Attacking the Center: Ari Caroline Reveals the NBA’s Blind Spot

We’re happy to have more amazing work from Ari Caroline (@aricaroline), Director of Quantitative Analysis and Strategic Initiatives at Memorial Sloan-Kettering Cancer Center. Today he’s bringing his awesome expertise to show the NBA’s blindside when it comes to centers.

As it should be! via Keith Coward

We love to criticize front offices, players, coaches, etc. around here. Understand we do this as hyper-critical fans. Ari Caroline has been using the data to do some great work. One of his key points — for the most part teams play the right players! That’s good news (well for teams other than the Nuggets) However, teams do make mistakes and Ari wanted to shed some light on that.

First, let’s check out how NBA teams allocate minutes for centers.

2012 NBA Center Distribution. Click for larger image.

Drilling down into the individual position of center we see that the same trend we saw league wide. With statistical certainty, if you’re a better center according to Wins Produced, you’re likely to get more minutes. In the lockout shortened season it works out such that each increase in WP48 of around 0.010 gains you an extra 189 minutes. Go NBA, right? Ari decided to highlight something important though.

The highlighted players are really good or really bad centers that break the mold. In essence, they should be getting more or fewer minutes and they’re not! (Seriously, this is one of the coolest images I’ve ever seen!) In case you didn’t click on the image, the players are

Editor’s note: I’m content to say Varejao (injury) and Ben Wallace (age) may be acceptable. However, there’s no real excuse for the rest of the names on the list.

That’s the NBA’s blind spot for centers, right there! These are the players that every NBA should be looking at (to play much more or bench permanently) Doing so can change a team’s fortune. It has a noticeable effect in fact.

Removing our outliers changes the trend. It still stays significant, and we can still give our NBA teams a pat on the back. Instead of 189 minutes per 0.010 WP48 it now jumps to 350 minutes per 0.010 WP48! The NBA could shut down a lot of criticism by just realizing they’re messing up on 5-10 players (if you believe a few names above have justifiable reasons for their low or high minutes)

Of course, there’s more than one position and thus more than one area to improve. Don’t worry! Ari promises he has more coming and I can’t wait.

-Dre

140 Responses to "Attacking the Center: Ari Caroline Reveals the NBA’s Blind Spot"

reservoirgod says:

September 4, 2012 at 1:16 am

I think there are good reasons for all of the productive centers seeing limited minutes. The biggest reason seems to be fouls. Only 6 players qualified for the mpg leaderboard averaging 4+ fouls per 36 mins: http://bkref.com/tiny/4tiIT.

Additional reasons:
Birdman: Old & injury prone. He can’t play major minutes on those knees anymore.
Koufos: On the same team as Birdman, then JaVale McGee.
Asik: Splits time with Joakim Noah
Splitter: Splits time with productive frontcourt players that foul less like Tim Duncan and Bonner.
Stiemsma: No explanation needed outside fouls, but also splits time w/ Garnett.
Ty Hicks says:

September 4, 2012 at 4:11 am

I would also add a note to Tiago Splitter, Omer Asik, and Greg Stiemsma, that they do(or did) play behind Tim Duncan, Joakim Noah, and Kevin Garnett. That only leaves Koufos and Chris Anderson as underplayed good players. Who did they play for again?
Mark Cuban says:

September 4, 2012 at 8:23 am

Im not a huge fan of WP for the NBA, but its still fun to read the craziness you guys post.

has it dawned on you guys that other than obvious superstars, the “best ” and “underused” players all tend to look pretty much the same for every position ?

Maybe you should start looking at WP for lineups ?

Coaching matters. Who you play with matters. Who you play against matters.
Andres Alvarez says:

September 4, 2012 at 9:26 am

Mark,
Great to hear from you again. To promote our own stuff – great videocast with VC Chris Yeh (http://wagesofwins.com/2012/08/30/wages-of-wins-videocast-chris-yeh-defends-the-magic/). We talk about this a bit on it. Arturo also has a great post on how complimentary players improve each other (I’ve bugged him for a link.) It’s not that we don’t value lineups. It’s that question one is: do you have good players? After that you can look at lineups etc.

I will say the following – http://jbrocato.wordpress.com/2011/05/27/the-best-to-never-win-it-all-and-why-im-rooting-for-the-mavs/ When you guys won it all, you had the best collection of Wins Produced players without a title.

This season you have a terrible collection of Wins Produced players (Sorry man, Dirk is old)

I believe there is value in lineups and systems, etc. However, you can’t get around the basic problems: you need good players to compete and you can’t give bad players major minutes. We’re trying to help with that and I believe Ari does a great job showing some of the problems. And yes, you should be worried about Kaman.

Thanks for indulging our craziness!
Ari Caroline says:

September 4, 2012 at 9:55 am

Mark,

I’ll touch on this a bit for my next post on guards. Coaches seem to recognize value in point guards vs shooting guards very differently. They are also the only two positions that are potentially interchangeable defensively without running into major matchup problems. You’ll have to wait a bit for that post, though. We’re still trying to cure cancer here.

Also, I’ve mentioned that it’s really easy to customize these charts so if you’ve got a couple of lineups that you’d like to compare using these visuals (or other visuals for that matter), send them to me and I’d be happy to run them.

Ari
Ari Caroline says:

September 4, 2012 at 10:10 am

Reservoir,

Absolutely, there is good reason that a lot of these centers were underplayed. The point of the visual is to tease out value that might be underutilized, even if it is for good reason. Houston recognized this in laying out for Asik and, perhaps, the Bulls were right not to match given that he had Noah in front of him in Chicago.

You’ll see some similar themes in the upcoming guard post. The Spurs (who are just bleeding brilliant), traded a very productive but underplayed George Hill to the Pacers where he’s shined. The Spurs realized that he never could get those minutes behind Parker and Ginobli so they waved their wand and turned him into Kawhi Leonard.

