Matthew Yglesias (July 10, 2007) - Rashard Lewis Revisited (Sports)

Imagine the plight of the Indiana Pacers coach. Everyone sucks, any position you play them at is not their optimum position, and using them in a second position results in it raining toads and blood from the sky.

Posted by steve duncan | July 10, 2007 1:36 PM

It's true that teams employing Lewis do well to remember that he's much more effective as a small forward than as a power forward (assuming that's true) but it's also true that it's better -- more useful to your coach and GM -- to be able to "play out of position" with a modicum of success than to be totally useless.

I think you're misunderstanding each other. The point is that Lewis would produce less rebounds than you would expect from a replacement player in the power forward spot. In this analysis, Lewis has more value as a 3 than a 4 not because of an inability to play the 4 but because his statistical contributions compare favorably with other 3s, but not with other 4s. The question is whether Lewis can match benchmarks set by the other players who play his positions.

It's a bizarre way of looking at it, of course. It says more about the talent distribution among positions in the NBA than anything else. It's a system where a players value is entirely dependent on statistical output relative to replacement players at his position, rather than to a independent concept of what a power forward should provide to your basketball team.

Posted by Freddie | July 10, 2007 1:53 PM

a independent concept of what a power forward should provide to your basketball team

which depends on the team and the system and the coach, of course. While I've felt for a long time that the traditional monikers (point guard, shooting guard, etc.) are poor categories for catching the different kinds of players there are, "player types" also don't really correspond to the "number" system except vaguely. There's this weird tautological thing where the most meaningful definition of a "4" is that he can defend other "4"s. I think Berri's missing Matt's point more than Matt's missing Berri's, though. Or, to be as I generally am uncharitable towards WoW, Berri looks to me like he's arguing within the contraints of his method, such that his concept of a player's value is identical with what of that value is captured by his system.

Posted by Quarterican | July 10, 2007 2:01 PM

We would all do well to remember that Berri knows next to nothing about actual basketball. He has some familiarity with basketball statistics, but a little knowledge is a dangerous thing.

Posted by Pooh | July 10, 2007 2:18 PM

I think this amount of analysis of Rashard Lewis is odd. He's a very good but not great scorer who rebounds poorly for his size and plays no defense. If you put Kobe and KG at the top of the pyramid and guys like Paul Pierce and Tracy McGrady in the level below, Rashard clearly fits into the level below that at best.

Adding Lewis to the Magic will make them a better team, but not that much better. Two years from now when they need to make another move to improve the team and are hamstrung by Lewis' stupid deal, the Magic will learn not to overpay for the best available player during free agency.

Mike

Posted by MBunge | July 10, 2007 2:33 PM

Rewis puts up Lewis' exact same numbers, but if his coach tries to insert him as a power forward he refuses to play. Which player is better to sign -- Lewis or Rewis? Berri says it's Rewis -- Rewis will have a better position-adjusted Wins Produced number. I say -- and basic common sense says -- it's Lewis.

More to the point, say Rewis is so unbelievably awful at the 4 that his coach would never even think of asking him to play there. According to Berri, that would make him better than Lewis as well.

Posted by JP | July 10, 2007 2:37 PM

I pointed this out on Matthew's original Lewis thread, but I'll repeat it here: Matthew and Berri are talking about two different things. Matthew is talking about a player's inherent ability as a basketball player. Berri, however, is talking about the player's actual output on the floor. These two things don't have to match each other. The player's output on the floor is significantly affected by a factor besides his inherent ability: the coach. And if the coach puts the player in a position in which he will not play up to his ability, they player will perform worse on the floor, no matter how good a player he is. To take an extreme example: if the coach plays Duncan at PG, he is going to perform poorly. He can dribble and pass, but he won't be a good PG. Now, Matthew comes along and says: look, this shows that Duncan is even more valuable than we thought - he can even play PG, even if poorly. But Duncan's actual performance will suffer, and Berri will say he's a worse player because, well, he didn't play as well. They are both right, and the only thing this shows is that the coach is most likely an idiot for playing Duncan at the PG (or the GM is an idiot for not having a better PG on the roster).

Posted by Al | July 10, 2007 2:38 PM

Every time you write "Lewis" I think you're talking about Lewis Scooter Libby. I'm confused. Perhaps you need a second blog to talk basketball?

Posted by Jim | July 10, 2007 2:39 PM

or a second blog to talk politics.

Posted by too many steves | July 10, 2007 2:43 PM

MBunge -

That depends - I'm not going to do the research right now, but what top flight free agent wings are due to come on the market in the next few years? There was a lot of speculation that Carter would be wooed by the Magic, but he stuck in NJ. Personally, I was convinced that at least *one* of the premiere wings from the 2003 draft (LeBron, Wade, Carmelo) would opt not to sign lucrative extensions in order to at least test the free agent waters when Orlando was positioned to make a big move, given that Orlando's budding nucleus at the time looked a lot more promising to me than the non-Bron/Wade/Melo nuclei of their respective teams. But I was wrong about that.

Posted by Quarterican | July 10, 2007 2:43 PM

Matthew is talking about a player's inherent ability as a basketball player. Berri, however, is talking about the player's actual output on the floor. These two things don't have to match each other.

True, but it's the former, not the latter, you should be interested in when signing free agents, no? And what good is fancy statistical analysis if it doesn't help you Moneyball the league, that is, beat the free agent market?

In this analysis, Lewis has more value as a 3 than a 4 not because of an inability to play the 4 but because his statistical contributions compare favorably with other 3s, but not with other 4s. The question is whether Lewis can match benchmarks set by the other players who play his positions.

Similarly, why do we need statistical analysis for this? So you should sign Lewis if you need a 3 but not if you have a good 3 and a hole at 4. Well, duh. And not really a useful thing to know, anyway. What you want to know is how valuable is Lewis compared to all other free agents I could be spending money on, and how much more or less valuable--at all positions. You have X dollars to spend on your whole team, not on each position.

Posted by Antid Oto | July 10, 2007 2:43 PM

I'm a little more amenable to what Berri is saying than it seems most people here are. But I do think the whole "value over replacement player" system of metrics create some really weird ways of looking at things. For instance, Berri believes that Lewis becomes a worse player when playing the 4 than the 3, not because he would play the position worse, but because his stats would compare less favorably to others at his position. So suppose that, when calculating the averages of the average power forward, instead of considering Tim Duncan a 4, you consider him a 5. And instead of considering Dirk Nowitzki a 4, you consider him a 3; and consider Garnett a 5, and Zach Randolph a 3. All of a sudden, the average numbers for a power forward in the NBA drop, and Lewis's numbers improve relative to the average. So without Rashard Lewis's game changing at all, he's become a more valuable player. Which is kind of strange.

Posted by Freddie | July 10, 2007 2:52 PM

I don't think you are looking at it correctly, Antid Oto. You are not signing the player based on his inherent talent as a basketball player (which is how Matthew is looking at it), you are signing him for his production on the floor. Again taking my absurd exaggeration to show the point: I would not sign Duncan to a max deal to play PG for my team, because his production there will be low, no matter how talented he is. Similarly, Lewis is not a max player if you play him at the PF position. And Berri makes a compelling case that Orlando will play Lewis at the PF a lot. The trick is to have the talent play at the position where they will be most productive. If ORL finds a way to play Lewis at the SF exclusively, then it is a good deal. But Lewis's ability to play the PF mediocrely is not much of an asset, since all the time he plays at PF reduces the time he would otherwise play at SF, where he is effective.

Posted by Al | July 10, 2007 3:16 PM

The point is that Lewis would produce less rebounds than you would expect from a replacement player in the power forward spot... The question is whether Lewis can match benchmarks set by the other players who play his positions.

I haven't looked at the actual numbers, but Berri argues that Lewis is an "average" power forward. If he is better than half of the players at that position (I know this isn't the technical definition of average, but it's probably a close enough estimation) there are many teams that would be happy to have him as their backup power forward. And even if he is only better than a few true power forwards, this still makes him more valuable than if he couldn't play power forward at all. If you lose all your regular power forwards to injury, it's probably better to have Lewis step in than someone who can't compete at the position at all. Not necessarily since you would be weakening two spots (3 and 4) rather than just one, but the choice would be there.

Where I think the communication problem may be (although I'm not going to go back and read everthing to make sure), is that Matt says playing two positions competently is better than one, even if you are only average at the second one, which is obviously true; Berri is saying (at least in the current post), that even though (a) Lewis is an excellent player, and (b) he can play power forward, it doesn't follow that he should get paid for being an excellent power forward. He should get paid for being an excellent small forward, with maybe a bonus for his power forward skills, which would still be less than for an excellent power forward. So his argument in this case is that Lewis is not as valuable as an excellent power forward, not that he is not as valuable as an excellent small forward who can't play power forward. I can't weigh in on the truth of this, but it is logical if you think power forwards are significantly more valuable than small forwards. And considering Orlando will be paying a premium, you would think it would represent his ability to play a premium position well. Instead, it only represents his ability to play well at one postion and play competently at a premium position.

Of course, this still leaves the problem of a rating system that says in fact Lewis is not as valuable as that small forward, let alone the power forward. Common sense tells us this is false. And IIRC this is what Matt's original complaint specifically addressed, not how brilliant he thought Orlando was for signing a two-position player. I'm not sure if Berri deliberately avoids this point or if he really just didn't get Matt's argument.

Posted by ibid | July 10, 2007 3:44 PM

Is it better that Lewis can play a little PF? Maybe, but only in that it gives you some flexibility in how you construct your roster. The point is, if Lewis plays PF, then someone else has to play SF, somebody not as good as Lewis the SF. Given that Lewis is an average PF, Lewis(PF version) plus JoeSchmoe(SF) doesn't provide the same production as Lewis(SF)+JimSchmoe(PF). I'm pretty sure that's what Berri's saying here, with the caveat that I'm assuming his stats actually measure Lewis' performance at the two positions accurately, which I gather is a point of substantial question.

In fact, Lashard Rewis might BE a better player, since your coach may be better off keeping him at SF and inserting the backup PF rather than be tempted to do something stupid and play Rewis (average PF) combined with a backup SF. It all depends on the relative merits of the players sitting on Orlando's bench available to play. I'm not sure there's any reason to increase Lewis' value just because you can now be flexible on who you sign at backup PF and SF.

Posted by Larry | July 10, 2007 4:12 PM

You are not signing the player based on his inherent talent as a basketball player (which is how Matthew is looking at it), you are signing him for his production on the floor. Again taking my absurd exaggeration to show the point: I would not sign Duncan to a max deal to play PG for my team, because his production there will be low, no matter how talented he is.

No, no team signs a player for his production on the floor, except for the lower-level, average or below guys they get to plug holes. These are two different classes of free agents we're talking about. One where the team is saying "crap, we need to get someone to play PG" and one where the team is saying "okay, where do I spent my money to improve the team as much as possible?" The second question is always (or should always) be answered first, which is why Berri's system can only be useful in marginal cases involving low-level, hole-filling players. Would you sign Tim Duncan to play PG? No, of course not. But if your hole is at PG, you already have a pretty good but not great PF, and you have the chance to sign Tim Duncan for max money, is there any chance you don't do it? No, because you look at the totality of the team. Nobody goes out and shops specifically by position. It just doesn't happen. That's why performance on the floor at a given position is not that helpful, especially if you suspect somebody else's coach has been playing a guy out of position and you don't plan to.

Posted by Antid Oto | July 10, 2007 4:19 PM

Al, I think Matt is looking at the numbers, he just disagrees with the system that punishes someone for playing a second position. The basis for saying Lewis is better is that he is competitive at two positions according to the statistics, not in spite of them. In fact, he even cites Lewis' and Rewis' numbers to present them as equals -- he doesn't say they are equally talented. His never argues that Lewis is better than Berri says because he is just a great player. He argues that Lewis is better than Berri says because Berri's system is flawed. Arguing that this specific method is flawed suggests that he thinks statistics can be useful when used properly.

I also think both ability and numbers are used to some extent. If a player put up great numbers but is getting older, he won't get a contract commensurate with a younger player with the same numbers. Similarly, even for his position, Oden was not picked for his numbers but for his potential. This how I interpreted Antid Oto's comment -- the point isn't that great players who don't produce are valued as much as less talented players who do put up the numbers, it's that ultimately you need to make decisions based on what will happen in the future, and there are many reasons to expect a player will be better or worse in the future that go beyond his numbers.

I don't disagree that Orlando may be handling things poorly, but that doesn't mean that Lewis is less valuable (or at least, less valuable potentially), just that he isn't being utilized properly, just as a team using Duncan as a point guard would not make him less valuable to a smarter team that knew how to use his abilities.

Posted by ibid | July 10, 2007 4:36 PM

I'd go a step further. If you have the chance to sign one of the premier players in the game, you do it. You figure out how to fit the team around him. Of course you wouldn't play Duncan at the PG, but you could ship off your decent but not great PF for a decent but not great PG and have improved the team.

Lewis isn't in this class of player.

The other thing to remember about Orlando's position is what they already have. Howard is a wonderful rebounder and will likely improve the next few seasons. So it is more tolerable to have a poor rebounding PF for the Magic, than, say, for the Lakers, where the center's rebounding is suspect at best.

All in all, I think this was a decent signing. I don't think it changes the landscape of the league, or anything silly like that.

