Referreed Journals

[EvanZ]

Would you guys take Millsap and Anderson over Aldridge? What about Young?

[mystic]

I wouldn't make a decision about that alone based on the boxscore data from this season. There are only played 17 or so games so far and some of the players had to deal with injuries. When I use the blended version of my SPM and RAPM (prior informed) I get:

Code: Select all

Name                Tm   OPM   DPM    PM
Dirk Nowitzki       DAL  2.97  1.75  4.72
Paul Millsap        UTA  2.46  2.04  4.50
Chris Bosh          MIA  2.07  1.84  3.91
LaMarcus Aldridge   POR  2.20  1.44  3.64
Josh Smith          ATL  0.80  2.33  3.13
Ryan Anderson       ORL  2.07  0.97  3.04
Kevin Love          MIN  2.58  0.44  3.01
Thaddeus Young      PHI  1.59  1.31  2.90
Kevin Garnett       BOS  0.91  1.63  2.53
Nick Collison       OKC  1.35  0.91  2.26
Al Harrington       DEN  1.53  0.59  2.11
Carlos Boozer       CHI  1.72  0.34  2.05
Blake Griffin       LAC  1.72  0.32  2.03
Elton Brand         PHI  0.53  1.46  1.99
Taj Gibson          CHI  0.22  1.55  1.77
David West          IND  0.78  0.74  1.51
Matt Bonner         SAS  1.48 -0.42  1.07
Lamar Odom          DAL  0.78  0.16  0.93
Luis Scola          HOU  0.52  0.39  0.91
Kris Humphries      NJN -0.26  0.57  0.31

I don't think that this is an unreasonable list and expecting everyone being healthy I would pick the PF in that order for my playoff run.

[mystic]

The divergance at least raises questions for me that might not come up if I relied on just one metric.

For sure. But what kind of questions? And is Berri even asking a question or is he drawing conclusions based on the differences?

Well, we have to take into account that scoring with the same efficiency is viewed differently in WP48, because it depends whether that scoring is coming from 2pt shots, 3pt shots or free throws. It also depends whether the player is a PF or a guard. Nowitzki's scoring for example is seen as being less valuable, because he is a PF and not a guard. Just imagine, if Nowitzki would be listed as SF (which is the position he would probably assigned, if his offensive game would only be looked at), he would have had a higher WP48 last season. As SF Nowitzki would be at 0.206 WP48 and with 10.74 WP, that would make him 27th respective 14th in the league (among all players with reasonable minutes above 480). Due to the positional adjustment he gets knocked down to 94th in WP48 (0.136) and 66th (7.1). Is it unreasonable to call Nowitzki's offensive game as a game of a SF rather than PF? Is Nowitzki's scoring really less valuable to a team, because he is a PF?

That Berri needs a positional adjustment is an issue related to the very basis of his metric. That this positional adjustment undermines his whole hypothesis at the beginning (calculating marginal values via league average and regression on win%) doesn't even come to his mind. So, how much stock should we put into such a metric, when the guy who built it doesn't even believe in the original hypothesis? When he doesn't believe in his marginal values?

[Mike G]

..That Berri needs a positional adjustment is an issue related to the very basis of his metric. ..

That about sums it up, doesn't it?

[EvanZ]

I don't think that this is an unreasonable list and expecting everyone being healthy I would pick the PF in that order for my playoff run.

So your metric is reasonably capturing the defensive component? Or do you just ignore that?

From what I've seen of Bosh and Millsap, I'd take Bosh any day of the week. I think he's shown recently what he can do when he is the 2nd option with Wade being out. I'd probably take LMA over Millsap as well. I'd also take Zach Randolph when healthy.

[xkonk]

Has anyone checked if other metrics find differences in production from different positions? I know Hollinger uses position adjustments for turning PER into points/wins added, and I didn't do anything fancy like weight by minutes but the average production in a couple other measures I looked at show differences as well.

[J.E.]

Has anyone checked if other metrics find differences in production from different positions? I know Hollinger uses position adjustments for turning PER into points/wins added, and I didn't do anything fancy like weight by minutes but the average production in a couple other measures I looked at show differences as well.