Ari
joshweil says:

September 4, 2012 at 10:43 am

@Mark

What kind of evidence would convince you that Wins Produced was a useful model for analyzing the NBA?
Neil Paine says:

September 4, 2012 at 12:26 pm

In addition to fouls, it seems to me that you can predict who will be on your list based on usage & the % of shots taken close to the rim:

Anderson Varejao – 16.9% Usg; 66% of FGA in paint
Chris Andersen – 14.8% Usg; 62% of FGA in paint
Kosta Koufos – 14.0% Usg; 89% of FGA in paint
Omer Asik – 12.6% Usg; 88% of FGA in paint
Ben Wallace – 6.6% (!!!!) Usg; 59% of FGA in paint
Tiago Splitter – 20.7% Usg; 79% of FGA in paint
Greg Stiemsma – 10.8% Usg; 41% of FGA in paint

Byron Mullens (bad) – 22.5% Usg; 29% of FGA in paint
Glen Davis (bad) – 22.4% Usg; 38% of FGA in paint
Chris Kaman (bad) – 26.6% Usg; 39% of FGA in paint

With few exceptions, the players on the “good” list have very low usages and rarely shoot outside a few feet of the basket, while the guys on the “bad” list have high usages and shoot lots of jumpers.

You could say the first group’s playing style is always preferable, but as Mark suggests, there are cases where a player from the second group would “fit” a unit better.
Chris Yeh says:

September 4, 2012 at 12:39 pm

Perhaps it makes sense to point Mark to Dre’s analysis of the 2011 Finals:
http://wagesofwins.com/2011/06/15/dont-blink-a-review-of-the-nba-finals/

Dre’s WP-based analysis shows that Kidd and Chandler played well, while Dirk showed some drop-off (perhaps due to injury), and that the key to victory was the unexpectedly fine play of Jason Terry and DeShawn Stevenson.

The tough part to accept is that despite the evidence of the highlight reels, Dirk was only the 5th most-important Mav in the Finals. In his defense, he had been the second-best Mav (behind Kidd) heading into the Finals, and he did suffer a serious injury.
Mark Cuban says:

September 4, 2012 at 1:02 pm

Dre makes my point. How many wide open threes do you think dsteve, jet or jkidd get without diek on the floor.

Jet was the only one who played any meaningfully minutes w out dirk. To say dirk was our fifth best player doesn’t pass the most basic of smell tests.
Who should have taken dirks minutes?

One of the huge failures of win share is that there is no context. Is the first basket as important as the last in a close game? A good team vs a bad team all count the same? Why the variance in playoffs?

Look at the win shares of our centers over the past. Seems to me they always got better and we’re considered way above average. Damp, Tyson, haywood, bwright all got better. Bwright is a star, the other three were 10game winners and Tyson more.

Are we that good with our centers or do we just use them differently. Or do we recognize that a good player is only as good as they contribute to a team game

Should I ask coach to consider giving moat of dirks minutes to bwright?
jbrett says:

September 4, 2012 at 1:54 pm

Mark,
Perhaps you could take a page from an in-state rival–find a few games to list Dirk as ‘DNP-Old.’ (Truth is a perfect defense.) ‘Rest’ him for the playoffs. Cut his minutes to 28 a game, and give Wright the other 20–plus a few at SF, or even C if it keeps Kaman off the court.

On the other hand, I see an opportunity for a twisted win-win scenario–play Kaman big minutes. If your argument has merit, he’ll become an above-average WP guy, and prove we’re all oversimplifying. If you’re wrong, and he’s just as bad as we think, you’ll at least get a decent shot at next year’s overall #1 pick!
Ari Caroline says:

September 4, 2012 at 1:58 pm

Neil, When would it ever make sense to ask a poor shooting center to leave the paint and shoot jumpers?

Mark’s argument is more convincing. A big (talented) body in the middle attracts defenders and frees up the real jump shooters. Admittedly, that would not be captured by WP (and presumably would be by Prof. Albright’s +/- regression model).

With that said, the best argument for WP is probably the Churchill argument. It’s the worst model except for all the rest. Despite having some serious gaps (that are probably fixable). WP is far more predictive of actual outcomes than the closest competitor.
Neil Paine says:

September 4, 2012 at 2:10 pm

As the token representative for Basketball-Reference here, I should note that Win Shares and Wins Produced are not the same stat. The correlation between Win Shares and Regularized Adjusted Plus/Minus (which, for all of its noise on an individual basis, does begin to separate players by their influence on the game when taken in the aggregate) is 0.531, far greater than the correlation for Wins Produced (0.380).

Having said that, I think Wright is a great example of box score stats overvaluing certain players. Box score numbers can really only paint a limited picture of how good a player was within a very specific role — to say nothing of how valuable he’d be if moved to a different role. For most NBA players, their value at any given time is almost completely dependent on who the other players are that surround them, a fact that calls into question the usefulness of many single-number metrics.
Steven says:

September 4, 2012 at 2:11 pm

Ari, when you asked, “When would it ever make sense to ask a poor shooting center to leave the paint and shoot jumpers?” Isn’t that just confirming Mark’s point (or was that the point?).

ie. Kaman was playing above average early in his career when his usg was < 20%.
Ty Hicks says:

September 4, 2012 at 2:13 pm

I think valuing performance against good teams and bad teams evenly is fine as long as teams are playing a balanced schedule. I suspect a bigger problem would be non-starters who play a lot of minutes against other team’s back ups.

Dirk has always struck me as a player who’s effect on the game is difficult to model. You just don’t see guys as big as him be able to play that effectively so far from the basket. Has anyone done a study on what happens to the rebounding rate of the player guarding Dirk?
Neil Paine says:

September 4, 2012 at 2:14 pm

@Ari – Well, let’s see, what if you need to open up space down low by luring the opposing big man away from the paint? Who better to do that than another big man who can knock down enough jumpers to keep his defender honest?

Perhaps a player like that would prefer to play closer to the rim and shoot easier, higher-efficiency shots, but that’s not his role on the team. In order to make his teammates more efficient, he’s sacrificing his efficiency stats. Now, maybe that’s true or maybe he just loves the J and is being selfish. But from the box score we can never know which is true.
Andres Alvarez says:

September 4, 2012 at 2:14 pm

Neil,
We do enjoy your representation. As a bit of curiosity, why are you giving a stat called WIN Shares, which purports to explain wins (and a stat we’ve well received around here) a test as to how it explains Adjusted Plus Minus, which by all measures does a poor job explaining wins and is very noisy? (Also, my understanding was Win Shares was Justin’s work. I also recall at Sloan you mentioned you weren’t a huge fan of it.)