Posted by Curtis | July 10, 2007 4:52 PM

And Berri makes a compelling case that Orlando will play Lewis at the PF a lot.

I have yet to see this compelling case. Lewis is a SF, always has been and always will. Seattle tried him at PF and were not impressed. The stats generated which put him at the PF recently come from times when Seattle was running a small ball offense to take advantage of matchups. No one signing Lewis as a free agent would expect to use him at PF for significant minutes. That he could fill in there for short spurts is just a little bonus which I think was Matt's point.

Posted by Col Bat Guano | July 10, 2007 4:55 PM

Lewis isn't in this class of player.

See, that's really the question you want answered. I agree, Lewis isn't in that class of player. But what you want your Moneyball-type statistical system to answer is who is in that class relative to who you already have on the team overall. Position limitation is not helpful in answering.

Put it another way: Magic Johnson is remembered for having played all five positions on the floor. He wasn't as good a center as, say, Patrick Ewing, but he was a better player. That's the kind of information you need, except you need help making those judgments at the middling-star level where Rashard Lewis is. At the top the decisions make themselves.

Posted by Antid Oto | July 10, 2007 5:15 PM

Ibid, I think we are on the same page for the most part. The issue is not with the player, but rather with how the coach uses the player. As mentioned above, I think Berri makes a compelling case that ORL's coach will use Lewis stupidly, just as SEAs coach did.

That said, I don't think Matthew's Lewis/Rewis example is helpful. The choice isn't between Lewis playing 30 minutes at SF and 10 minutes at PF and Rewis playing 30 minutes at SF and 10 minutes pouting on the bench. Rewis would simply play 40 minutes at SF. And, since the example is that Lewis and Rewis are equally adept at SF, that means that Rewis IS ACTUALLY a better player on the floor for the 40 minutes that they both play.

Posted by Al | July 10, 2007 5:29 PM

I like Al at 238's post. But I do believe the WoW authors assert that their model has predictive power. The debate since they made that assertion is over how much such power, both absolutely and relative to other models offered. The jury is still out, I submit. The Iverson trade provided an experiment that, I think, supports the WoW model. That's why we talk about the Lewis move, it's another experiment. Lewis may prove to be the best 3 and the best backup 4 on the Magic and if so he has justification to be paid the value of both. (He still may be overpaid - I think backup 4's tend to get low figure salaries.)

I agree with Al that coaching decisions and quality of personnel affect the performance of a player. It's not something that any model captures and it may be uncapturable, if that's a word. (Please note: I am familiar with Dave Berri's WoW analysis that KG's teammates are the reason he has not been on a winner and this is a different argument. I am saying that a player's teammates and coach affect the individual's performance, because the coach and the things one's teammates affect the tasks the individual is called on to perform and thus indirectly affect the performance delivered.)

Matt and others make the point, with which I also agree, that versatility of an individual player has value to a team and that, to the extent the WoW model claims a predictive power as to an individual's future productivity, it does not necessarily capture (or weight properly) the value of any versatility (VoV?) he may have that a new coach or complement of teammates may take advantage of. Again, I don't know if that is a flaw unique to WoW. Hopefully WoW will be refined, or other models will develop, to capture "VoV." Of course, it's easy to see versatility when a team uses a player in a versatile manner (e.g. a Don Nelson system), but then there are plenty of other teams where it's hard to know if it's the players' lack of versatility, the coach's or gm's focus on other traits, or a distribution of talents that renders the versatility of any one player unnecessary.

Posted by MT | July 10, 2007 5:59 PM

Matt's last point is his best. Almost all of these statistical measures ignore defense, when defense is the single most important thing that players do on the court. Bill Russell proved this 40 years ago-- it's the way he won 11 NBA titles-- and it isn't any less true today.

If you can't stop the guy or zone you are supposed to be guarding from scoring, your team can't keep up with that except on rare days when it shoots an extraordinarily high field goal percentage.

Just about everything that people get wrong about basketball, they get wrong because they ignore defense. If it hadn't been for defense, Jordan couldn't have scored 28 points a game and won championships. If it hadn't been for defense, the Pistons of 1989 and of 2004 could have never beaten the Lakers. If it hadn't been for defense, Willis Reed's heroics would have made no difference. If it hadn't been for defense, Wilt Chamberlain would have retired without an NBA title.

Posted by Dilan Esper | July 10, 2007 6:20 PM

"Matt's last point is his best. Almost all of these statistical measures ignore defense, when defense is the single most important thing that players do on the court. Bill Russell proved this 40 years ago-- it's the way he won 11 NBA titles-- and it isn't any less true today."

Rebounding in these statistical measures is a proxy for defense.

It's not a particularly good proxy, but it is a proxy nonetheless.

Steals and blocks are more obviously defensive statistical measures.

-----

More generally...

Berri : Basketball :: Jonah Goldberg : Foreign Policy

There's little point in quibbling with Berri's details because he's such an idiot on the broad strokes of basketball.

Posted by Petey | July 10, 2007 6:45 PM

I'm getting into this conversation late, but I really feel the need to say something. First of all, I can't believe we're actually talking about Berri again. The guy has things totally wrong to begin with and when he engages someone who disagrees he does it in the most dishonest fashion possible.

Berri makes no case whatsoever the Lewis will be playing at PF. He observes that if Darko leaves, they might be more inclined to do it. That's hardly a strong case. Nor is it really relevant to the argument MY was making about how good Lewis is (as opposed to whether the Orlando signing was a good one based on need).

Lets look at a couple of Berri's claims:

1) We should not give players credit for playing alternative positions.

This is obviously wrong. He brings up AI: we should not give him credit for taking so many shots if he misses a lot of them. First of all this is debatable if we don't start of by assuming WoW is a complete explanation of basketball talent. But, secondly, it misses the point. Being able to sub in at a different position is valuable even if the level of play you offer is below average. Sometimes a team can't even field a 'replacement level' player at that spot on the floor. Thus you are adding value if that guy can fill in for you during a tight spot. The important thing to recognize is that the minimum a player brings is his ability to play his best position.

2) Players are asked to play positions they aren't good at all the time.

The first problem with this claim is the example he gives: Eddy Curry. How is this even an argument? Does he think NY has a better solution at center? Secondly, this is supposed to imply that teams play players at the wrong positions... but only based on WoW analysis. Maybe, just maybe, Berri's system is broken and some of the people paid to coach the game know more about it than he does.

I don't think MY is missing Berri's point at all. We are all very clear on the point Berri is making- Lewis is not a great PF. Nobody here is claiming that Orlando's needs at that position don't factor into the evaluation of that signing. What we are saying is that if Berri thinks Lewis is a great SF but a below average PF, then he must be a jack-ass to suppose that if Lewis had not logged any minutes at PF last year this would make him an overall more valuable player.

One of the obviously stupid things about Berri's system is that ironically, b/c of the limited stats it looks at, it can't take into account a player's ability to play a specific position in a team context. Because his system values rebounding and scoring efficiency, based on his system, we should just field 5 PF and C and have a great team. But this is obviously wrong for the reasons MY outlines (defense for starters). The only interesting question to fall out of all of this is: are Lewis' better than average rebounding stats (for a SF) exaggerated b/c he was asked to play some PF last season?

Posted by mpowell | July 10, 2007 6:47 PM

"First of all, I can't believe we're actually talking about Berri again. The guy has things totally wrong to begin with and when he engages someone who disagrees he does it in the most dishonest fashion possible."

Yup.

That pretty much sums things up. Berri is both an idiot and intellectually dishonest. As mentioned above, he's the Jonah Goldberg of basketball.

-----

Off topic, but I read a wonderful summary of Z-Bo's game today (no link since I can't post links):

"He is a black hole," the executive said. "I was watching a tape last season where he caught the ball in the post and spun around twice. He did a 720! I had to call everybody into the room to show them. But I tell you what, when he shot the ball it went in. He's just never going to pass."

Dude may have a one-dimensional game, but he's really good at that one dimension.

I'm actually getting semi-amped about watching the Knicks this year. They're going to be one weird team.

Posted by Petey | July 10, 2007 7:09 PM

First off, Al has this one right: Obviously, Lewis is a better all-around talent if he can play more than one position. However, the opportunity cost of playing him at PF (a small increase in productivity at the PF position (over the average player) versus a large drop in productivity at the SF position) makes it counter-productive to take advantage of the average PF skill set. That's all Berri says in his post, not that Lewis is in anyway a worse player. His commentary is only about how his game translates into wins, and playing him exclusively at SF will maximize this.

Second, whatever his intellectual dishonesty may be, his model (which incorporates defensive efficiency which over the course of a season should act as a decent indicator of defensive prowess per player) seems to work exceptionally well at predicting wins. Nobody, in discrediting him in their comments, has addressed that seemingly glaring inconsistency.

Posted by Jake | July 10, 2007 7:54 PM

The most important number in terms of Lewis is his max contract, which by any estimation is wildly overpaying for a good but not great player. And considering that he's replacing Grant Hill, who is still a pretty good ballplayer, Lewis' addition doesn't really seem to offer a marked improvement for the Magic over where they were last season.

But if you're talking Lewis as a 3 vs. a 4 you also have to take into consideration the fact that Lewis is leaving the power forward strong West and going to the East where his numbers at the 4 would likely be hugely improved, as the competition at the position will be much worse.

Posted by The CarpetMuncher | July 10, 2007 8:03 PM

"Second, whatever his intellectual dishonesty may be"

Someone really ought to write a FAQ. Berri has been thoroughly discredited by hoops stats community.

-----

"his model (which incorporates defensive efficiency which over the course of a season should act as a decent indicator of defensive prowess per player)"

Say what?

"his model ... seems to work exceptionally well at predicting wins."

Say whaaaaaaaaaat?

Dude, while I understand that meth can seem fun at first, it really ends up rotting your brain.

Posted by Petey | July 10, 2007 8:30 PM

"And considering that he's replacing Grant Hill, who is still a pretty good ballplayer, Lewis' addition doesn't really seem to offer a marked improvement for the Magic over where they were last season."

Well, no.

Judging both by stats, as well as by common hoops sense, Rashard is a sizable upgrade over Grant.

Posted by Petey | July 10, 2007 8:39 PM

Jake -

I don't remember the details of how this worked at the moment, and to be honest I'm not going to look it up, but a bunch of the other bigwig basketball statheads at apbrmetrics kept trying to engage Berri in some discussion/debate about WoW and finally got frustrated, from what I recall. Dan Rosenbaum in particular worked out that you could do, I think, anything you wanted with the weights in WoW and as long as you applied Berri's Team Adjustment mojo (which is what I don't recall in any detail) the predictive power would remain basically unchanged.

"If you wanted Ron Artest to have a low rating, you could add in a coefficient of -1000 for jersey number and your metric would predict team wins just as well as Wins Produced."

There's more on the stats front, but I don't recall the details, I just remember being more persuaded by, oh, the array of guys arguing contra Berri than I was by Berri. And that's part of the problem: Berri's a bad spokesperson for his method, because he's reclusive, defensive, condescending, insulting, argues tautologically all the time, and continually says things that indicate he doesn't have a deep understanding of basketball or even why you'd want to do statistical analysis of basketball. At some point he said he wasn't interested in doing large-scale predictions of an upcoming season (perhaps I'm misremembering the context, but we're coming to the important part) because otherwise why play/watch the games if you knew who was going to win in advance?

Posted by Quarterican | July 10, 2007 8:44 PM

"argues tautologically all the time"

That's a big part of my problem with him.

Posted by Petey | July 10, 2007 8:57 PM

And good work on the FAQ, Quarterican.

Posted by Petey | July 10, 2007 8:59 PM

I don't know anything about Berri other than what I read in the posts Matthew links to, so I am not going to take a side in this argument. But. I do feel as though his analysis, often which is counterintuitive, has some predictive value. He cites the Iverson trade, which I think is a good example. He was more right about than anyone - MUCH more right than Petey, BTW. Denver was worse with Iverson than they were with Miller, and the Sixers were better with Miller than with Iverson (and that's true even eliminating the games that Melo was suspended). I am not aware of anyone else who viewed the trade that way.

Posted by Al | July 10, 2007 9:25 PM

"I do feel as though his analysis, often which is counterintuitive, has some predictive value."

You ought to look at what Berri is actually doing. He's got a rather simplistic model that does a few things differently than all the other models.

- He wildly overvalues rebounds, which produces such results as saying that Rodman "produced more wins" than Jordan.

- He weights for team wins, which has some advantages, but also has some significant disadvantages.

- He incorporates his weird static positional structure, which leads to massively weird conversations like the current one trying to determine if Lewis is a '3' or a '4'.

He then takes this odd model and says a variety of things which the model doesn't even vaguely support, all the while citing the model as his basis. And when challenged, he changes the subject.

Read the fucking FAQ.

-----

"Denver was worse with Iverson than they were with Miller, and the Sixers were better with Miller than with Iverson"

Huh. I could've sworn the Sixers won more games in '05-'06 than in '06-'07, and that the Nuggets won more games in '06-'07 than in '05-'06. Damn internets are lying to me again.

Posted by Petey | July 10, 2007 9:52 PM

"MUCH more right than Petey"

And FWIW, I think I'm clearly on record in these virtual pages at the time of the trade as saying it would be disastrous for the 76ers to hold onto Miller because having a solid '1' to run a team of young wings would result in winning enough games to take them out of the Oden-stakes.