I think absolute defensive RAPM values are generally higher for forwards/centers than guards. I think that's fine. If you did a similar analysis for soccer/hockey I'm sure you'd find higher absolute values for the goalies. Why would we want to bring that rating down to everyone else's?

[mystic]

So your metric is reasonably capturing the defensive component? Or do you just ignore that?

Not quite sure what you are referring too, because that list isn't sorted by the defensive component. And to be quite honest, deviding SPM into offense and defense is more something to play around. SPM is first calculated, after that the split to defense and offense happens in accordance to team-related offense and defense.

From what I've seen of Bosh and Millsap, I'd take Bosh any day of the week.

Well, from what I've seen ...

I'd also take Zach Randolph when healthy.

Randolph is missing, because he didn't play enough minutes. I also have to admit that I copied and pasted the wrong columns, those who have the data from this season alone. Because of the sample size from this season, I wanted to show a blended version of last season and this season numbers. That's how it looks:

Code: Select all

Dirk Nowitzki       DAL    5.25
Chris Bosh          MIA    3.58
Paul Millsap        UTA    3.44
Kevin Garnett       BOS    3.37
LaMarcus Aldridge   POR    3.17
Ryan Anderson       ORL    2.87
Zach Randolph       MEM    2.70
Thaddeus Young      PHI    2.62
Kevin Love          MIN    2.61
Josh Smith          ATL    2.50
Lamar Odom          DAL    2.15
Blake Griffin       LAC    1.98
David West          IND    1.98
Carlos Boozer       CHI    1.92
Nick Collison       OKC    1.81
Luis Scola          HOU    1.59
Elton Brand         PHI    1.45
Taj Gibson          CHI    1.38
Al Harrington       DEN    1.21
Matt Bonner         SAS    1.10

I took out the offensive and defensive components, because it seems like that leads to confusions.

@xkonk

Why should we expect to have equal "production" at each position? How do you devide the positions anyway? We have players playing different position on the court on offense and defense, we have players playing multiple positions depending on the matchups, and so on. The positional adjustments brings only in another variable, which is hardly accurate to measure.
And it wouldn't surprise me at all, if nearly all metrics are ending up with the big players having a bigger "production" than small players.

[Mike G]

Has anyone checked if other metrics find differences in production from different positions? ... the average production in a couple other measures I looked at show differences as well.

I think absolute defensive RAPM values are generally higher for forwards/centers than guards. I think that's fine...

Production is mostly an offensive phenomenon, while defense is the suppression of the opponent's production.

As long as a metric insists on making a 'position adjustment' of some kind, it's admitting that it doesn't really measure player values. Otherwise, why not just play your 'best' players, even if it's 5 bigs on the floor?

Guards, like all positions, have value that isn't seen in boxscores -- bringing the ball up the floor, for starters. Bigs can't do that, yet the only boxscore 'reward' for the guard is the occasional turnover.

Sometimes the player, his teammates, and his coach wouldn't agree that he's a C or F, a F or G. If you have to hash that out before your metric can achieve a rating, you haven't got much of a metric.

[xkonk]

Let me run through what I understand to be the rationale behind having a position adjustment, and then someone can tear it apart.

At the base of the issue, I think, is the question of how good do I need to be to help my team. Do I have to be a 'good' player? Do I have to be above league average? The answer, really, is that I just have to be better than the guy I play against. I can be the ninth-best guy on the court, so long as the tenth-best is the guy on the other team I'm going against. Kind of a broad generalization, but it seems to be reasonable. Regardless of my team's quality or the quality of our opponents, if our guys are all better than the counterpart on the other team, we should win. Or, as long as we have a bigger advantage at some match-up than the other team has on a different match-up, we should win.

We know an average team should win half of its games and from that you can figure out that in terms of wins per 48 minutes an average player has a production of .1; that works for Win Shares too. But a basketball game is played by 5 guys who need to serve different roles; you need a ballhandler, people who can shoot accurately, people who can rebound, etc. So what you really mean is that five average people who fill those roles (or who are asked to fill those roles, at least) have a production of .1. And it doesn't matter if your ballhandler is, in some objective sense, worse than your rebounder or the other team's rebounder. He just needs to be better than the other team's ballhandler. If my ballhandler is above average for all the ballhandlers in the league, then he helps my team be above average.