As to your other things. Those are good theories and hypothesis. However, without being tested for significance or impact, they remain just that. It’s boring I know but I stick behind basic point that Wins Produced does explain wins and by virtue of the fact that basketball players are more consistent year to year than most sports has a good year to year correlation.
dberri says:

September 4, 2012 at 2:25 pm

just a general comment…
one does not test one model by regressing it on another model. If you take such an approach, which model are you testing? And how do you know?
Ty Hicks says:

September 4, 2012 at 2:29 pm

How about you do a Dre(or Arturo) vs Neil podcast? That would be fascinating. I also wouldn’t complain if Mark wanted to do it. He seems like a bright guy. :-)
Ari Caroline says:

September 4, 2012 at 2:57 pm

In Neil’s defense (correct me if I’m off, Neil), I believe his intent in comparing Win Shares with Adjusted Plus Minus was relevant because the Mavs, from what I’ve understood, have already bought in to the Adjusted Plus Minus model. It wasn’t meant as a claim of inherent superiority for the Wins Share model.

Is this correct, Neil, or did I misinterpret?
Chris Yeh says:

September 4, 2012 at 3:17 pm

@Mark,

I’ve written (an as of yet unpublished) post about what I call the Magic Johnson effect, which compares the 1991 and 1992 Lakers, who had to suddenly deal with Magic’s retirement during the preseason.

The details will come out when Dre publishes the post, but essentially the numbers suggest that Magic had an effect outside the box score.

The drop from Magic to his replacement (the immortal Sedale Threatt) was about 12.5 wins, but the Lakers actually dropped 21 wins in terms of WP. Without Magic, the performance of both James Worthy and Vlade Divac fell off a cliff.

It would be really fascinating if we could look at WP numbers based on the number of minutes played by a particular lineup. It would be great to see if over time, Jet’s WP48 was significantly higher when he was in the lineup with Dirk, versus his Dirk-less minutes.
Andres Alvarez says:

September 4, 2012 at 3:27 pm

Chris,
You get full apologies from the editor on neglecting your piece and full promises that it will go up this week :)
Arturo says:

September 4, 2012 at 4:09 pm

This thread is pretty surreal. Let’s jump right in :-)
Neil,
There is a reason High usage bigs who shoot inefficiently from outside are condemmed by the model. It’s bad basketball. Throw in below average at everything? Yeah, not good. I will argue that not doing a bad thing is as important as doing good things.

Mark,
I love Dirk and his ridiculous efficiency regardless of usage (shown for 2011). High usage,high efficiency players will have a positive impact on other players (See James, Lebron). However, Dirk’s numbers were off last year for him particularly rebounding. Where I in the room, I might have argued for giving him some run at the three particularly in the regular season and teaming him with Wright and Haywood. Given that Miami did something similar with Lebron by turning him into a PF or Center this could have been successful. Still could be.

I agree, coaching does matter. You can’t really change who a player is but you can put him in a better position to succeed (see the Spurs picking up retreads and optimizing their value). You can do the opposite to the opposition (see turning Lebron into a jump shooter in the 2011 Finals). Most importantly the coach controls the minutes and that’s the biggest variable there is in basketball.
Ari Caroline says:

September 4, 2012 at 5:01 pm

Correction: I mistakenly gave Chris Albright the credit for the +/- model created for the Mavs. It was Wayne Winston, who co-authored several texts with Albright including Practical Management Science which maintains a prominent place on my home office shelf.

I have nothing but respect for Prof. Winston and would love to see his model at work. I have no doubt that it outperforms many of the less-predictive published +/- models.
Mark Cuban says:

September 4, 2012 at 6:48 pm

As far as Waynes model, its been amazing at predicting “good players” from NBA players who have been in the league for 3 or more years. It doesnt predict every player that is good, but the ones it says are good are always good.
Its also very good at helping us determine which lineups win games. Of course it requires some data, so with lineup changes we will have this year, it wont really kick in immediately, but it works.

You guys are like the hedge funds that always back test their data and determine that they can beat the market. Markets adjust and so do teams.
For goodness sakes, 1 yr you had Damp, who I liked having on our team as a .238 and Dirk a .74. Is there anyone on the planet who trade dirk for damp ?
And BTW, it wasnt just 1 season Damp out did Dirk, it was 3 consecutive years.

1 question. You had dirk as a below average player for 3 of last 4 yrs.. WHy is every good coach in the league stupid enough to double a below average player all 4 years? -
Andres Alvarez says:

September 4, 2012 at 7:01 pm

Mark,
I will stress there is one key aspect of the WP model. We explain with the stats how each player contributed. This is based on the rules of basketball and the very simple model of point differential (how many points did you score vs. how many points your opponent scored) Yes in 03-04 when Dirk had a down scoring year we ranked Dampier higher – http://www.thenbageek.com/players/compare?career=1&direction=asc&player_ids%5B%5D=63&player_ids%5B%5D=375&sort=name&utf8=%E2%9C%93

I’d say the following. If your goal is to buy and sell players on the NBA market, then yes you present a good case. If your goal is to win games, then it makes sense to look at the things that win games. You’re a very smart guy that knows there are flaws in markets. It’s not enough to sell me that conventional wisdom is always right. Also, we’re not saying Dirk isn’t a good scorer. Quite the contrary. We’re saying when you look at the required production from an NBA PF that Dirk’s total contribution is below average. And a big part of that is he’s an older player and those tend to decline.
Mark Cuban says:

September 4, 2012 at 7:25 pm

Just a simple question. Why do most coaches continue to double dirk and why have they done so for the past however many years if he is in most years a below average player .

Has nothing to do with conventional wisdom. Coaches watch film. THeir scouts watch film. Players watch film. You ever see anyone raise a stink about dirk not deserving to be doubled ?