Posted by Petey | July 10, 2007 10:01 PM

Petey:

Rebounds, blocks, and steals are terrible proxies for defense. (Rebounding position, and coming off your man for blocks and steals, can actually come at the expense of good defense.) Ideally, you'd want to come up with a measure like "points scored against" or "field goal percentage of assigned man", though it would be really tough with zone defenses. There isn't a good statistic.

Which means you have to actually watch film to determine who the best defenders are, and what matchups favor them. That makes it almost impossible to do unless you spend a lot more time than your average basketball fan on the internet can afford to or would desire to spend.

But that's what matters. We have overplayed statistics because they are available, but only actual observation of gameplay will lead you to the best players in the league.

Posted by Dilan Esper | July 11, 2007 1:16 AM

"Rebounds, blocks, and steals are terrible proxies for defense ... Which means you have to actually watch film to determine who the best defenders are"

No doubt.

In addition, the various available stats are semi-terrible proxies even for offense.

This is why the assertions Berri makes for his model are ludicrous.

"only actual observation of gameplay will lead you to the best players in the league."

I think that you need both eyeballing gameplay as well as stats to make non-ludicrous evaluations of players. Eyeballing only is almost as bad as stats only. Both will lead you to some very erroneous judgments.

It's hard to eyeball the difference between a scorer with a 58% TS% and a scorer with a 54% TS%.

-----

Finally, I will note that for frontcourt players, rebounding is actually an underrated proxy for defense. That doesn't mean it has anywhere near perfect correlation, but you tend to find the best frontcourt defenders also having very high rebound rates.

It's no coincidence that Bill Russell averaged over 22RPG for his career.

The reasons for the correlation are worth thinking about, if you like to think about hoops stats. And the overweighting of rebounds in his model are one of the only things Berri has done that is actually worth pondering, (even though Berri took it to an absurd level.)

Posted by Petey | July 11, 2007 1:36 AM

Quaterican - "That depends - I'm not going to do the research right now, but what top flight free agent wings are due to come on the market in the next few years?"

That doesn't matter. What only matters is how good Rashard Lewis is. He's a good, not great, scorer who doesn't rebound much and plays no defense.

Mike

Posted by MBunge | July 11, 2007 10:13 AM

MBunge -

I misread you, I think. I took this:

Two years from now when they need to make another move to improve the team and are hamstrung by Lewis' stupid deal, the Magic will learn not to overpay for the best available player during free agency.

to mean "two years from now, when the Magic realize that Lewis isn't the guy to be their #1 wing scorer," and thought you weren't making any sense, because I don't think anybody clearly better (right now) is plausibly coming on the free agent market in the near future, and if they need to trade for an upgrade at the wing, Lewis' big contract will make matching salaries easier, if anything. I thought this because my perception of the Magic has been essentially that they're an elite wing scorer away from really competing (presuming they resign Darko / otherwise acquire a serviceable big to play w/Dwight Howard).

But now I think I misread you, and what you meant was something more along the lines of "what happens when the Magic realize they can't be a championship team without a better backup point and a defensive specialist / sixth man, but have no financial flexibility because Lewis eats up so much cap space," in which case presuming Orlando's management won't pay tax, you may be right.

Posted by Quarterican | July 11, 2007 10:21 AM

First time poster here, so take it easy on me.

Dave Berri's assertion is that Lewis' weak rebounding hurts his team while playing PF.

To make his point, Berri has taken Lewis’ stats, and compared them to his “average SF” and “average PF”. Looking solely at rebounds, Lewis’ 8.1 rbs/48 is higher than the average SF (7.6 rbs/48), but lower than the average PF (11.4 rbs/48).

The problem with this approach is that Lewis’ 8.1 rbs/48 is really the the combination of all the rebounds he grabbed while splitting time between SF and PF. By the way, using 82games data, it appears that Lewis played 1771 minutes at SF, and 576 minutes at PF.

So how did his rebounding stats look while playing those positions?

At SF, Lewis grabbed 7.2 rbs/48. At PF, he grabbed 10.3 rbs/48. This paints a somewhat different picture when comparing to the “average” players.

However, why stop there? Why should we compare Lewis’ stats against an average NBA SF or PF when we have data that shows how he performed against the player he was actually guarding?

For comparison, here are the rbs/48 for Lewis, his man, and Berri's average player:

At small forward:
Lewis 7.2
His man 5.7
Berri's average SF 7.6

At power forward:
Lewis 10.3
His man 10.5
Berri's average PF 11.4

At you can see, Lewis out-rebounds his man at SF, while holding his own versus PFs.

Based on this comparison, I don’t think it’s fair to say that Lewis’ rebounding (or perceived lack thereof) hurts his team.

Posted by Rasta | July 11, 2007 10:36 AM

"but have no financial flexibility because Lewis eats up so much cap space," in which case presuming Orlando's management won't pay tax, you may be right."

I think he's pretty clearly wrong over the next few years.

Orlando is only mildly overpaying for Lewis. And Lewis and Howard together are only going to be taking up a bit more than a third of Orlando's space under the tax threshold.

In short, I don't think he has the slightest clue of NBA salary math works. He's like the mouthbreathers who think the US is going bankrupt due to foreign aid and earmarks.

I will say that if Orlando does a sign & trade with Orlando, and thus gives Lewis a 6yr deal instead of a 5yr deal, I'll be a lot less impressed with the signing than I currently am. The problem with the Lewis contract lies in the out years, and the out years in a 6yr deal are going to be significantly worse than in a 5yr deal.

But if they can hold it to a 5yr deal, it's really a no-brainer. It's not easy to get elite talent in its prime.

Posted by Petey | July 11, 2007 10:43 AM

The internets aren't lying, Petey, you just don't know what you are doing. The Iverson trade didn't happen in the offseason - it happened midseason. Comparing the 2006 season to the 2007 season is thus completely inappropriate.

PHI: post trade was 29-28, pre trade was 6-19
DEN: post trade was 31-28, post Melo suspension was 25-20, pre trade was 14-9

Posted by Al | July 11, 2007 11:58 AM

"PHI: post trade was 29-28, pre trade was 6-19"

Of the 25 "pre-trade" games, Iverson played in 15. The Sixers lost 9 of the other 10.

Ugh, Al. Please don't bring your sophistical style of argument to hoops. Save it for defending Scooter Libby or other political topics.

Philly began the season with Andre Iguodala running the point, an experiment which, had it continued, would have led to a season seriously challenging Boston and Memphis for worst record in the league.

Post-trade, they replaced Iguodala with an above average '1', Andre Miller, running the point, and thus managed to put together a .500 record while playing against Eastern teams that were more interested in lottery position than in winning games.

Denver, on the other hand, didn't get their rotation and responsibilities worked out until late February, at which point they went 19 wins and 9 losses for the balance of the season.

You really ought to try watching the league, Al. It's fun stuff. I know you're getting up there in years, but I bet you can still get a feel for this stuff if you watch. On the other hand, if you like manipulating numbers, you ought to buy Berri's book instead.

Posted by Petey | July 11, 2007 12:28 PM

Petey - "Orlando is only mildly overpaying for Lewis. And Lewis and Howard together are only going to be taking up a bit more than a third of Orlando's space under the tax threshold.

In short, I don't think he has the slightest clue of NBA salary math works. He's like the mouthbreathers who think the US is going bankrupt due to foreign aid and earmarks.

I will say that if Orlando does a sign & trade with Orlando, and thus gives Lewis a 6yr deal instead of a 5yr deal, I'll be a lot less impressed with the signing than I currently am. The problem with the Lewis contract lies in the out years, and the out years in a 6yr deal are going to be significantly worse than in a 5yr deal."

Guess what, Petey. The deal that just came over the AP wire is a sign-and-trade that gets Lewis a 6 year deal for AT LEAST 110 million dollars. Combined with the contract extension they're going to give Howard, that will tie up over HALF (not a bit more than a third) of Orlando's salary cap in Lewis and Howard. As a practical matter, Orlando is giving Lewis two more years and at least 70 MILLION MORE DOLLARS than Detroit is giving Chauncy Billups.

You see, what separates mouthbreathers like you from guys like me is I understand the reality of the NBA marketplace, while you only concern yourself with theory. The reality of this situation is that the FA and his former team essentially conspire to get the the biggest contract possible out of the bidding team. Lewis wanted as much money as possible and the Sonics didn't want to let him go and get nothing in return, so they effectively bargain together against the Magic. Lewis gets an insane contract, Seattle gets a conditional second round pick and a 9 million dollar trade exception (which could very likely end up being part of a KG deal before the next trading deadline) and Orlando gets screwed.

Seattle didn't really want to pay Lewis 15 mil a year. No other team was really willing to pay him that. Yet, Orlando is going to be paying alot more than 15 million dollars a year for the next 5 years to an absolutely untradeable player, who was only the second best player on a crappy team last year. That's the REALITY of what happens when you get obsessed with grabbing the "best available talent".

Mike

Posted by MBunge | July 11, 2007 1:26 PM

"Combined with the contract extension they're going to give Howard, that will tie up over HALF (not a bit more than a third) of Orlando's salary cap in Lewis and Howard."

As I wrote, and as you quoted me:

Lewis and Howard together are only going to be taking up a bit more than a third of Orlando's space under the tax threshold.

Do you actually find that confusing? Do you just not read the things you quote?

The cap is not particularly important. What is important is the tax threshold, since that actually constrains almost every team in the league.

And Lewis and Howard together will indeed take up a bit more than a third of Orlando's space under the tax threshold.

Compare and contrast with actually dire situations, like Philly's last year with two players taking over 60% of the tax space. Or New Jersey's this year, with three players taking almost 70% of the tax space.

You don't get the math, but you still have emphatic views of how you feel the numbers must add up. That's what makes for a mouthbreather.

"No other team was really willing to pay him that."

You, sir, are a dumbass.

Posted by Petey | July 11, 2007 2:11 PM

After reviewing all the comments above I have to agree with Al. His detractors (namely Petey) seem to resort primarily to name calling and condescension as debate tactics and the WoW model as a whole is much more convincing than the conventional (and subjective) wisdom espoused by Matthew and others.

Posted by Will | July 11, 2007 2:53 PM

Petey, ya got me. The argument that having two players taking up over 30% of a team's luxury tax space is a GOOD thing was so amazingly stupid that I couldn't really believe my eyes that someone was making it. I guess I hadn't quite accustomed myself to your level of stubborn dumbassery. My bad. By the way, Orlando says they still want to resign Darko and if they do, that'll likely be over 40% of luxury tax room tied up in 3 players - one of which was the second best player on a crappy team and the other might never be any better than Kwame Brown.

The salary cap, idiot, does in fact matter quite a bit. That's because once you get over it, which the Lewis and Howard deal will pretty much do to Orlando for the next 6 years, you're greatly constrained in your ability to add talent to your team. Who's going to trade or do a sign-and-trade for anyone else on Orlando's current roster, because that's the only way the Magic can add any more talent.

And where exactly did you find a list of teams willing to give Rashard Lewis a 5 year, 75 million dollar deal? I didn't see any of the other teams under the cap getting in line to do that, and the teams over the cap don't have anything Seattle would want in a trade so it doesn't matter what they want.

Orlando is giving Rashard Lewis a 6 year contract worth at least 110 million dollars. That's the reality, not your theory. Are you going to defend that, or are you going to chicken out like you previously indicated?

Mike

Posted by MBunge | July 11, 2007 3:09 PM

"His detractors (namely Petey) seem to resort primarily to name calling and condescension as debate tactics"

I actually find a bit of name calling and condescension are wonderful tactics when I know precisely what I'm talking about and someone else is clueless.

I tend to be far more polite when that particular dynamic isn't present.

Posted by Petey | July 11, 2007 3:22 PM

By the way, Petey, I just did some calculations and Lewis' 6 year, at least 110 million dollar deal and Howard's expected 5 year, 85 million dollar deal averages out to about 35 million a year for the two of them (a little less in the first few years, a bit more in the last few). So unless the NBA Luxury Tax Limit increases to well over 100 million dollars (over 40 million more than it was last season), those two deals will take up way more than a third of Orlando's luxury tax space as quickly as the season after next.

Mike

Posted by MBunge | July 11, 2007 3:25 PM

Petey - "I actually find a bit of name calling and condescension are wonderful tactics when I know precisely what I'm talking about and someone else is clueless."

You know, I respect a guy more if he's just as ass all the time. Being an ass only when you think you're right is not only counter-productive but it's a fairly weenie way to behave.

Mike

Posted by MBunge | July 11, 2007 3:28 PM

"Orlando is giving Rashard Lewis a 6 year contract worth at least 110 million dollars. That's the reality, not your theory. Are you going to defend that, or are you going to chicken out like you previously indicated?"

Chicken out? Are you 12 years old?

I thought I was pretty clear upthread. The 5yr deal seemed like a no-brainer to me from Orlando's perspective. The 6yr deal is a bit more troubling.

They're going to be paying $25m for a 35yo in that final year.

But, since they had competition willing to offer Lewis the 6yr deal, it was that or nothing. And if I were Otis Smith, I think I would've thought about it for 30 seconds, swallowed hard about the final year, and done the deal.

They'll definitely be overpaying in the last year or two, but that seems a reasonable tradeoff for what they're getting. In a lot of senses, Orlando didn't really have a choice here. Their future without signing Lewis simply wasn't very appetizing.

