take as many shots from near the rim as the average center do so because of team demands. Either they aren't good at shooting near the rim (otherwise they would do so like an average center) or someone else on the team is better at it than they are. In either case it seems fair to me that a metric might dock the center for not being a better close-range shooter. But that was a poor digression. Let's try something else.

Going back to Stallion's Player X and Player Y, let me make one small change which is just to give them the same total number of shots. We could even go so far as to make all of their boxscore stats the same; they are identical except that player Y has replaced some number of close-range 65% accuracy shots with further away two-pointers at a lower accuracy. This changes nothing about Stallion's concern but makes the next part easier.

Is there a metric which would claim Player Y is equally/more valuable? Probably not WP, especially if X and Y were given the same position adjustment. But definitely not PER, because all it sees are two otherwise identical players, but one has a lower shooting percentage. I don't see how Win Shares could find otherwise. Even SPM or ASPM methods would find Player Y to be inferior. That's because all are based on the boxscore and none know that Player Y shoots from more places on the court, or that he shoots them more accurately than most players. All the boxscore sees is that Player X took N two-point shots at 65% and Player Y took N two-point shots at 50%, or whatever.

Thus you don't have a WP problem, you have a boxscore problem. And, even though few people seem to like the position adjustment, it seems to be the only boxscore metric that has any way to account for the problem. Yes, it will overrate players with limited offenses who only shoot near the rim and underrate players (particularly centers) who shoot from other locations. But given that most centers shoot well because they stay near the rim, and most small forwards shoot less well because they shoot from other places, the position adjustment will bring centers and small forwards more in line by assuming centers will shoot better. Obviously an imperfect solution, but more of a solution than some other metrics.

[mystic]

Obviously an imperfect solution, but more of a solution than some other metrics.

Which other metric does need a positional adjustment, because it is so heavily biased towards certain positions as WP?

Good point, Mike. I had sort of skimmed over the questionable comments.

Don't worry, I didn't see your comment as a homage to my writing skills and my political correctness, I'm sure you just referred to the basketball-related content.

[Guy]

I simply stated that with WP's position adjustment, it's possible that both of those players would be equally rated because small forwards are expected to shoot more poorly

The position adjustment has very little to do with shooting efficiency. The vast majority of the position adjustment represents the difference in rebounds by position -- remove rebounds and "productivity" by position is close to even. (This is the reason Berri had to add a position adjustment: if he didn't, his metric would value big men so much more than guards that no one would take it seriously at all.) Differences in turnovers and assists (for PG) make up the rest.

Thus you don't have a WP problem, you have a boxscore problem.

Actually, you have both. The boxscore data is limited, as you observe. But WP makes the problem worse by making two related assumptions: 1) usage has no value , and 2) 50% shooting has zero value. Note that in WP, a player who who takes two 2P shots per 48 min. and makes one of them has the same value as one who takes 30 shots and makes 15. Similarly, a player who takes 20 shots and makes 9 is less valuable than a player who does not shoot at all! Because the players who are tasked with taking lower-percentage shot tend (not always, of course) to be higher usage players, WP penalizes them much more than is required by the limitations of boxscore data (and the reverse for low usage players).

That's the kind of explanation someone needs who has no clue what he is even talking about.

One can quarrel with Mystic's diplomatic skills, but he identifies the critical problem with Berri: professional basketball is simply a much more complex game than Berri realizes. To Berri, it's simple: put the ball in the hoop, grab rebounds, don't turn the ball over. Do those things, and you are valuable; fail to do them, you are not. In reality, the game is much more complex, with varying assignments for players based on their skillsets, and multiple interaction effects among teammates. NBA basketball is chess, while WP is modeling a game of checkers. Maybe his next book will be Berri on Chess:

In the early stages of championship chess matches, we find that fully 62% of the moves involve moving a pawn, and an overwhelming 74% of pieces captured in this pivotal period are pawns. However, our regression analysis reveals that pawns are not actually high-value pieces, but in fact are the least valuable piece on the board. These so-called "grand masters" are far too preoccupied with these small, inefficient pieces, while neglecting the much more valuable "big men" (not to mention the Big Lady) who are the true game changers in chess.

[xkonk]

PER has a position adjustment for its 'value added' (and thus 'estimated wins added') features. Not that it's any kind of support for having an adjustment, just answering the question.

[EvanZ]

Mark Cuban commenting at WoW:

http://wagesofwins.com/2012/09/03/attac ... lind-spot/

[mystic]

Mark Cuban commenting at WoW:

http://wagesofwins.com/2012/09/03/attac ... lind-spot/

Yeah, seems like the Mavericks are making their decisions upon WP48.