Why would a good coach of a team double a below average player consistently ? I know I cant think of players the Mavs have doubled over the years that I thought were below average, the exception being someone who cant pass out of a double team. One of your most highly rated centers being one. Bad decisions turnovers

and FWIW, dirk would tell you he had a sub par year last year. No prep time ..

I can also tell you that in our planning, we see players playing into their mid 40s starting in about 15 years. So the guys coming in now, could play 20 plus years if they work at their games and follow our health and medical regimens
Mark Cuban says:

September 4, 2012 at 7:33 pm

my last reply didnt include the not equal symbol so it should read

Bad decisions do not equal turnovers.

adding: that you can force someone to make a bad pass to someone in a position where he cant take a good shot, negatively impacting that players efficiency . A player does that enough , it has an impact on the game. His minutes get reduced if that coach wants to win .

One of your highest rated centers is someone who falls into that category. You double team him because he has no clue what to do. He wont turn it over, but you can bet he someone will be heaving the ball up against the shot clock

and in all candor, Chris Kaman used to have that problem. He was asked to do too much as a passer. He would get double teamed and didnt make great decisions. We wont have the same requirements and it will be far tougher to double team him with most of our line ups.

System matters
Vivek says:

September 4, 2012 at 8:06 pm

@Mark

Little less about the debate, but more about something else I’ve always wondered. As a Nets fan, I witnessed the Mavs completely rob us of Jason Kidd; the deal was terrible for us, and Dre loves to remind me about it…..
How did you manage to pull that off? You basically traded Harris (an average PG) for one of the best players in the game.
Mark Cuban says:

September 4, 2012 at 8:16 pm

Vivek
We knew that Jkidd would get better because of the players the Mavs put around him. And he would make the other players (other than Dirk, We are fans of WP and knew that dirk was a below average player and would probably get even worse with JKidd) much better as well.

It was worth the price to pay.
Ari Caroline says:

September 4, 2012 at 8:18 pm

@Dre

Might it also be possible that Dirk is penalized for lower rebounds and shooting % because he is misclassified as a PF (yes, I know how tall he is)? His game in recent years is much closer to that of a swing forward and I could easily see the Mavs playing him with two other bigs if they had them. Defending a young, agile SF might be an issue, though.
Arturo says:

September 4, 2012 at 8:24 pm

@Mark,
Systems can matter but players (like employees in any other field) tend to be who they are. Chris Kaman has not been a productive center. You’re banking on being able to alter his trayectory and we’re saying it’s an unlikely proposition.

Haywood and Chandler where both productive before you got them on your roster.

He’s also been injury prone.Based on the evidence at hand He’s a high risk move.

I will cede that an $8 million gamble on size, given the fact that mavs no salary commitments in 2014/15 may just be the right play. Need to hit some long odd plays to make it happen.
joshweil says:

September 4, 2012 at 8:25 pm

@Mark

For the past three seasons Dirk was an above average offensive scoring threat but below average rebounder. That’s the short answer to your question how could a player be rated below average yet still draw double teams.
Vivek says:

September 4, 2012 at 8:33 pm

Mark,
To be fair, the players New Jersey and Phoenix put around Kidd also made Kidd a very productive point guard. Kidd is definitely a player that WP has consistently complemented, but your savvy in getting him was incredible; it was one of the best trades of the last 5 years (the Nets can testify), and helped you get your ring. You must’ve known that you were completely ripping New Jersey off, and I respect you as one of the best informed owners in the league.
However with Dirk, it’s not like here at WoW we’ve claimed Dirk has always been below average. He draws double teams because he can still score. But he rebounds poorly for a power forward and struggles defensively. In 2007, Dirk was one of the best PFs in the game; now at 34, he isn’t.
Mark Cuban says:

September 4, 2012 at 8:37 pm

Joshweil, he must be a really, really bad rebounder . So far below average that he canbe a top 5 offensive threat and on balance be a below average player

but if thats what the numbers say. THat is what they say.

Of course , the fact that we play more zone, and more different types of zone than anyone in the NBA might have something to do with his rebounding numbers as well.

And of course it would be inconceivable that his rebound rates would vary depending on who he was on the court with. But we dont need to track that because good players are just good players. Average players are just average players, even if they are just a little bit better at some things than others
Jon C says:

September 4, 2012 at 8:41 pm

This thread has certainly moved past this point, but I’d like to chime in about using WP model as a basis for analyzing a specific NBA playoff series. Obviously, we have the small sample size issue. But that isn’t what I want to address.

The WP model, as I understand it, is based on taking a regression of actual NBA wins in comparison to actual NBA stats. It is basically a model based on all the teams playing all the other teams. So it smooths out match ups. However, in any particular series we have team A against team B. Maybe a model based only on “hundreds” of games just between those teams would have some subtly different weighted values. Maybe in A versus B, offensive rebounds aren’t so valuable because both teams have offensive rebounders who tend to rebound and kick out for a rest versus rebounders who score put backs.

For a particular NBA team, a specific WP model could be computed and it might show something different than the generic league wide model.
Mark Cuban says:

September 4, 2012 at 8:41 pm

Vivek, in 2007 dirk was a top pf, but in 2008, he suddenly lost it ? Then in the playoffs last year he suddenly gets it back ?
Vivek says:

September 4, 2012 at 9:06 pm

Mark, in the playoffs last year, your team had Shawn Marion, Tyson Chandler, and Jason Kidd. All 3 of those guys were playing at a star level, in addition to Dirk. Yes, Dirk was good, but I don’t think it’s fair to attribute him as the one main reason as to why that championship was yours. The Dallas Mavericks have done a great job in getting players that are on good deals and are star players, and that’s why the media loves to say “LeBron choked, that’s why Dallas won (Skip Bayless)” or things about how the Mavs won with only one player leading the charge.
And the sudden drop off in game actually happens enough with older players, they start shooting worse (mysteriously happened to Deron last year, but I think you knew about that too) and still continue to shoot upwards of 20 shots per 48 minutes a game. That is detrimental to a team. Dirk can still be a star player, no doubt, but his game is so relient on the mid range jumper which is inefficient. He’s over 30, so it becomes tough.
Jon C says:

September 4, 2012 at 9:14 pm

Well is it fair to say that a model that predicts wins based on stats compiled by all the teams in the league should necessarily predict wins based on a fairly unique player like Dirk? The WP model is based on hundreds of games that don’t involve any players that are remotely similar to Dirk. So it might not be a great predictor for Dirk. But you have to admit that seven footers who can survive on mid to long range two pointers are pretty darn unique. Actually anyone who can survive on long range two pointers in the NBA is pretty darn unique.