Posted by Petey | July 11, 2007 3:32 PM

"You know, I respect a guy more if he's just as ass all the time."

Jesus. You are 12 years old, aren't you.

Posted by Petey | July 11, 2007 3:34 PM

I have no problem with tying up a ton of money in just two players if they're worth it, but Rashard Lewis isn't. In fact, Petey, given how convinced you've been in the past that small forwards are rarely game-changers, I'm surprised to hear you defending the deal, even apart from the fact that Lewis simply isn't that great. Is Orlando going to contend for a title in the next two to three years because of Lewis? No, of course not. So why tie so much to him? Why not accept sucking for another year while Howard develops and then see what you can add through the draft, free agency, or trade next year, especially at point guard?

As to the arguments about Berri's "system," I'll just repeat what I said initially: statistics are useful to GMs insofar as they help you evaluate deals in the context of the whole team. Knowing that Lewis is a better small forward than power forward doesn't help you do that in any significant way.

Take the shift from considering batting average to OBP and slugging percentage as important indicators of offensive contribution. Does it make any difference to know how well a guy bats at various spots in the order? It shouldn't.

Posted by Antid Oto | July 11, 2007 3:50 PM

"Why not accept sucking for another year while Howard develops and then see what you can add through the draft, free agency, or trade next year, especially at point guard?"

When Howard's extension kicks into effect, their cap space will disappear.

The Magic are in a use it or lose it situation. This is their only summer to be able to grab an elite player.

And I'd rather pay Rashard $15m than Gerald Wallace $10m. It seems a far safer bet to me.

-------

"In fact, Petey, given how convinced you've been in the past that small forwards are rarely game-changers, I'm surprised to hear you defending the deal"

I hear you loud and clear.

But at the end of the day, I think this is Howard's team and they'll get as far as Howard will carry them. But Howard may not be a real scorer. So he needs a scoring running mate.

To get historical, I think the model here is Howard as Moses Malone and Lewis as Dr. J. That title winner was Malone's team, but since Moses had offensive limitations, having a guy like Erving could provide synergy.

(Lots of ways that's an imperfect model, but you get the point.)

A SF isn't the best guy to base your team around, but it's not a bad second star to have when your main guy is a bruiser inside.

And I think Lewis is a bit more elite than the consensus around here has it, as well.

Posted by Petey | July 11, 2007 5:01 PM

Petey, you expect me to believe that the fact that the Sixers went 5-10 prior to Iverson quitting somehow supports your view that the Sixers didn't improve by trading him for Miller? By any measure, the sixers were better with Miller than with Iverson. But, as we know, you are a member of the reality-based communty where up is down and 5-10 is better than 29-28. But you're right that the Sixers screwed up by gwetting the better player in Miller - they should have tanked the season to get Oden... Because we all see how well that worked for the Celtics.

Posted by Al | July 11, 2007 9:36 PM

Also, the rest of the NBA seems to agree with Lewis isn't that "elite", seeing as how he's only made a single all-star team. I can see that in an up-and-coming elite player who isn't known widely yet, but not a supposedly "elite" player who's been in the league for 9 years already.

Posted by Al | July 11, 2007 9:49 PM

This thread is probably dead, but I do want to recommend my own work here for possible inclusion in the anti-WoW FAQ.

It takes a little while for me to develop my argument but, ultimately, I argue that (1) WoW overvalues rebounds, even using it's own logic and (2) this fact can be seen with an example of two teams, one of which uses possessions more efficiently but, in the process, generates more defensive rebound opportunities for the opposing team, and the team that is less officient, but gives it's opponents fewer rebounding opportunities. The second team rates higher in WoW calculation despite being the less efficient team.

I still believe that there are good ideas worth salvaging from Wages of Wins, but once I figured out the above I feel completely conviced that it needs revision.

Posted by NickS | July 11, 2007 11:45 PM

NickS - As your correspondent on that thread, I have to add (to this possibly dead thread) that your arguments were somewhat unfounded. You failed to understand the distinction between Win Score and Wins Produced, which is why the numbers weren't coming out right for you, a point I made at the time. There is no discrepancy when he calculates his numbers using the full method. There may be valid criticisms of his method, but I don't think that specific post you linked to qualifies as such.

Interesting thread here with many insightful comments, especially before Petey got involved. Berri is being misunderstood I think, as frequently happens. It seems to me almost everyone here agrees Rashard Lewis, while a great player, doesn't deserve a max contract, which is Berri's main point. It also seems unlikely that this move will catapult the Magic into championship contention. And if it doesn't, the opportunity cost of Lewis will probably be a true superstar who could do so a few years down the line, if paired with Howard.

To me Ibid got it exactly right when he said:

"a) Lewis is an excellent player, and (b) he can play power forward, it doesn't follow that he should get paid for being an excellent power forward. He should get paid for being an excellent small forward, with maybe a bonus for his power forward skills, which would still be less than for an excellent power forward. So his argument in this case is that Lewis is not as valuable as an excellent power forward, not that he is not as valuable as an excellent small forward who can't play power forward. I can't weigh in on the truth of this, but it is logical if you think power forwards are significantly more valuable than small forwards."

And I would add that in Berri's system, power forwards and centers are in fact considered significantly more productive than small forwards and guards.

Posted by Flint | July 12, 2007 12:43 PM

"And I would add that in Berri's system, power forwards and centers are in fact considered significantly more productive than small forwards and guards."

This seems like a flaw in the system (or a symptom of the flaw, which is that he ridiculously overweights rebounds). Does anyone who's ever actually watched a basketball game think that power forward, in general, contribute a lot more than point guards?

Some stat dude whose name I can't remember did an analysis years ago to try to determine the most valuable positions. He looked at the correlation between having an elite player at a certain position, and team performance -- how many games did the teams with the best point guards win, as opposed to the best centers, etc? He found that good point guards and good centers were the most valuable, followed by small forwards, then shooting guards and power forwards. No, I can't find a link for this, nor can I even remember the magazine I read it in. I think it was back in the early 90s Jordan era, though, so things might have changed.

Posted by too many steves | July 12, 2007 12:52 PM

Too many steves: Not only would I agree that point guards and centers are most valuable, they are also scarcest. You can always overpay for a good center or point guard for both reasons. I would not agree that power forwards are less valuable than small forwards, but I think it's a much smaller difference in any event.

Posted by Antid Oto | July 12, 2007 1:14 PM

Flint, it's true that we never convinced each other. You say that the numbers work out properly when the more complicated Wins Produced metric is used rather than Win Score.

But in my May 20 5:20 post I tried doing the calculations using the Wins Produced weights and ended up with exactly the same results. I don't remember you ever finding a flaw with that, you just responded by saying "Clearly, Wins Produced is going to come out clean any way you slice it. There may be fatal flaws in his approach, but the numbers adding up is not going to be it. I dont think he gets the book published or an economics professorship without being able to get that right." which begs the question.

I never did respond to your point that it does work if you only include defensive rebounds and not offensive rebounds, but the whole point of my argument was that the sum of the weights that he gives to offensive and defensive rebounds is off, so that either weight could be correct in isolation, but that they can't both be correct.

Posted by NickS | July 12, 2007 1:17 PM

Apologies, for going somewhat off topic (and decending into minutia) in my last comment, but I happen to think it's interesting.

Posted by NickS | July 12, 2007 1:19 PM

NickS -

When calculating Win Score, one does not account for opponent 3 pt and 2 Pt field goals made. That is the source of the error you correctly identified in your posts. I actually hadn't really thought it through when we were posting. I thought Win Score should work out correctly. But it obviously doesn't. It was designed as a quick and dirty metric to capture a snapshot of performance in one real NBA game. And it does an excellent job at that. But you can create a hypothetical scenario where the numbers get wacky.

However, with Wins Produced, the numbers work out. Opponent Field goals made are deducted from the player's total. This happens at the team level in Berri's work. (the defensive adjustment people get worked up about.) In your scenario, which was 1-on-1, they should be deducted from the players total And if you do so, the numbers do work perfectly,

I think my May 20 post at 621 was my clearest response (i didn't have too many.) If you calculate the numbers in that fashion, i.e. with deductions for made field goals, they do work out correctly.

Hope that is a satisfactory response...

Posted by Flint | July 12, 2007 2:11 PM

I guess my challenge to you, is to get the numbers to work for my example. You got the numbers to work out by removing one element of the example (offensive rebounds) and that my point, made at 1:17 above, was that the numbers can work including either offensive rebounds or defensive rebounds but not both.

What I would say is that if you can get the numbers to work for that example using both offensive and defensive rebounds than you should post it in that APBRMetrics thread.

I would be quite pleased if you could get the numbers to work for my example.

Posted by NickS | July 12, 2007 2:19 PM

"Also, the rest of the NBA seems to agree with Lewis isn't that "elite"

So why were teams lining up to offer him a $110m deal then?

Posted by Petey | July 12, 2007 7:09 PM

"WoW overvalues rebounds, even using it's own logic"

I have no idea if you are correct about this or not, NickS, but even if you are wrong and WoW follows its own internal logic, I still the way it weights rebounds is plainly absurd.

Posted by Petey | July 12, 2007 7:12 PM

NickS - Honestly, getting into all those fractions and decimals doesn't really interest me. It was a very untidy, non real world example, and my math isn't that good. But if you do what i did in the comment I cited, the numbers will come out right.

As i said, there may be many problems with Wins Produced, but the numbers not adding up as you suggest they don't just isn't one of them. I can't remember accomplished statheads like Rosenbaum making that point either. The defensive adjustment may be a fudge factor, but it at least is one the makes the numbers square. Rebounds may be overvalued, but its not because of a numbers glitch.

Petey - Your answer. Because NBA GM's are morons. Rashard Lewis gets 110. Gerald Wallace gets 57. Orlando would have been much better off with Wallace. Berri had him at a .335 last year, and NBA Babble has him at a .245 this year. He is already a better player than Lewis. And he would have left them with money to acquire a third piece. You will say that Lewis is the perfect complement with his outside shooting range, unlike Wallace, but I don't think thats the case.

The Bobcats are looking good, Okafor, Wallace, and Richardson. Just imagine if they had taken Roy rather than Morrison...

Posted by Flint | July 12, 2007 10:21 PM

"Orlando would have been much better off with Wallace. Berri had him at a .335 last year"

But I think Berri's system is fundamentally flawed, and that his numbers are not a particularly good measure of player value.

"He is already a better player than Lewis."

Any system that massively overweights rebounds is going to undervalue a player like Lewis.

The Berri dittoheads don't have much grasp of how things actually work.

"And he would have left them with money to acquire a third piece."

If Orlando had wanted to sign Wallace, they'd have had to pay him quite a bit more than $57m, obviously. The real comparison isn't 110 vs 57, it's 110 vs 80. And no one with a higher IQ than Berri would take Wallace over Lewis in that situation.

Posted by Petey | July 12, 2007 10:42 PM

Petey -- Yes, I do think that my arguments are, largely, an attempt to add rigor to a relatively common sense critique. I also think it's ibteresting to delve into Wow a little bit to try to separate some of the ideas that I think are important and worth salvaging from some of the sloppiness.

Flint -- I do appreciate your willingness to be a guide through some of the details of WoW that I had forgotten, but I'm starting to get annoyed by the feeling that you aren't willing to engage my argument.

1) If I construct an example that I say reveals problems, it isn't a response to construct a different example and say, "it works in this case." That just raises the obvious question of why would it work in one case and not in another -- a question that I am quite interested in trying answer and which you don't seem to think it at all interesting.

The defensive adjustment may be a fudge factor, but it at least is one the makes the numbers square.

2) This may be true but is also not convincing. You seem to be accepting Rosenbaum's argument that a team adjustment can make any set of numbers add up on the team level, not matter how messed up the original numbers are. That is not a convincing argument for Wages of Wins.

3) Enough with the aguments from authority. It was annoying when you said, "I dont think he gets the book published or an economics professorship without being able to get that right." Which just argues, "because Berri has qualifications everything he says must chech out." But to argue, "I don't think your criticism is correct because it isn't a criticism that Rosenbaum has made" utterly fails to convince me of anything because I believe that I am making a new criticism. What's more, I don't think that Dan Rosnebaum has a monopoly on criticisms of David Berri.

Finally, Flint, I'll say again what I said in the APBRMetrics thread -- as far as I'm concerned you should be excited about the questions that I'm asking. If you think Wages of Wins is trying to do something interesting in explaing basketball, than me asking questions is just a way to make it better. If someone said to me that they found a problem in the way that DeanO calculated Ortg, I'd want to know. I use Ortg as a summary statistic all the time, and if it had a problem, and there was a way to make it better. I'd rather have a better Ortg than a flawed Ortg.

You seem more interested in defending the idea that Berri must be correct in what he's written than in actually working through the strengths and weaknesses of his approach.

Posted by NickS | July 12, 2007 11:40 PM

Petey - Why would they have to pay Wallace 80 million rather than 57?

NickS - I hesitate to tackle your scenario, with one guy shooting 33% on threes, the other shooting 50% on twos, each with a 20 % rebound rate, etc. I tried doing it originally, and it gets extremely messy.

If you want to post a reasonable game sequence on the thread I will take a stab at scoring it. But I don't see the point in an example where you have one guy who is scoring 33.3 points, then rebounding 13.32 shots, then converting 4.4356 of those second chances. It's just bizarre, and I feel like I am so likely to make a mistake in figuring out the basic numbers that the exercise is basically worthless.