I couldn't have said anything of that better, but Cuban is actually making my point, the stuff I basically preach the last couple of years also on RealGM. It is a team game, and how a 5on5 unit is working against a 5on5 unit matters, and when a player is the key to the success of that unit, he might as well be really valuable, especially when we see in Nowitzki's case that he is the member of basically all the best lineups the Mavericks have while the teammates around him change. Nowitzki is a +/- star for a reason, and if a boxscore metric fails to recognize such things, there is something wrong with the boxscore metric, not with the success of the lineups with Nowitzki. In 2011 the Mavericks with Nowitzki and without Chandler were at +9.5 (per 48), while they were -1.7 with Chandler and without Nowitzki. And then you look at WP48 and that metric tells you that Chandler is twice as valuable as Nowitzki per minute? Something doesn't add up, and it was explained multiple times what is wrong with WP48. Maybe Cuban opens their eyes ...

Also, funny how Cuban says that they should use the WP48 for lineups. When I checked the performance of the different metrics to predict lineup performances, WP48 was as bad as PER, both way worse than my SPM approach or Win Shares. That can be seen as another proof that WP48 is not evaluating players at all.

Actually, you have both. The boxscore data is limited, as you observe. But WP makes the problem worse by making two related assumptions: 1) usage has no value , and 2) 50% shooting has zero value.

That is not really an assumption they made, that is just based on the flawed formula they used. In order to get the efficiency, they used team related formulas to calculate possessions. When you use such formulas, they will dicate the treshold for the shooting. Also, because both teams have roughly the same possessions and they obviously have a combined 100% usage on each team, that factor gets eliminated. As I pointed out some years ago, using team-related formulas is the problem here. That usage has no value and shooting 2pt shots only adds value with above 50% is a direct result of the used equations.

[Guy]

When I checked the performance of the different metrics to predict lineup performances, WP48 was as bad as PER, both way worse than my SPM approach or Win Shares.

Mystic, could you post a link to this research?

It's unfortunate that Ari Caroline, in the linked WOW thread, repeats the canard that WP predicts wins better than other metrics. Presumably, he is not familiar with the research showing that, although WP does indeed "predict" the past very well, it does a poor job of predicting future (out-of-sample) wins....

[mystic]

Mystic, could you post a link to this research?

I wouldn't call it research, but here is a link to an earlier version: http://bbmetrics.wordpress.com/reality_check/

PRA is the predecessor of my current SPM. In much ways the same.

I collected the data from 2007 to 2011 and found the following correlation coefficients:

WP48 to NetPts per 100 possessions: 0.568
WS/48 to NetPts per 100 possessions: 0.674
PER to NetPts per 100 possessions: 0.546
SPM to NetPts per 100 possessions: 0.667

It's unfortunate that Ari Caroline, in the linked WOW thread, repeats the canard that WP predicts wins better than other metrics. Presumably, he is not familiar with the research showing that, although WP does indeed "predict" the past very well, it does a poor job of predicting future (out-of-sample) wins....

That always reminds of that FM48 metric I created for the RealGM crowd. Just took the scoring per 100 poss for each player and added a defensive adjustment. The metric completely fullfilled the requirements Berri set for WP48 (explaining in hindsight, year-to-year correlation, etc.). That those people are ignorant to the fact that most players will just be used in the same fashion each year and that the high correlation between wins and WP has nothing to do with the ability to evaluate players, is really the sad thing here.

[DSMok1]

Also, my research is similar, Guy:

"I compared some of the other stats to the same 8 year-average RAPM data set I mentioned above, using weighted correlation. For ASPM, this is in sample, so the results WILL BE INFLATED.

Code: Select all

                 R^2 Values
	              to RAPM 	to O RAPM 	to D RAPM
PER 	          0.458 	   0.554 	    0.017
Win Shares/48 	0.531 		
O Win Shares/48 		       0.598 	
D Win Shares/48 			                 0.450
Wins Produced    0.380 	   0.217 	    0.142
ASPM 	         0.627 		
O ASPM 		                0.735 	
D ASPM 			                          0.492

"

http://godismyjudgeok.com/DStats/aspm-and-vorp/

[Guy]

I collected the data from 2007 to 2011 and found the following correlation coefficients:
WP48 to NetPts per 100 possessions: 0.568
WS/48 to NetPts per 100 possessions: 0.674
PER to NetPts per 100 possessions: 0.546
SPM to NetPts per 100 possessions: 0.667

Interesting. Did you use same season data for the metrics (i.e. are players' 2007 WP being used to project the performance of a 2007 lineup)? I suspect the spread in your correlations would be even higher -- to WP's disadvantage -- had you used players' prior-season ratings on the metrics.

I always thought that comparing WP to performance of specific lineups would be a good way to document WP's shortcomings. Nice to see someone has done it. Of course, the WOW crowd would just object that you didn't control for quality of opposition, home/away, humidity, etc. etc. and then dismiss the results -- ignoring the fact that all the other correlations had the same limitations. As you probably know, WP also does a relatively poor job of predicting next-season team wins ("Sport Skeptic" looked at this a year or so ago).