Generally speaking, I suspect, the WP model is very effective. The Churchill quote above might be right on the money. But unique players who do something very different than the vast majority of players who form the basis of the model, might not be fairly graded one way or another by the model.
joshweil says:

September 4, 2012 at 9:55 pm

Last year Dirk shot below his lifetime average. At 56.4 TS%, he’s not a top 5 offensive converter.

You’re right all those thing vary.
Ty Hicks says:

September 5, 2012 at 2:03 am

Dirk, for his career, is a slightly below average rebounder. However, that low average comes entirely from his poor offensive rebounding. His defensive rebounding is actually well above average. The model, rightly in my opinion, gives more credit to offensive than defensive rebounds. In Dirk’s case however, that might not be entirely fair. You would expect Dirk to get fewer offensive rebounds because he plays much further from the basket than an average PF. But, his defender also has to play much further from the basket. So, I suspect, Dirk’s net effect on his team’s offensive rebounding is more complex, and possibly beneficial, than it might first appear, and probably can’t be easily modeled.
Charles Connaughton says:

September 5, 2012 at 3:24 am

It’s not as one dimensional as ‘double team the other guy’s best player’. WP really likes Tyson Chandler, but no one in their right mind is going to double team him. Players can be good for many reasons, but they only draw double teams when they have particular offensive skills.

I do think that WP48 undervalues volume scorers somewhat; turnovers increase with usage as a rule, and while point guards often have big assist totals to compensate a shooter will not. Given two perfectly average shooters, WP48 will tend to prefer the one who shoots less – as he will touch the ball less, and thus turn it over less.

There are definitely second order effects from players making their teammates better, and team composition in general. There’s no question that’s important; and with the right tools, I don’t doubt that you could tease those out. The point isn’t that WP48 is a perfect model – there are no perfect models – but that it’s remarkably good for a low data, first pass approximation.
wiLQ says:

September 5, 2012 at 6:22 am

@Mark,
“Just a simple question. Why do most coaches continue to double dirk and why have they done so for the past however many years if he is in most years a below average player”

WP or WS doesn’t explain anything in this regard because it’s a total metric and OFFENSIVELY Dirk is very good and he has a history of making shots from anywhere on the court [he is above average on threes, close to the basket and a rare player with very efficient long twos[!]]. Plus he also is very good at drawing fouls so that’s why he needs to be doubled.
His TOTAL value has slipped because of his defensive numbers which have nothing to do with the decision whenever double-team him or not.
Mark Cuban says:

September 5, 2012 at 8:19 am

Well put Charles.

WilQ, my point is that in order for his net number to place him below average on a continuous basis, his defensive and rebounding deficiencies have to be so bad that he genuinely hurts his team in those areas more than he benefits from offensive abilities when compared to an average player. Thats ridiculous on all fronts.

Charles has it right. its a nice piece of information. its interesting, but its not something that an NBA team can use to build a team.

There are statistical tools that can help teams win games. That help us help our players improve their games. That is what we focus on
Chris Yeh says:

September 5, 2012 at 8:36 am

First, let me point out how amazingly cool it is that Mark, the NBA’s most thoughtful and progressive owner, is actually spending his time debating the merits of various models with the readers of a blog aimed at sports economists and stat heads.

The Mavericks and their fans are very lucky to have such an involved owner.

As far as Dirk is concerned, for most of his career, WP has accurately rated him as a well-above average player. 5 times, he produced more than 10 wins, including 14.1 wins in his 2007 MVP season. (Incidentally, that was also Jet’s best season; he played like a star and produced 12.6 wins.)

WP says that Dirk was great, and was good as recently as 2011, but that he had an off year last year, a fact that Mark freely admits.

I think Mark’s point is that WP undervalues Dirk, which is quite possible with unique players. I’ve previously explored the possibility that decoy effects matter (e.g. “The Kobe Effect”) and that certain great players contribute in other ways.
Chris Yeh says:

September 5, 2012 at 8:44 am

(continued)

The Magic Johnson effect theorizes that Magic was more valuable than WP indicated (which already pegged him as the greatest player of all time) because of the open layup attempts he produced for the likes of James Worthy.

It may be that Dirk’s ability to pull a primary low-post defender away from the basket (he is the original “Stretch 4″) creates value that is captured by his teammates, rather than in his own statistics.

Moreover, Dirk’s WP numbers in the 2011 playoffs show him to be performing at an extremely high level; it was only his Finals injury that hampered his numbers then.

At any rate, the thing to do is for someone smarter than I to address Mark’s criticisms directly, with numbers. Here’s how I enumerate them:

1) Zone defense affects Dirk’s rebounding rate.
2) Dirk’s rebounding is lower because he is surrounded by players who handle that duty

1) Is hard to test, but 2) isn’t–where do the Mavs rank in rebounding, and how does Dirk’s presence affect it?
Chris Yeh says:

September 5, 2012 at 8:57 am

(continued, yet again)

In Dirk’s MVP season, the Mavs were an above-average rebounding team–41.9 TRB vs their opponents 38.1, for a net rebounding margin of 3.8.

In 2011, the Mav’s championship season, were an average rebounding team–41.4 TRB vs 40.7 for their opponents. In other words, the Mav’s rebounding had gotten worse.

I’ll buy that some teams (like the 2007 Mavs) are so loaded with rebounding, that they can allow scorers to focus on scoring. But that’s not true for the 2011 Mavs. This isn’t a case where Dirk’s teammates were able to crash the boards themselves, and his contributions were unnecessary.

In contrast, the 2011 Mavs’ Finals opponent, the Heat, had a TRB margin of 2.9 (not as good as the 2006 Mavs, but considerably better than the 2011 Mavs).