But, also, really, the numbers work. There is no secret flaw. Rebounding is not overvalued because of a numerical glitch. It's overvalued because the rebounder gets credited for what is a team defensive achievement. (And fwiw, i dont believe that, but that is how the argument goes I think.)

Posted by Owen | July 13, 2007 1:12 AM

"Why would they have to pay Wallace 80 million rather than 57?"

Because unless Wallace simply wanted to go to Orlando for reasons other than money, Orlando would've had to outbid Charlotte, as well as whatever other teams got involved with sign & trade deals if the deal became too rich for Charlotte.

It would've become a bidding war, just as the Lewis deal became.

If Wallace could get $57m with only one team bidding, what do you think would've happened with multiple bidders?

"Rebounding is not overvalued because of a numerical glitch. It's overvalued because the rebounder gets credited for what is a team defensive achievement."

Rebounding is overvalued due to Berri's subjective whim.

One has to credit Berri's understanding of what works in basketball in order to credit his choice of weighting.

Posted by Petey | July 13, 2007 1:24 AM

"If Wallace could get $57m with only one team bidding, what do you think would've happened with multiple bidders?"

I guees I don't understand why it didn't become a bidding war. What incentive did Wallace have to keep the bidding down? Why wouldn't Orlando at least have used Wallace as leverage against Lewis? Gerald Wallace is a young, excellent player. I think his services on the open market ought to have been worth, if not the max, something closer to it than he got. Perhaps he didn't offer enough at Power Forward in their mind...

As for the weight of a rebound, it's hardly a subjective weight. He derives it by regression analysis. And FWIW he use the same weighting for rebounds relative to other stats as Hollinger. It's his accounting of missed shots and his position adjustment that make the real difference.

Posted by Flint | July 13, 2007 8:26 AM

"I guees I don't understand why it didn't become a bidding war."

Because with Orlando committed to Lewis, there was no one more than $8m under the cap. You can't have a bidding war if it's not possible to have any bidders.

"As for the weight of a rebound, it's hardly a subjective weight."

Bullshit.

Read up on what Berri is actually doing, rather than just accepting his lies at face value, and you might actually get a clue about this.

And the core lie that is the underpinning for all the other lies is that his weighting is somehow "derived from the numbers". It's not. And until you understand that, you won't get why he's a flim-flam man.

Posted by Petey | July 13, 2007 8:42 AM

Ok. I see that logic. I just don't quite understand why Orlando was so much higher on Lewis that Wallace, and why, considering they had the most avalable cap space, they didn't play the two off against each other. But obviously their view of Wallace's talents relative to Lewis' is more in line with yours than mine.

As for Berri, you honestly make me laugh. I would love to hear what you think of Sabremetrics...

Posted by Flint | July 13, 2007 8:51 AM

NickS - I hesitate to tackle your scenario, with one guy shooting 33% on threes, the other shooting 50% on twos, each with a 20 % rebound rate, etc. I tried doing it originally, and it gets extremely messy.

That's where I started from because I wanted a scenario in which the two teams were the same in some regard. But the simplest example is based on that, but doesn't have any of the fractions.

Team A:
39-115 from 3-point range
15 offensive rebounds
(44 defensive rebounds)
(117 points scored)

Team B:
56-111 from inside the arc
11 offensive rebounds
(61 defensive rebounds)
(112 points scored)

Each team uses 100 possessions. Team scores 117 points, team B scores 112 points.

Win Score for each team (by my calculations in the 5:20 post in the linked thread)

Team A: .204
Team B: .226

Team B has a score that is about 10% higher despite having a lower score.

The reason why I make a big deal out of this is because (1) It's an example that's constructed to reveal a theoretical problem -- the fact that Berri has the value of an offensive rebound and a defensive rebound add up to more than the value of a possession and (2) Berri argument is based on the idea that he starts with results at the team level. Various people have criticized the ways in which he goes from team stats to individual stats (and I have my own criticisms of that) but I'm arguing here that if you can construct an example in which it doesn't get the team stats right (when Win Score can't accurately tell you which team won the game) than the problems involved in rating players on team stats becomes secondary. It isn't even accomplishing it's starting point which is to have an accurate accounting of how team's win games.

Posted by NickS | July 13, 2007 11:03 AM

Gosh darn it, now I am hoist by my own petard.

Looking over the numbers I see a mistake in the linked post, I have team B as missing 44 field goals when, of course, they miss 55 (having gone 56-111 from the field)

With that correction the numbers are:

Team A: .204
Team B: -.148

It does correctly identify the winner of the game though, now, the difference seems larger than you would expect from the difference on the scoreboard.

I'm still convinced that, on a theoretical level, the weights for an offensive and a defensive rebound can't add up to more than the value of one possession, but I believe that this example doesn't prove the point. I'll try to come up with an example that does.

Posted by NickS | July 13, 2007 11:18 AM

Watch this space, BTW, because I will try to find another example.

The reason why the example above fails in Win Score is because, as someone else pointed out in the APBRMetrics thread, Win Score double counts a missed shot, because it gives a penalty to the team that misses the shot and a bonus to the team that gets the rebound. The reason why it work in Wins Produced is that Wins Produced also double counts a made basket in that it gives a credit to the team that hit the basket and a penalty to the team that allowed the basket. So the fact that missed baskets have to entries on the ledger and so do made baskets means that they are scaled to each other. But if there is a stat that doesn't get a double entry (Turnovers? I will need to check) that could be used to construct an example that will fail in Wins Produced.

Posted by NickS | July 13, 2007 11:31 AM

Okay, I've found my copy of WoW and looked over the values for Wins Produced and found that, I hate to admit it, Flint was right that the values do work out as long as you don't include steals or blocks (which is actually reasonable).

Wins Produced is set up so that a game between two teams is a zero sum game. Barring rounding errors, every action results in a credit for one team and a matching debit for the other team, so that the total of the two teams will add up to zero.

One of the things that this means is that both WinScore and the popular discussion of Wins Produced are somewhat misleading. Most people talk about a rebound being worth a possession in the WoW formula, but it's actually worth half of a possession. Which means that Wins Produced does follow the basic rules of a linear weight system -- that the value of a missed shot and an offensive rebound are equal, and that the value of an offensive rebound and a defensive rebound add up to the value of a possession (all 3 of those events are valued at half a possession).

With that out of the way, I want to move onto another, and slightly more confusing area of Wins Produced. (and, again, I'm largely trying to find a way to clearly express an complaint that many people intuitively have about Wins Produced).

What is the baseline for Wins Produced? This is made more confusing because of Berri's various adjustments, but we know that Wins Produced, the linear weight produces a raw score that is than adjusted for team defense and for position.

We also know that the production of an average player is defined as .1 wins/minute. We also know that a player that shoots with an eFG% of 50% receives no credit for their scoring because they are only scoring at an "average" rate.

But wait, isn't that odd? An "average player" produces Wins at a rate above zero, but an average scorer produces no wins at all. In other words, a player receives credit for being an average player, but that credit comes entirely from their non-scoring contributions.

Why should a player receive credit (the .1PAWS/min of an average player) for being an average rebounder at their position, but receive no credit for being an average scorer at their position.

Look at it this way, any linear weight system has, by definition, a 0 point for a player who stands on the court and doesn't do anything. A player who is on the court but gathers no rebounds, takes no shots, never turns the ball over will, by definition have a raw WinScore of 0. A player who is on the court and does things at an average rate will have some positive score, which is than adjusted for team defense and position.

But the odd thing about Wins Produced is that it makes no distinction between a player who stands on the court and does nothing, and another player who goes on the court and gets no rebounds, assists, steals or blocks but shoots 15-30.

I would argue that a reasonable linear weights system would want to be able to distinguig between those two players.

I would argue that if you're going to say that an average player produces .1win/min that the ability to both score at an average rate for the position and the ability to generate RASB at an average rate should both go into that .1win/min. But, Wages of Wins does not believe that.

That is my other problem with wins produced.

Posted by NickS | July 13, 2007 1:17 PM

Petey - This one is for you.

http://sports.aol.com/fanhouse/2007/07/11/rashard-lewis-wont-let-gerald-wallace-eat/

NickS - I am glad we could resolve the question amicably.

I am not exactly clear on why a rebound is worth half a possession, or rather why you think that. But I definitely agree that in the WOW an offensive rebound, defensive rebound, and a fg attempt have the same value.

I think the answer to your second question lies in the position adjustment. Wins Produced basically is a measure that assesses players according to how far above and below average they are for their position for each stat, while weighting each stat overall according to the values suggested by his regression. I don't believe (anymore, thanks to you) that the breakeven point for scoring is 50 efg%. I think its the average for the position. I also believe, and I am not privy to the inner workings here, that raw scoring is an element also.

If that's true, in your example, it would seem to me the player who goes 15-30 would be more valuable in WP, since he he is matching or closer to the average for his position for scoring efficiency and output.

The average center scores 17.7 points per 48, and to the extent a player is close to that, he receives credit, or at least more credit than a player who doesn't score at all.

I am not sure that helps, or if its even totally correct, but that's my stab at it.

Berri has promised a response to this post today, bo hopefully that can be more for you to digest...

Posted by Flint | July 13, 2007 2:57 PM

This might sound funny, but I don't believe WoW overvalues rebounds at all. But it does undervalue the impact of scoring. It does this by penalizing the shooter for every shot attempt, unlike other metrics that penalize only for missed shots.

My analysis indicates that a whopping 96% of all points scored are negated by Win Score's formula. To illustrate, consider Kobe Bryant's 2005-06 season (sorry, I haven't updated my database for 2006-07 yet).

Kobe averaged a league leading 35.4 ppg. Those points are all included in his Win Score. The next step is to subtract 1 point for each FGA, and 0.5 points for every FTA. At this point, before considering any other stats, Kobe's Win Score from shooting activities has been reduced from 35.4 to 3.1. That's a 91% reduction in his offensive output.

Consider that most other metrics (PER, NBA Efficiency) penalize the shooter for MISSED shots, not for every shot attempted. In these cases, the scorer's offensive output is reduced somewhere between 40% and 50%. Kobe's 35.4 points would be reduced to 19.8 points, an fairly typical 44% decrease.

Getting back to Win Score, the formula then adds or subtracts points (or half points) for rebounds, steals, blocks, assists, TOs and PFs. If you accept that these weights are all basically reasonable (which I do), the fact is that they far outweigh the positive impact of scoring 30+ points.

Consider an extreme example. Say Kobe scores 35 points on 32 shot attempts, but does absolutely nothing else statistically in the game. Compare his Win Score of 3.0 to another player who does absolutely nothing except to grab 3 rebounds. Using Win Score, these two players are equal.

This, of course, is ridiculous.

However, you cannot say that rebounds have been overvalued. They have not. Instead, you would have to conclude that the positive impact of scoring has been severely penalized.

Win Score advocates would counter this argument by saying that PER and other metrics do not penalize inefficient shooter enough. This is true, and needs to be addressed. The WS approach, however, is not the answer.

Posted by Rasta | July 13, 2007 4:26 PM

Rasta -- I think we agree. Saying that "rebounds are overvalued" is shorthand for saying that "rebounds are overvalued relative to scoring and to the overall rankings."

Posted by NickS | July 13, 2007 4:30 PM

Scoring 35 on 32 actually isn't that good; in 05-06, Kobe averaged 35 on 27 shots.

By the way, when I was looking up that Kobe # on Basketball Reference, I noticed that in 86-87, Larry Bird scored 28 points on 20 shots a game. His TS% was .612. Holy shit, that's amazing.

Posted by too many steves | July 13, 2007 4:48 PM

in 86-87, Larry Bird scored 28 points on 20 shots a game. His TS% was .612.

In 83-84 Adrian Dantley scored 30.6 points on 18.2 shots with a TS% of .652 -- one of the greatest scoring seasons ever.

Posted by NickS | July 13, 2007 4:52 PM

"As for Berri, you honestly make me laugh. I would love to hear what you think of Sabremetrics..."

I've got nothing against Sabremetrics in baseball. And I've got nothing against APBRmetrics in hoops.

What I do have something against is Berri's flim-flam strategy of concealing the subjective choices made in his system behind a false appeal of "scientificky" pseudo-objectivity.

There is nothing inherent in the game of hoops that leads one inexorably to Berri's weightings and model. Subjective choices have been made, and Berri is loathe to discuss those subjective choice, or even to admit that subjective choices even exist, all in the effort of trying to preserve the "scientificky" veneer around his model.

Figure this out, and you'll understand the core fraud behind Berri-ism.

Or to put it another way, if you like APBRmetrics, you should find yourself offended by Berri-ism.

Posted by Petey | July 13, 2007 4:57 PM

Nick,
Yes, I think we are on the same page regarding WoW. I especially like your example:

"But the odd thing about Wins Produced is that it makes no distinction between a player who stands on the court and does nothing, and another player who goes on the court and gets no rebounds, assists, steals or blocks but shoots 15-30."

That pretty much sums up my main complaint regarding Win Score.

That said, I think Berri has done some really fascinating and even groundbreaking work. The direction my analysis is heading is largely influenced by the concept of Wins Produced.