So, when wins produced has a position adjustment, it's claiming that players are good or bad for their role. I agree that the positions may not be especially accurate given the diversity of player ability now and cross-matching on defense, but it's an attempt to account for the fact that some players are more similar to, and more likely to match up with, certain other players. If you'd prefer, you could have different labels and have an adjustment for 'slashers' and an adjustment for 'unathletic big men' or whatever, so long as it turned out that the wins added up properly. But there should be an inherent understanding that even though Spencer Hawes and James Harden have similar ratings (for example), they are not objectively equally productive. It's that Hawes' advantage over the types of players he generally plays against is the same size as Harden's advantage over the players he plays against, and the claim is that those advantages are equally valuable.

[Guy]

Wherefore art thou, Motherwell? Now that you've seen the data, are you concerned by Wins Produced's failure to predict future team wins? And if not, why not?

It should also concern you, IMO, that the WOW community has said nothing about Alex's findings. My guess is they will continue to simply ignore them, because they are profoundly inconvenient. Or maybe we'll hear that such findings should be ignored until they appear in a refereed journal.

Also interested in your thoughts, xkonk.

[xkonk]

Also interested in your thoughts, xkonk.

Assuming the sport skeptic analyses hold up, it does seem like WP is not the best option for predicting performance. But that isn't really what I'm interested in; I'm interested in finding out which parts of the Wages of Wins enterprise people disagree with and why, the same way I'm interested in the pros and cons of APM and RAPM and all the rest. It's hard to tell sometimes if a comment means 'I have an issue with position adjustments', for example, versus 'I have an issue with Berri'. I'm familiar with the defensive rebounds issue, since that's been all over the place, but if anyone has thoughts on my last comment I'd like to read them.

[EvanZ]

Also interested in your thoughts, xkonk.

Assuming the sport skeptic analyses hold up, it does seem like WP is not the best option for predicting performance. But that isn't really what I'm interested in; I'm interested in finding out which parts of the Wages of Wins enterprise people disagree with and why, the same way I'm interested in the pros and cons of APM and RAPM and all the rest. It's hard to tell sometimes if a comment means 'I have an issue with position adjustments', for example, versus 'I have an issue with Berri'. I'm familiar with the defensive rebounds issue, since that's been all over the place, but if anyone has thoughts on my last comment I'd like to read them.

Yes, that. The "enterprise" of it. Honestly, at this point, the metric itself is not as repugnant as the enterprise behind it.

[Guy]

Assuming the sport skeptic analyses hold up, it does seem like WP is not the best option for predicting performance. But that isn't really what I'm interested in; I'm interested in finding out which parts of the Wages of Wins enterprise people disagree with and why

The two questions are closely related. If WP can't predict well, then by definition it isn't measuring what it purports to measure.

if anyone has thoughts on my last comment I'd like to read them.

This is a long discussion. But my quick take is that the problem is not the position adjustment per se (which improves WP). It's that the need for a position adjustment tells us there's a problem with the underlying productivity metric (adjP48). The metric says that guards are only half as productive as big men. Berri then says "well, teams need both big men and guards, so I'll make them equal in value to the big men." But why does a team "need" a SG, or a PG, or any particular mix of positions? It doesn't say that in the rules. So if it's true, it must be because a SG is delivering some kind of essential value to the team that isn't measures, or is mismeasured, by adjP48. How else can one justify saying that .400 adjP48(PF) = .200 adjP48(SG)? [There is also no logical connection between Berri's claim that teams "need" different types of players, and his conclusion that all positions are equal in value.]

When Berri initially got his productivity results, he had two reasonable choices: 1) believe in his metric, and say that big men are twice as valuable as guards, or 2) believe that positions are equal, and go back to the drawing board to figure out why his productivity measure claimed otherwise. Instead he took door #3: just define the positions as equal, with no empirical or theoretical justification, and then keep using the metric as is. But this creates a basic mathematial fallacy. Berri tells us that .400 adjP48(PF) = .200 adjP48 (SG), but also that .600 adjP48(PF) = .400 adjP48 (SG). And that does not compute....

[EvanZ]

The position adjustment is basically a kludge to smooth over the rebounding problem.