Dirk may still be more valuable than the stats, but the current Mavs aren’t so flush with rebounding that an individual PF’s rebounding is irrelevant.
Chris Yeh says:

September 5, 2012 at 8:58 am

(continued, one more time)

In fact, with the loss of Tyson Chandler in 2012, the Mavs actually became a below-average rebounding team. Their opponents outrebounded them, 43.9 to 42.8 TRB.

Given that fact, it’s hard to argue that Dirk’s lack of rebounding doesn’t hurt the team.
Chris Yeh says:

September 5, 2012 at 9:08 am

(the last chapter)

Almost forgot–all numbers courtesy of Patrick:
http://www.thenbageek.com/
Eugene Earnshaw says:

September 5, 2012 at 10:39 am

Mark’s comments point out the basic flaw with wins produced (and box-score based metrics generally): they don’t account for interaction effects between players. If JET knocks down an open look because Dirk drew a double team, JET gets 100% of the wins produced. If Harden misses a shot because DWade was defending him tough, and Haslem pulls down the rebound, Haslem gets all the credit, even though much of the ‘value’ of a defensive rebound is because they correlate with missed shots.

The wins produced posse make a big deal of how their metric is derived via regression analysis, and that is a perfectly fair point — if we assigned WP numbers to LINEUPS. But assigning WP numbers to individual players ASSUMES that players are responsible for producing 100% of their own box score numbers. But this is a false assumption, and as a consequence Wins Produced lacks validity.
Kai Xiong says:

September 5, 2012 at 10:55 am

I’ve gotta go with Mark on this one. I’ve just been getting into Wages of Wins and I think Wins Produced is a great measure, certainly better than the rest I’ve seen but it does seem fail to capture some keep aspects that I think intuitively would have a great impact on the game and how superstars are rated. First, Mark’s point on the double team – Great players, the Kobe’s, the Dirk’s will be double teamed and any team with a decent center will be rewarded for it (Offensive rebounds, tip-ins). I believe if you give partial offensive rebounding credit to shooters, it would aid the value of volume scorers. This makes sense statistically as well, offensive rebounds are dependent variables, you can only rebound a miss, so if we could correlate above average OR to player/team proportional FG% that could give you a point value factor for misses and space created. It may not be significant but something makes me think that on a team like the Lakers or Mavs, it would be.
Kai Xiong says:

September 5, 2012 at 11:04 am

Next point, bail out shots, I know you guys compare things on the aggregate and macro level with box score vs league averages but context is important. I’m sure if you look at players that are designated bail out options and factored in something to account for that, certain players would get efficiency boosts. Gonna use Kobe as an example (I’m a lakers fan and watch their games so I have the best feel for their situation), Mike Brown’s offense wasn’t very creative to say the least and the good teams would double the post on every possession. The stagnation caused a lot of late shot clock situations and someone ends up chucking it. Sure its low percentage but it’s >0 so it’s still better than a violation. The person usually left to bail out a badly executed possession is Kobe. System matters. I just rewatched spike lee’s Kobe doin’ work during his championship season and boy does system matter.
Kai Xiong says:

September 5, 2012 at 11:08 am

Honestly, even based of the age model, http://www.thenbageek.com/players/compare?career=1&direction=asc&player_ids=137&sort=name&utf8=%E2%9C%93, a 50% decrease in WP48 can probably be attributed to more than just standard regression.

Offensive creation, space creation, and bad offensive schemes should be factors in WP. Stop overrating rebounders, you can’t have a offensive rebound unless you have proper conditions. A shooter shooting when they are double teamed with an open man under the basket can be just as good as an assist.
Richie Sullivan says:

September 5, 2012 at 11:09 am

One thing I’d like to point out, and it’s probably useless this far down the thread, is that Dirk’s production last year after he took four games off basically returned to his career numbers. That’s impressive in its own right, but the early season numbers also drag that season’s statistics down in a way that leads people to the false conclusion that Dirk has lost more than he really has. At any rate, this next full season, before which Dirk didn’t even have Team Germany duties, will be telling.
Andres Alvarez says:

September 5, 2012 at 11:10 am

Kai,
Slow down :) As bartender I may have to cut you off.

Now you present some good cases and some good anecdotes. I appreciate both and that’s how research should start. Next question, do you have numbers backing up your claims? Also, you say we overrate rebounding. How do you rate rebounding?
James Washington says:

September 5, 2012 at 11:13 am

I’m surprised that up until this point we have not talked about Kevin Durant. He is a high volume jump shooter/scorer who WP loves. Why is this? Because he makes the increased shots he takes at an above average rate. Shooting efficiency is the quickest way to increase or torpedo your WP. If u make the shots u take u will be forgiven for not doing other things well except turning the ball over. Ray Allen is another example.

http://www.thenbageek.com/players/compare?career=1&player_ids=228&player_ids=10&utf8=%E2%9C%93

There is a lot of Red in the charts but for the most part both are still above average to star players because they shoot very well.

When we talk about efficiency we are talking about doing things better than average. To be a better than average shooter/scorer u have to make the shots you take. That is why Kobe was below average last year, he does many things well but he did not deserve 24 shots a game at the rate he was making them and hurt his team by taking them.
James Washington says:

September 5, 2012 at 11:29 am

If a player hits an open shot because a double teamed teammate passed it to them the double teamed teammate gets credit for an assist. Now the NBA does not count “Hockey assists so in that regard if that is what has transpired I agree. As far as taking shots and missing and a teammate getting a put back, why should Chandler/Wright not get the credit? If Chandler/Wright was running back down the court to get back on defense because he thinks Dirk is going to make the shot (which he then missed) and the defense gets the rebound than the team as a whole loses out on the extra possession.
For people who argue that shot taking (and missing) is an offensive positive; I thought we proved that wrong with Allen Iverson. The only way we can think it is positive is if we are comparing it to a shot clot violation. Then the question becomes is it better to take the shot clock violation and get back on defense as a set defense or take a bad shot a possibly let the opposing team get a fast break or
James Washington says:

September 5, 2012 at 11:30 am

hope your teammate gets the offensive rebound.