Some points of distinction are:

1. Penalizing shooter for missed shots only
2. No positional adjustment required
3. Compare the players performance to the actual player he's guarding (using 82games data)
4. Adjust the players metric up/down depending on number of offensive attempts compared to opponent
5. Player's relative value is determined by whether (and by how much) he outplays his opponent

Posted by Rasta | July 13, 2007 4:58 PM

"Scoring 35 on 32 actually isn't that good; in 05-06, Kobe averaged 35 on 27 shots."

He also averaged 10.2 free throw attempts. Hence the 32 shots in my example.

Posted by Rasta | July 13, 2007 5:01 PM

Berri said he would address this comment, but it looks like he is saving it for next week.

Rasta, I applaud any attempt to innovate. Berri often says on his blog and in his book, that he doesn't think his will be the final word on basketball statistics.

Vis a vis points 1-5, I dont care what you do with missed shots, but in my mind, its very clear that the two most common metrics, EFF and PER, overvalue scoring and undervalue efficiency. So your new metric should incorporate this change, as Berri's does.
2. IF you can dispense with positional adj. more power to you. It seems to me to be a very useful concept though.
3. Berri actually has already attempted to do this, finding the differences were marginal. See what you think at this post.

http://dberri.wordpress.com/2007/03/26/incorporating-defense/

4. I don't really understand this one.
5. This one I dont think is really relevant. Basketball is not five games of one-on-one. But will happy to see what you hope to achieve when you finish your work.

I think one of the biggest pluses of Berri's work is that Wins are the basic currency of his work. That makes it much more useful than anything done before. You should see if you can follow his pattern.

Petey - If i could figure out what you mean, perhaps it would help. But your comment is, like always, vague and vituperative. My impression is that the foundation of Berri's work is a rigorous regression analysis of the value of stats at the team level. This doesn't seem like a very subjective starting point. The assumption that these weights can be translated to the individual level is perhaps iffy, but I am a firm believer for the moment. I think the next couple of years, if he sticks with it, should provide an excellent test of his method. Every transaction provides new opportunities, which is sort of exciting. And the way his method is structured at the least is rigorous enough that it can be falsified.

So basically, until it's disproven by more reasonable people than you, I will just enjoy it and what it has to say about my favorite players, Renaldo Balkman and David Lee.

Posted by Flint | July 13, 2007 7:46 PM

The assumption that these weights can be translated to the individual level is perhaps iffy, but I am a firm believer for the moment

But that seems like a big assumption. Look at Renaldo Balkman, whom I love, it seems obvious that if you could have a team that matched his statistics (more Steals than TOs, as many Blk as TOs, shooting 50%, getting offensive rebounds on 70% of missed shots) that would be an amazing team. But it seems like a stretch to think that 5 Renaldo Balkman's could put up those numbers as a team.

Secondly, it seems like a big stretch to assume that calculations at the team level are a reliable way to distribute credit for team success to individuals on the team. Look at Phoenix, for example, they are the 4th best team at the league at avoiding TOs. But Steve Nash commits tons of turnovers. According to 82games.com he not only leads the league in bad pass TOs, he has 50% more bad pass TOs than the second place person in the league (Kidd), and he was 3rd in the league in total TOs. Just imagine how low TO a team Phoenix would be if they could get rid of Nash. They'd lead the league by a mile. It's amazing that they're willing to keep having letting him run that team when he throws the ball all over the place like that.

Obviously Steve Nash is a unique player so he may not be representative, but it seems clear that, even though Nash has an average Ast/TO ratio, the fact that he is able to handle the ball so much helps everyone else have such low TO numbers.

Any linear weight system will wrestle with problems like that but, I agree with Petey, that part of what's frustrating about Berri is that he doesn't seem to want to talk about the problems of player rating. He's happy to talk about his successes, but he has given the impression of actively avoiding any conversations with the APBRMetrics crowd about the limitations of his system or linear weights in general.

Posted by NickS | July 13, 2007 10:21 PM

Berri makes quite clear that basketball requires several different kinds of players. His system is basically built around that notion actually. He has responded to and elaborated on the 5 Balkman arguments in the book and numerous times in his blog. It really shouldn't be an issue at all, but as a pseudo WOW Huxley, I can say I encounter that argument almost every time I have this conversation on a message board. And when I hear it, it tells me that perhaps my correspondent hasn't read Berri closely enough.

Also, re 5 Balkmans, it would be an amazing team to watch, incredible defense, horrendous bricks. lots of dreads.

Re Nash, he was ninth in the league amongst qualifiers in ast/to, and fifth amongst starting point guards, so he was no slouch and in fact was well above average on that metric. Nash is also extremely highly rated by the WOW metric, suggesting perhaps that the method is taking account of this value he adds to a team. But your general point is that Nash helps his teammates, and that is correct. Berri has actually addressed this point in this post.

http://dberri.wordpress.com/2006/10/21/the-phoenix-suns-in-2005-06/

Berri has public relations issues, and he doesn't react well to strident criticism. The fact that he thinks he may have come up with a better, simpler method for evaluating basketball players rubs a lot of people the wrong way. I respect your criticism about his lack of engagement. But as someone who follows his blog very closely, I can sincerely say that he has made an enormous effort to respond to comments and to answer politely framed questions. But there are many more Petey's in the world than NickS'. Having been in the middle of a number of these WOW debates, I can understand why he doesn't want to engage his critics. In general, they are a very hostile, rude bunch with little real understanding of what he is trying to do. You are a real exception in having actually purchased and read a copy of the book, and having made a sincere attempt to understand how the system works.

Beyond his content, I think Berri's style and manner, also touch a deep nerve. I think there is room for improvement there. But to be fair to him, there have been a lot of vicious things said about him that were totally unfair. Dan Rosenbaum went on to a WOW sister site, NBA Babble, and heckled anonymously. It was childish stuff. Berri hasn't always risen above that kind of behaviour, but certainly of late he has made huge progress in that respect.

However, I do think he assumes that the kind of APBR debate you envision will be completely unproductive. My judgment, knowing what I know about his critics in general and about the atmosphere even at APBR, which I think is an absolutely wonderful forum generally, is that he is completely correct in that assessment.

I think he expects that over time the truth will out, and he won't convince people through some intellectual lightning struck. He is also far more comfortable I expect with academic review than blogosphere review.

The truth is that I dont think Berri was really interested in trying to be the Bill James of basketball, as many are these days. He began I think just looking for a useful place to do sport econometrics as a way to analyze whether people act rationally. He was really just looking to create a performance metric to help him analyze decision making. Everything that came after was more than he expected I believe. It certainly isn't making him rich, and he is already a respected tenured professor who has organized the biggest Sports Economics Conference in the world the last two years. Rosenbaum himself said he couldn't carry Berri's jock academically. And its not like he is getting famous here. This is a little niche. My sense is that he enjoys it, but that actually a lot of his close colleagues think his blogging is taking away from far more profitable pursuits.

One last thing. A lot of people on message boards also evince a low personal opinion of Berri, which I think is ridiculous. From the reaction he incites you would honestly think he is some sort of evil warmonger. Yet another reason, I am more than happy to be on his side...

Anyway....

Posted by Flint | July 14, 2007 12:11 AM

"My impression is that the foundation of Berri's work is a rigorous regression analysis of the value of stats at the team level. This doesn't seem like a very subjective starting point."

I'm well aware that that is your impression. But, as previously stated, it's a rather clueless position.

Berri's "rigorous regression analysis" would support an infinite number of models. It doesn't somehow inevitably lead to the model that Berri has chosen based on his subjective choices.

"So basically, until it's disproven by more reasonable people than you, I will just enjoy it and what it has to say about my favorite players"

Feel free to enjoy it. As all individual stats hoops models are, it's a subjective system. If it dovetails with your subjective understanding of hoops, feel free to enjoy it.

Just don't confuse it with the pseudo-objective wrapping that Berri has tried to clothe it in.

And FWIW, it will never be "disproven". It can't be. Similarly, it will never be "proven". The very fact that you would use words like that are an indication of how fundamentally you misunderstand the issues here.

"From the reaction he incites you would honestly think he is some sort of evil warmonger."

He's not an evil warmonger. But he is a fraud who repeatedly misrepresents what he is doing.

If you don't understand why the blogosphere might have a problem with intellectual dishonesty, I don't know how to help you.

Or to put it another way, the problem isn't Berri's model. I think it's a rather primitive model that makes some bad choices, but all individual stats hoops models are incredibly imperfect.

The problem is Berri's lies and obfuscations in explaining what he's doing. That's what makes him worthy of nothing but scorn.

Posted by Petey | July 14, 2007 9:44 AM

Flint, thank you for the statements of respect. I will say again, that I've gotten frustrated (and I'm sure I've frustrated you), but I do appreciate you continuing this conversation for this long, because without you to argue for WoW I wouldn't have spent this much time trying to really follow the details of the WoW argument and doing so has really helped my understanding of linear weight ratings in general.

I will also add that, the last time we went through this my copy of Wages of Wins was in a storage box but, this time, I've pulled it out and will be referring to it.

That said, I'm going to be making the same argument that Petey makes above which is (1) there is nothing that requires you to divide credit between scorers and rebounders the way that Berri does and (2) If you look at the central problem as dividing credit within a team, you would never chose to do it the way that Berri does.

The Balkman and Nash examples don't necessarily get you that far, except they raise that question of distribution of credit within the team, and now I will tackle that in more detail.

To start, what I am talking about when I talk about distribution of credit? I use Berri's framework that there are two things that a basketball team does, acquire possession of the ball and then attempt to score the ball once they have possession. I think that's a very good way to think about it. I also think it's helpful to imagine that, within the team, two separate people are performing those two tasks. Taking the team as a whole, of course, there's no way to separate them. The team gets the ball, and attempts to score, and you tell how much the team as a whole

To use a, possibly dumb, analogy consider that you're eating hamburgers with a friend and start arguing about which has mroe calories, the hamburger or the bun. As long as you take the hamburger as a whole, you can tell how many calories there are in the hamburger total, but there is no way to tell how many of those calories come from the patty and how many from the bun without trying to separate them. If we could know either for sure, we could subtract that from the total and have a good estimate of the other, but without being able to separate them we're just speculating.

I can't actually tell you how many calories each has, but I will argue that Berri hasn't done anything to pull them appart, he's just looking at the hamburger and saying, "essentially" I know how many calories there are in the hamburger and I think that 95% of those calories come from the meat patty. And I think that, as far estimates go, that doesn't seem plausible.

So, why do I say that Berri isn't separating them. He's talked about the regressions he's done, and you've talked about that as the reason for distributing credit between scoring and rebounding the way that he does.

If you look at page 93, you can see that Berri examines the possibility of performing regressions against individual statistics and rejects it.

To see what he does instead, look at page 109, where Berri summarizes his method.

"1. Regress team wins on offensive efficiency -- points scored per possession employed -- and defensive efficiency -- points allowed per possession acquired"
"2. From this regression determine the value of a collection of statistics tabulated for the individual player and team, in terms of wins."

So the regression he does is a regression of team wins against team point differential (he is, essentially, replacing the pythagorean estimate of wins with a linear estimate).

He then calculates the contributions of varios stats to point differential.

You can see this in page 100 where he says (using X to avoid cluttering the quote with decimals), "From our analysis, each additional point a team scores, holding all else constant, increases wins by [X] . . . With these values in hand we can now estimate the value of a player's actions associated with scoring, acquiring, and maintaining possession of the ball."

You can then see that he uses the value of X as a basis for assigning value to scoring and possession statistics. He never uses regression analysis to try to separate the twp functions of scoring and possessions, he has just, as I said, run a regression of Wins against Point Differential at the team level.

So, the second part of my argument is that, knowing the value of a point scored you can divide credit between scoring and getting possessions however you want. I can make this argument in more detail, but I will start by saying that you can do that because I've done it.

So the final part of my argument is to say that, if you believe you can divide credit however you want between scoring and getting possession, that the way that Dave Berri does it is not how you would intuitvely do it. If on a team one person gets a rebound and another person scores a basket you and I might differ on how much credit we would assign to each of them but I don't think if either of us would come up with the method of distributing credit that Berri uses and, therefore, the obligation is on him (or people defending him) to explain why ditribute credit that way and, rather than defending the choice as a choice, he's presented it as a logical inevitability which I don't think it is.

How does Berri divide credit? We've been over that, he says that if a possession has an expected value of 1 point to a team, that the player gaining or losing possession of the ball is credited with that entire expected value, and the player who attempts to score the ball is only credited if they can do better than the expected value.

To use another analogy, which shouldn't be taken too literally, this is like Dave Berri saying that, the value that a pilot provides by flying an airplane from one point to another without crashing is equal to the chance that, on average, an airplane crashes, then looking at the statistics and saying that airplanes are safer than cars and that, therefore, a pilot isn't providing much value and that airplane pilots should be paid less than taxi drivers -- while ignoring that part of the reason airplanes rarely crash is because we spend a lot of money on training and compensating the people that fly them.

Berri says that the expected value of a possession is 1 point, gives no credit to the fact that teams have spent a lot of time and money finding, recruiting, and training people with the ability to turn possessions into points. Yes, performance needs to be compared to some baseline. You wouldn't say that any pilot who doesn't crash should be credited with saving the plane from disaster, and paid $30M per trip. Similarly you wouldn't say that the person who scores the basketball deserves all of the credit, some of that must be shared with the person who gets the rebound. But surely they deserve some credit for merely being able to perform up to average standards in a very demanding position, and berri does not do that.