As for the value of the defender. There was a past post on the effect of defense on the other team. I believe it showed Lebron should be DPOY. Now I am not sure this is totally incorporated into the number but WP should be used with other metrics to make final decisions. It shows you who is above average and why with a box score and can be combined with other metrics if you are targeting specialist to use for specific things on your team.
Kai Xiong says:

September 5, 2012 at 11:38 am

Haha, I’ve had a lot on my mind. You guys got me thinking a lot as you can see. I haven’t yet found the time to do the numbers but maybe once I get the chance we can see if something sticks. Who knows, I could be dead wrong and Kobe is really a chucker :). But just some straight intuition like I mentioned finds some aspects lacking. About overrating rebounds, rebounds are definitively dependent and there needs to be some credit given to the person missing. Also, I think you guys have a Yay, small sample size issue as well (though I’m sure you’re aware of this). Low minute guys that do things well in limited minutes obviously will look better on paper projected out. I don’t think it’s a pure projection, I’d expect some decay specially if they foul a lot. ASIK anyone? *We’ll see how he does as a starter :). But I got a lot of math to do, and after doing it, we’ll see if my intuition holds. Maybe you can even run the articles
reservoirgod says:

September 5, 2012 at 12:16 pm

@JonC:

I do what you suggested with Wins Produced for every HEAT game. I simply use the avg. Win Score for each position in the game to estimate the wins produced by each player. For more info & 2 years of box scores with est. wins produced, see miami-heat-index.blogspot.com/p/heat-produced.html.
Mark Cuban pipes up for Chris Kaman | Dirk Nowitzki Jersey says:

September 5, 2012 at 12:19 pm

[...] The conversation (which is served with a spicy side dish of “how dare you question Dirk Nowitzki?”) is playing out now in the comments of the Wages of Wins blog. [...]
Alex says:

September 5, 2012 at 12:28 pm

Hey guys, I actually looked at the rebounding numbers of the players guarding Dirk in the 2011 playoffs. I assumed the main defenders he had on him were Aldridge, Pau, Ibaka, and Bosh/UD. The NBA advanced stats tools only have rebounds per 36 minutes and % of the team’s total rebounds, so I’ll use those (I’ll have them in boards per 36/% of team’s boards format).

For Aldridge, with Dirk on the court he posted 5.7 boards per 36 minutes/22% of his teams boards. With Dirk off the court, he had 9.4/29%. With Aldridge on the court, Dirk posted 7.5/26%.

For Pau, with Dirk on the court he posted 8.5/30%. With Dirk off, he posted 14.1/39% (Not the biggest sample size, but the trend definitely shows). With Pau on the court, Dirk posted 9.4/29%.

For Ibaka, with Dirk on the court he posted 6.9/23%. Dirk off, he posted 14.4/36% (Sample size still small but still…). For Dirk, with Ibaka on the court he posted 4.9/19%.
Alex says:

September 5, 2012 at 12:38 pm

(Continued…)
These are by far the lowest numbers and they are in a fairly large sample size (125 minutes). This is interesting since Aldridge and Pau are two of the best offensive big men in the game and Ibaka pretty much has a 15 foot jumper and dunks. Would have to watch more tape of this series to figure that out. Maybe they didn’t guard each other as much as I thought and the zone the Mavs used to counter KD and Westbrook probably had a lot to do with this too.

For Bosh, with Dirk on the court he put up 7.0/24%. With Dirk off, Bosh put 3.4/13%. Really small sample size and just seems like an outlier (Is Ian Mahinmi Bosh’s arch nemesis???). For Dirk, with Bosh on the court, he put up 8.9/30%. For Haslem, with Dirk on he put up 6.0/20%. With Dirk off he put up 7.9/32%. For Dirk, with UD on he put up 7.2/26%.

So while Dirk is never putting up Kevin Love/Z-Bo/Dwight rebounding numbers he is almost always outrebounding his man (per 36 minutes) and grabbing a higher % of his te
Alex says:

September 5, 2012 at 12:48 pm

(Last One…)
So while Dirk is never putting up Love/Z-Bo/Dwight rebounding numbers he is almost always outrebounding his man (per 36 minutes) and grabbing a higher % of his team’s rebounds. The per 36 numbers are always up for opponents when Dirk is off and they are always grabbing a higher % of their team’s boards.

So while Dirk isn’t pulling boards down like Moses Malone, he is turning 20-10 type guys (Aldridge, Pau, Bosh) into basically average rebounders and even worse when you consider how prominently these teams rely on those guys to rebound for them. The difference in what these guys put up when Dirk is on the court vs. when he is off, added to what Dirk is actually giving you makes Dirk at least an average rebounder if not an above-average one. I imagine if you expanded this to full seasons the trend would continue alas that is beyond what limited stats I have access to.

Hope this helps the debate somehow.

- Alex
Will Larsen says:

September 5, 2012 at 1:08 pm

@Kai,

I think Mark’s point wasn’t just that the high-production, low-minutes guys are skewed because of the small sample sizes, but because they’re mostly matched up with backups.

@Mark, and Kai,

I’m coming in a bit late to the discussion, but I want to come back to Mark’s point about Dirk being more valuable than the numbers show because he draws double teams. While this argument passes the eye test in this specific case – Dirk is a good passer out of the double team, and has such unique size and shooting form to get off good shots even when doubled – I’m curious whether you’d extend it to a player like Kobe. Kobe is still regarded by WP/WS as an above-average player (until last season), but certainly not a star; most of his value comes from above-average 3%, above-average rebounding for the 2 spot, and the fact that he gets a lot of star calls.

But watching a lot of Lakers games, I always got a feeling that they’d be better off with Kobe playing a secondary role, and maybe…
Will Larsen says:

September 5, 2012 at 1:09 pm

…even benching him in crunch time. Yep, he makes a lot of long Js, but not nearly as many as he misses. He’s an inefficient scorer, turns the ball over a lot, and the numbers have been showing this for the past three or four years. Now, finally, the public narrative is turning against hero ball. Is Kobe still worth more than his WP? It doesn’t mean anything to draw double teams if you’re going to shoot over them, and shoot poorly.
reservoirgod says:

September 5, 2012 at 1:10 pm

@Chris Yeh:

Comments like “Dirk was only the 5th most-important Mav in the Finals” do the Wins Produced model a disservice. I use the model, I like it, but context matters and if I had a criticism about WP analysis is that context gets ignored on this blog sometimes.