The position adjustment serves as a proxy for this, to some extent, because the functions of scoring and rebounding are divided unequally between positions. But Berri uses the position adjustment in a way that conceals the problem.

It's as if he says "let's assume that, on average, taxi drivers should be be paid the same as airplane pilots, and we'll compare taxi drivers to other taxi drivers and pilots to other pilots and see how they compare to other people in their class." and then finds that the top taxi drivers complete a lot more trips than top pilots and, therefore, that even if the averages are equal that top taxi drivers are better than top pilots (the analogy breaks down a little here because the roles aren't separate and good big men score as well as good guards, but you can still see the analogy). In other words he's assuming away the problem of dividing credit between different actions/roles on the court, rather than addressing it directly.

That was way too long, but I'm agreeing with Petey 's last post completely. I think that believing that the problem of dividing credit between scoring and rebounding as something that Berri has solved mathematically is either (1) wrong or (2) assuming away one of the crucial questions.

Posted by NickS | July 14, 2007 9:52 AM

That last comment was too long, and I'm not sure the analogies helped so I'll summarize again. I made three points above which were (1) I'm willing to concede, for the sake of argument, that Berri's standard of value at the team level is correct. (2) It is entirely possible to take that same standard of value at the team level and divide credit among players differently. (3) Berri's decisions about dividing credit are highly ideosyncratic and there is nothing in his method that implies or requires that division.

That said, I wanted to mention something else that I came across reading through WoW that was one of those moments that just made me shake my head. Berri analyzes the 04-05 Chicago Bulls and comes to the conclusion that Ben Gordon was the worst player on the team. Predictably berri says that Gordon didn't shoot that well, turned the ball over too much, and was a below average rebounder, and that the bulls would have one almost one more game if they had replaced him with a bucket of warm spit (total Wins Produced, -.7).

Let me look at the scoring on that Chicago team. There isn't a standard metric that combines TS% with turnovers, so I'm going to show the shot attempts (FGA + .44FTA), and TOs for the top 7 players on the team, and then I calculate a percentage based on treating the TOs as additional missed shots. Obviously this will slightly understate the PGs since they will have more passing TOs than other players but I want to emphasize how bad that Chicago team was at scoring:

Hinrich 15.8 Shots, 2.6 TO, 16.1 pts, 44%
Chandler 7.0 Shots, 1.5 TO, 8.5 pts, 50%
Duhon 6.4 Shots, 1.5 TO, 5.9 pts, 37%
Gordon 14.4 Shots, 2.3 TO, 15.1 pts, 45%
Nocioni 8.6 Shots, 1.7 TO, 8.4 pts, 41%
Davis 6.7 Shots, 1.3 TO, 7.0 pts, 44%
Curry 13.8 Shots, 2.6 TO, 16.1 pts, 49%

Gordon had, by far, the highest usage of any player on that team and yet he was still the 3rd most efficient scorer, behind Chandler and Curry.

Who else should have been taking those shots on that team? Ben Gordon doesn't rebound enough, and does turn the ball over too much, but it seems weird to select that team as an exanmple to demonstrate the power of your analysis and then say that Ben Gordon was worse than useless on that team. It's moments like that which makes me think that Berri sometimes just takes pleasure in being a contrarian and provoking people.

Posted by NickS | July 14, 2007 12:51 PM

"Predictably berri says that Gordon didn't shoot that well, turned the ball over too much, and was a below average rebounder, and that the bulls would have one almost one more game if they had replaced him with a bucket of warm spit"

As stated, I think Berri's model is kinda lousy. It does particularly badly on measuring folks like Gordon and Iverson for pretty obvious reasons.

But, again, all models that try to assign a single number to players are going to do badly in many, many cases. The humble PER produces a number that more closely approximates reality than Berri's number does, but it's still laughably far off for a player like Bruce Bowen.

But again, my hostility to Berri does not lie with my problems with his model, but instead lies with the fraudulent nature of his claims about his model.

-----

"It's moments like that which makes me think that Berri sometimes just takes pleasure in being a contrarian and provoking people."

I believe Berri takes pleasure in getting publicity to sell more books and get himself published in the New York Times and advance his career, and that he's willing to play fast and loose with the truth to accomplish those aims.

I've got nothing against contrarians. I love Malcolm Gladwell, who on a certain level specializes in the same contrarian schtick as Berri. But Gladwell isn't willing to trade in intellectual dishonesty to generate his publicity. He's an honest contrarian. Berri is a fraudulent contrarian.

Posted by Petey | July 14, 2007 1:39 PM

But, again, all models that try to assign a single number to players are going to do badly in many, many cases. The humble PER produces a number that more closely approximates reality than Berri's number does, but it's still laughably far off for a player like Bruce Bowen.

Agreed, and I'm just saying that part of what irritates people about Berri is that he would choose an example in which his system does poorly as one of his highlighted examples. Hollinger would never use Bowen as an example of how his system was so much better than the conventional wisdom.

In fact, here is what Hollinger wrote about Bowen in his first prospectus (page 162):

"Bruce Bowen is the bane of stat-based rating systems. After watching him play it is clear that Bowen is a great defensive player. However, unlike a lot of great defenders, he doesn't pile up many steals or blocked shots. . . . [two pages of analysis] . . .His true value is probably closer to 13.00. Let's use that number for argument's sake. ... Bowen doesn't rate as a truly good player. He simply does too little on offense and the boards to rate as anything better than an ordinary substitute."

To be fair to Berri, after talking about why Gordon was a terrible player he does spend a page (p 153) defending his rating against the criticism that it's unfair to scorers, concluding

"Although our measure emphasizes the importance of possession factors, scorers still generally score high in Wins Produced. Gordon, though, did not, causing us to question how he received accolades from both members of the media and NBA coaches."

Uncharitably his defense is that because Kevin Garnett and Tim Duncan rank highly in his system and score a lot of points (never mind that their scoring is not the reason they rank well) that means that Gordon, who scors a lot of points, isn't underrated.

Posted by NickS | July 14, 2007 2:31 PM

I have one more question for Flint.

If I can convince you that it's possible to use Berri's assessment of value at the team level, but distribute value differently to individuals, I want to ask about your intuition about how value should be divided up.

Let's take Caron Butler as an example, he plays on team that's slightly above average when healthy, is himself slightle better than an average player in a number categories, but isn't outstanding at any one category other than steals.

He will, pesumably show up as an above average player by any ranking system.

His stats per game are (rounding off)

19 points on 7-15 shooting (4.5-5.2 from the line)
2.3 Orb, 5.3 Drb
3.7 assists, 2.9 to, 2.1 steals

His value will be the sum of:
Passing value (assists, and 40% of tos)
Scoring value (points and 60% of tos)
Rebounding Value (offensive and defensive rebounds)
Steals.

What fraction of his total value would you assign, just using your intuition, to each of those categories?

Do you think his scoring should account for 80% of his value as a player? 40%? 10%? Should his rebounding account for 80%? 40%?

Just eyeballing it, if you were to assign percentages to each of those categories that add up to 100% where do you think his value as a player comes from?

Posted by NickS | July 14, 2007 3:43 PM

Petey - Why would you call him a fraud? That''s just uncalled for. Berri is a tenured economics professor. For the last three years he has been the lead organizer of one of the largest Sports Economics Conferences in the world. Clearly his peers don't think he is a fraud. He has had work published in numerous academic journals, a distinction his most prominent academic critic has yet to achieve. He is currently collaborating on a number of papers with other economists. None of them apparently think he is a fraud. His work has been discussed by a number of eminently respectable public intellectuals like Gladwell and Yglesias without the whiff of a suggestion of intellectual dishonesty. Honestly, where do you get off saying this stuff? Is it all the money he is making off it? The appearance on Letterman? The young women breaking down his door?

And really, your criticism is once again vague and vituperative. You begin by insulting me, and then you continue with an argument which could be leveled against almost any use of model building to understand complex real world phenomena.

Basically, a la Colbert, you are dead to me...

NickS - A few things. First, those are good posts, the first one was a doozy though. Second, I can't volley everything back. When your posts are that long its well nigh impossible to give a fair response. Also, APBR is probably a better forum. Third, I don't think it's all that idiosyncratic to give credit to the player who collected a stat. That seems a reasonable move. Fourth, there are other systems, they may be less primitive, but they are also much harder to use and understand I think.

On Gordon, I dont think Berri is alone as a stathead who has a low evaluation of his offensive performance that year. WinShares has him as being worth .66 of a win offensively.

More generally, on the topic of scorers. A very common argument is that without the high usage, high volume scorers, many team's offenses would grind to a halt. I don't think this is so. Neither does Berri. He addressed this question in the following post.

http://dberri.wordpress.com/2007/03/05/creating-shots-in-philadelphia/

I think a lot of players in the NBA could increase their usage and scoring rates. However that is not the case with rebounding. It's much more difficult to improve at. It's relatively much easier to become a better scorer. And that in my mind, is why the argument that Bulls needed Gordon to score, so his value is understated etc, is incorrect. That said, as Berri would say, "scoring is a crucial part of the production process."

I think I also look at it as a two way street. The role players make it possible for scorers to score, as much as scorers score points and create space of role players to do their thing without having to touch the ball much. When David Lee went out, Eddy Curry's performance nosedived, because it became much easier to double team him.

David Lee is my favorite player. The knock on him is that although he was second in the league in TS%, he isn't a good offensive player. He doesn't have a jumpshot, can't drive, etc.

My view is that this guy had a TS% of 65%. If Eddy Curry were gone and they needed more scoring, David Lee could score. His Ts% might fall by a lot. Maybe to say 57%, or even to 53%, which is Zach Randolph's career rate I think. But he could up his scoring rate. It's probably much easier than most people think for David Lee to be a 20-10 guy.

Vis a vis Bowen, I agree with you and Hollinger. It's a point I made on the WOW blog. No model is perfect. Model are simple approximations of reality that aren't completely accurate but which often allow us to improve our decisionmaking on average a great deal. That's what they are. The fact that it doesn't work perfectly doesn't mean we wouldn't be much better off on average using the model than human observation and bias.

Which brings us back to Rashard Lewis, which is where we atarted, I think almost any stat based system will tell you that the Magic overpaid. If you look at Wallace's numbers, they look very similar. And he is three years younger. For 18.3 annually the Magic could (possibly) have had the services of Varejao at PF and Wallace at SF. And that would have been much more useful than what they have now.

At the very least they should have used Wallace as leverage against Lewis, or driven up the cost for the Bobcats, who have the makings of an extremely good team with Sean May returning. If Jason Richardson can play as well as he did in the second half of last year, after he recovered from his injury, they will be much much better than last year.

My 12 cents...

Posted by Flint | July 14, 2007 4:20 PM

When your posts are that long its well nigh impossible to give a fair response. Also, APBR is probably a better forum.

I understand. I am planning on editing eveything I've posted here and posting it to ABRMetrics when I get a chance. But I posted it here as a rough draft because there was some discussion going on and I hoped you would respond.

You're welcome to pick and chose things to respond to in my long post, but let me highlight one part of your response . . .

Third, I don't think it's all that idiosyncratic to give credit to the player who collected a stat.

Part of my point, however, is that you can use the results of Berri's regression on Team Efficiency Differential, while assigning different weights to the individual stats.

I'm happy to walk you through that argument in however much depth you want, but I want to start by just saying again that the relative points for scoring and rebounding aren't derived from regression. They're just how Berri has chosen to weight the elements that go into offensive and defensive efficiency.

I'm saying that I believe that I could construct a table of values for shots made and missed, offensive and defensive rebounds and turnovers (like the one on page 103) with different weights but constructed such that for any sequence of play between two teams, that the difference in the calculated values between the two teams would be the same in my system as it is in berri's system.

That is to say, again, that I could the total the same, and change the distribution of value.

Which brings me back to the question about Caron Butler. If you answer that I'll go through and calculate how both Wins Produced and PER distribute value and we can talk about both of those relative to your intuition.

As far as you comments about scorers, I actually largely agree with you. I think that the conventional wisdom (and the NBA efficiency rating) overrate high volume low efficiency scorers.

I could say more about Gordon, but really it's off the main topic that I wanted to explore, and I probably shouldn't have brought him up, but it just struck me when I saw the section on him because I remember, at the time, defending his low efficiency as still being higher than the rest of the team.

Posted by NickS | July 14, 2007 5:02 PM

"..you can use the results of Berri's regression on Team Efficiency Differential, while assigning different weights to the individual stats.

I'm happy to walk you through that argument in however much depth you want, but I want to start by just saying again that the relative points for scoring and rebounding aren't derived from regression. They're just how Berri has chosen to weight the elements that go into offensive and defensive efficiency."

I don't have the book in front of me. However, what you wrote above is not my understanding. So please, elaborate on this point, in whatever depth you please.

"As far as you comments about scorers, I actually largely agree with you. I think that the conventional wisdom (and the NBA efficiency rating) overrate high volume low efficiency scorers."

You are actually agreeing with Berri, and to be fair many other statheads. I think this is the main point. And its non trivial. Players who score are very often overrated and overpaid for that skill. It's very similar to batting average and obp in baseball in my mind. People overrate hits and underrate getting on base. Something to do with the glamour of hitting 300 I think...

Posted by Flint | July 14, 2007 5:15 PM

I should have said people use to overrate ba relative to obp, things have changed a bit...

Posted by Flint | July 14, 2007 5:17 PM

I don't have the book in front of me. However, what you wrote above is not my understanding. So please, elaborate on this point, in whatever depth you please.

Which part of that do want me to elaborate on? Constructing an alternate table of values (like the one on page 103 of WoW) or the fact that the relative weight given to scoring and rebounding in Berri's table is not derived from regression?

I can elaborate on either, but I'll expand a little on the second part of that. I would point you to the bit that I quoted from page 109, in which step 1 is the regression and step 2 is assigning values. I am saying that I could match his step 1, and do step 2 differently.

Posted by NickS | July 14, 2007 5:23 PM

I just saw that you don't have the book in front of you, so to save you reading through my massive comment again here is the relevent quote again.

""1. Regress team wins on offensive efficiency -- points scored per possession employed -- and defensive efficiency -- points allowed per possession acquired"
"2. From this regression determine the value of a collection of statistics tabulated for the individual player and team, in terms of wins."

and also

""From our analysis, each additional point a team scores, holding all else constant, increases wins by [X] . . . With these values in hand we can now estimate the value of a player's actions associated with scoring, acquiring, and maintaining possession of the ball."

I'm saying I can keep "those values in hand" and estimate different values for a player's actions.

Posted by NickS | July 14, 2007 5:30 PM

I'm saying that I believe that I could construct a table of values for shots made and missed, offensive and defensive rebounds and turnovers (like the one on page 103) with different weights but constructed such that for any sequence of play between two teams, that the difference in the calculated values between the two teams would be the same in my system as it is in berri's system

I now am with the book. I have read through the relevant pages and endnotes.The above is what I have an issue with. Sorry if I am a bit dense. Even if you could do that, and I will take you at your word that you can, what would be the point? The point of his exercise is to see if you can estimate wins using individual stats. In order for this to be possible, there has to be a real relationship between the stats and wins. Otherwise, at the end of the season, you will not be able to estimate with any degree of accuracy how many games a team won. Nor will you be able to forecast with any even reasonable degree of accuracy. He says that their is a close correlation, something he didn't find in football.

What exactly are you saying? Are the values he derived from regression not the correct ones? Are they arbitrary? Should the values be different for some reason? If you think different values would be better, why do you think that?

I am going to dinner, so I won't respond till tomorrow, sorry if you find me hopelessly confused....

Posted by Flint | July 14, 2007 8:42 PM

Starting from the bottom first.

sorry if you find me hopelessly confused....

Not at all, let me again express my sincere appreciation for your willingness to continue the conversation. If it weren't for you, I would be thinking about this as much as I am, and I enjoy thinking about it.

Also, I have had my moments of wanting to score points at the expense of WoW, but that's not as interesting to me as whether I can explain what I'm thinking to you.

At this point I have a bunch of ideas relating to Wages of Wins and they all make sense to me, and I think they all fit together, but I want to explain them to you as a way to test whether they do, in fact, fit together.

Which is just to say that I need you in this conversation. If I'm not making sense I'm hoping that you will tell me so I can try again.

What exactly are you saying? Are the values he derived from regression not the correct ones? Are they arbitrary? Should the values be different for some reason? If you think different values would be better, why do you think that?

I don't think the values are incorrect, I think are arbitrary. I have been thinking since May about whether I can figure out a non-arbitrary standard on which to base the values and I haven't thought of one. So, at this point, I think there are different values that I like better, but my first goal is just to argue that there are a range of possible values and that all are as "true" as each other. So then it's a matter of making an argument for one set of values over another. But, first, the goal is to think about what it means to change those values.

The point of his exercise is to see if you can estimate wins using individual stats. In order for this to be possible, there has to be a real relationship between the stats and wins

True enough. I want to argue that the relationship between stats and wins is a two stage relationship. The first step is a relationship between stats and point differential, and the second stage is between point differential and wins.

To see this, here are two quotes from Wages of Wins.

"From our analysis, each additional point a team scores, holding all else constant, increases wins by [X]."

"From our regression we learn that each additional point scored creates [X] wins. One might think that a shot from beyond the arc must be worth three times this value. Unfortunately, to take a shot you must use a field goal attempt . . . a possession employed is costs a team [X], and a field goal attempt is worth one possession. Consequently the net value of making a three point shot is worth [2X] wins."

I'm saying that the logic here is (1) a 3-point shot has a net value of 2 points (2) 1 point is worth [X] wins therefore (3) a 3-point shot is worth [2X] wins.

I'm saying that if you believed that a 3-point shot was worth 2.5 points instead of 2 points that you could still follow steps (2) and (3) to conclude that it was worth [2.5X] wins. So there would still be a relationship between the stat and wins.

Let me know if that makes sense, I have another idea for you to think about that I will post later tonight.

Posted by NickS | July 14, 2007 10:10 PM

I want you to know that I've been thinking all evening about this topic, and it's entertaining and gets complicated.

So, since I'm still working out the details myself I'll make this as simple as possible.

Here is a chart based on the Wins Produced values. I have multiplied everything by 30 to elimiate the decimals, and I'm creating two columns, one for the value that is applied to the offensive team for a given action and one that is applied to the defensive team for a given action, since many actions have a credit to both teams.

Off Def
FGM +1 -1
3PM +2 -2
FGX -1 0
OR +1 0
DR 0 +1
TO -1 +1

As you can see most actions are credited in both columns except a missed field goal and a rebound.

What if you looked at this and decided that it was unfair that the defense was penalized for allowing a made basket, but receives no credit for forcing a miss. Lets shift half of the value for a missed basket to the defense (and adjust offensive rebounds so that they are still the inverse of a missed basket).

This would get you the following table:

Off Def
FGM +1 -1
3PM +2 -2
FGX -.5 +.5
OR +.5 -.5
DR 0 +1
TO -1 +1

Obviously this will still preserve the relative score of the two teams. In this case both teams' score will be higher than it was before, the offense because it is being penalized less for missed baskets, the defense because it is getting credit for forcing missed baskets. But the difference between the credits of the two teams will remain the same, because we haven't changed the value of the action, we've only changed the accounting.

Now, I'm not saying that's an improvement, but it would make a significant difference in the player rating that would be produced because you would be turning a missed field goal from an event in which the entire value of the event was applied as a penalty to an individual player to one in which 1 player gets a (smaller) penalty and five players get a (much smaller) benefit. Since the defensive credit is divided among the entire defending team.

Would that be just as legitimate a way to rate players?

For what it's worth I'm tenatively saying that I believe it's just as legitimate a rating as the original Wins Produced, but I'm thinking about it.

Posted by NickS | July 15, 2007 2:05 AM

NickS - I think I understand your point. You could design a system where rebounding credit was divided amongst the players on the court in such a way that rebounding's effect was diminished, but the accounting system would maintain its integrity.

I understand your point. At the end of the day, But the fact you could design sych a system is one thing, and that's system's ability to measure a players value in wins is another. I dont think a system that is arbitrarily constructed will tell you the true value of a player. And I do believe such a thing exists, i.e. a player having worth independent of the players around him on the court. I think a lot people think of basketaball teams like a jigsaw puzzle, where all the pieces have to fit together just so, but I think its simpler than that.

Also, I think theory only takes you so far. I think over the course of the next few years we will get a better and better sense of whether Berri's system evaluates players better than conventional wisdom. If he keeps knocking it out of the park like he did with IVerson and Miller, I think he will win people over.

Sorry not to give you more, but I am exhausted from my weekend's activities...

Posted by Flint | July 15, 2007 10:07 PM

"I dont think a system that is arbitrarily constructed will tell you the true value of a player."

Glad to see that you're beginning to see the light.

"Why would you call him a fraud? That''s just uncalled for."

I'd call him a fraud because it's an accurate way of describing him. Same reason I'd call the sky blue. Saying impolite but true things is not uncalled for in this type of situation.

"Berri is a tenured economics professor."

I hate to break it to you, but tenured professors can behave in a fraudulent manner. So can accredited doctors, lawyers, and indian chiefs.

Credentials really aren't a guarantee of honorable behavior. Sorry if I'm disillusioning you in any way.

"His work has been discussed by a number of eminently respectable public intellectuals like Gladwell and Yglesias without the whiff of a suggestion of intellectual dishonesty."

You're dead wrong about that. You just haven't been paying very close attention to the debate.

"Basically, a la Colbert, you are dead to me..."

Cheers.

If you're a smart fellow, and if you continue to pay attention, you'll eventually get to where I am. Hell, you seem to have already accepted that an arbitrarily constructed system won't give you the "true" value of players.

I understand that this stuff is difficult to parse for the nescient, but if you approach if with your brain rather than your gut, you might be able to figure it out. Berri's work has an excellent "truthiness" quotient. You just need to move beyond truthiness to actual truth in order to see the problems in his work and the fraud in how he presents it.

Posted by Petey | July 15, 2007 10:24 PM

"I have been thinking since May about whether I can figure out a non-arbitrary standard on which to base the values and I haven't thought of one."

A non-arbitrary standard is not possible. Simple as that.

There just isn't enough nearly information in individual hoops stats to accurately represent the finer points of individual players' contributions to the whole. And given the nature of the game, better stats would only ameliorate, rather than solve the problem.

Certain well thought out models are going to better approach reality than other poorly thought out models, but no single model will ever "best" approach reality. Or put another way, there is no non-arbitrary standard to be found, no matter how long you search.

It's actually a very similar scenario to economics models that attempt to represent the entire economy. No economist can tell you with reliability that a 33% top marginal tax rate is abstractly "better" than a 35% top marginal rate, although they can certainly tell you that a 98% or 2% top rate will both cause some problems.

Posted by Petey | July 15, 2007 10:42 PM

A non-arbitrary standard is not possible. Simple as that.

Yes and no. Any model is a simplification, and a linear weight system is necessarily a huge simplification. So you are always going to be abstracting out huge chunks of the game. The best you can hope for there is to make good decisions about what you're abstracting out and be aware of that when you try to use the stat.

But it is certainly possible for elements of a model to be non-arbitrary. See here for an example of trying to build non-arbitrary standards. That doesn't get you to a player rating system, but it does argue for certain stanards for estimating possession, or the relative weight of offensive and defensive rebounds.

That's really what I would mean by "non-arbitrary."

Flint -- I have a version of this argument worked out that I'm going to post to APBRmetrics soon, and I invite you to post there. I'm also happy to continue this discussion here.

let me just respond to one part of what you say, because you're getting ahead of me here.

"I dont think a system that is arbitrarily constructed will tell you the true value of a player. And I do believe such a thing exists, i.e. a player having worth independent of the players around him on the court."

You're correct, that I will, at some point try to argue that a different set of weights could be preferable to Berri's. But, at this point, all I'm trying to argue is that Berri's weights are, themselves, arbitrary.

You try to contrast something that I would propose with Berri's system by describing a different proposal as "arbitrary" and implying that Berri's are an estimate of "true value."

I'm just saying that we can agree or disagree about whether Berri's has a good estimate of "true balue" but I want to show that his estimate is, in some sense, just a result of choices that he makes because they look good to him. There is nothing that I can find in his model or in the math that he describes that forces someone to use his weights.

If that's true, and I believe it is, than Berri is one the same level as anyone else who takes his starting point of estimating wins based on point differential, and has a set of weights that will add up to a team point differential. In that case, Berri has his proposal and it's up to him, or people like you who like the number's he's chosen to argue for why they are better, but there's no reason to presume that Berri's weights are better than anyone else's.

Posted by NickS | July 16, 2007 12:20 AM

"But it is certainly possible for elements of a model to be non-arbitrary."

No doubt.

"That doesn't get you to a player rating system"

Again, no doubt.

-----

I guess my larger point is that while we're in total agreement that Berri's system relies on arbitrary decisions, (or perhaps more accurately to my way of thinking, subjective decisions), I'm asserting that there is no way for anyone to come up with a player rating system that doesn't rely on subjective decisions.

While there is plenty of non-arbitrary work that statistics can perform in providing a better view at what happens on the court, any attempt to synthesize those stats into a rating of an individual player's worth will always remain an art, not a science.

The formulas in the paper you link to may be the brushes and paint of that art, but it is still an art.

-----

"I'm just saying that we can agree or disagree about whether Berri's has a good estimate of "true value" but I want to show that his estimate is, in some sense, just a result of choices that he makes because they look good to him. There is nothing that I can find in his model or in the math that he describes that forces someone to use his weights."

The core of what I find offensive about Berri is the lengths he is willing to go to obscure this rather simple point.

Much of the rhetoric he has used to popularize himself is based on strongly implying the very opposite of this rather simple point.

Posted by Petey | July 16, 2007 3:59 AM

"The core of what I find offensive about Berri is the lengths he is willing to go to obscure this rather simple point."

Or to put it another way, how did Flint become so convinced that Berri arrived at his weightings through some non-subjective process?

I'd argue that Berri has worked very hard to convince his readers of that falsehood, and therein lies the intellectual dishonesty and fraudulent behavior.

Posted by Petey | July 16, 2007 4:09 AM

It took me longer than I thought it would, but I've posted a more worked out version of this question at APBRMetrics.

Posted by NickS | July 17, 2007 8:25 PM

Rashard Lewis Revisited

Comments (112)

Search

Books by Matthew Yglesias

Recent Comments

Recent Posts

Weekly Archives

Categories