My est. wins produced numbers had Dirk as the 2nd most productive Mav in the 2011 Finals (Tyson Chandler was 1st). How? Dirk’s shooting efficiency was close to the avg PF in the series (45% for Dirk, 46% for avg PF in 2011 Finals) but he scored 9.6 more pts per 48 and was above avg in the other areas of the boxscore: getting to the line (where he shot 98%), rebounding (+2 per 48), +0.7 in net poss per 48, blocks, assists and committed 2 less fouls per 48. See http://miami-heat-index.blogspot.com/2011/06/heat-produced-how-2011-finals-were-lost.html

Tyson Chandler & Dirk outclassed the HEAT frontcourt. That’s what drove the decision to not even guard Stevenson & let him shoot 81% (as Cuban indicated).
reservoirgod says:

September 5, 2012 at 1:13 pm

I burst out laughing when I read this:

“The correlation between Win Shares and Regularized Adjusted Plus/Minus (which, for all of its noise on an individual basis, does begin to separate players by their influence on the game when taken in the aggregate) is 0.531, far greater than the correlation for Wins Produced (0.380).”

That’s PRICELESS…
Kai Xiong says:

September 5, 2012 at 1:39 pm

@Will – Matching up to backups I’d say is a result of sample size. The larger the sample (more minutes played) we are doing this per season, so that’s the only way to increase sample size, the more exposure to starters. So that was my point anyway.

Regarding Kobe, a lot of parallels are there between his game and Dirks if you think about it. Perimeter and post-up threats that command double teams and defensive zoning/adjustments by mere presence. What hurt them this season is shooting efficiency. I’d say the main reason for both their drops is a product of perceived overall team quality and resulting team system.
http://www.thenbageek.com/players/compare?player_ids=63&player_ids=137&season=2011&utf8=%E2%9C%93 vs http://www.thenbageek.com/players/compare?player_ids=63&player_ids=137&season=2010&utf8=%E2%9C%93. Everything is almost identical except field goal percentage. And guess what, prepare to be mind blown
TheNBAGeek says:

September 5, 2012 at 1:44 pm

Man! I always miss the good argument threads!

@Mark:

Great to see an owner chiming in. I agree with much of what you say about context. But you also cheated by using the “Everyone doubles Dirk” argument, because that ignores much of the context of WP ;-).

I will agree that WP does a very poor job of measuring interactions. I’ll also agree with Ari that a) these can be fixed and b) that WP is simply the least-terrible metric we’ve found.

I think you’re correct that WP gives coaches/managers a good starting point and useful information, and digging into the WHYs are those decision-makers’ real jobs.

And if Kaman and Mayo (both players we do not value highly) have a career year, we’ll be sure to mention that you called it. What interests me is, if the Mavs do not win, and those players’ WP remain stable, will you blame your system? Or agree that sometimes WP is telling most of the story?
Kai Xiong says:

September 5, 2012 at 1:54 pm

Kobe Shot C in 11-12 http://www.82games.com/1112/11LAL6.HTM. Lots of late clock situations, not very effective. Now lets look at the PJ era with champs/def champs 10-11 http://www.82games.com/0910/09LAL5.HTM and 09-10 http://www.82games.com/0910/09LAL5.HTM pretty uncanny stuff. Now for the mind blow. 04-05 http://www.82games.com/04LAL6A.HTM look at the clock distribution. Shot distribution is rather consistent, Kobe is mainly a jump shooter and a decent one, but in a poor offense with a lot of late game situations, his efficiency suffers. In more limited amounts, he actually isn’t a bad late clock shooter, but I’m sure it doesn’t scale up to any players favor.
Julien Rodger says:

September 5, 2012 at 2:47 pm

I have a number of major issues with WP

- It is obviously reliant on rebounding. Rebound is tied to context arg. as much as raw ppg. A player who grabs a rebound after boxing out his man, does NOT add the same value to the play as a player who doesn’t box out his man, but manages to get the ball. It’s possible a player can get a DRB to a statline on a play while simultaneously lowering his TEAM’s chance at grabbing the DRB. Let’s say Javale McGee gets a rebound on a play by running to where his Faried is. Doing this a) Increased the chance that he will take a rebound from Faried if it bounces toward both of them, thus not increasing the chance of getting that rebound in that area if Faried was going to get it, and b) Vastly decreased the chance of his team getting the DRB if it bounces to his original pos.

Or put it this way: Good perimeter d is based on rotating properly, NOT gambling for steals. 1 could argue McGee in above play = “gambling” rebounding to pad stats
Julien Rodger says:

September 5, 2012 at 2:53 pm

(cont.) Is it possible we’re looking at rebounding all wrong and that we should treat it like positional defense vs stls, or cornerbacks preventing receptions instead of going for interceptions in the NFL. That it’s the positional play and blocking out as a team that matters and not the raw numbers.

M. Cuban outlined most of the problems I have with the “scoring” portion of WP

What would interest me is a mix of Adj. +/- and WP. Meaning taking the principles of WP (splitting up the ratio of eff., reb., etc. on wins), but instead of using raw numbers, using the players +/- numbers in those areas. So instead of using context based box scores for them, they use the best teams can do to estimate IMPACT in efficiency, rebounding, etc., which would be much much different than the raw stats IMO, and likely far more accurate
Julien Rodger says:

September 5, 2012 at 2:57 pm

The fundamental problem of WP is that it doesn’t account for the difference between a player’s individual stats vs his impact. Every boxscore stat is based on context and may or may not be indicative of the player’s actual value/impact, instead of his teammates. The idea of WP is not bad, but it uses the wrong input (boxscore stats)
Julien Rodger says:

September 5, 2012 at 2:59 pm

WAR works in baseball to an extent (I say to an extent because OPS is still a bad stat due to not accounting for diff. between walks/singles, and arbitrarily treating OBP and SLG as 1:1) because every play is individually isolated. The same concept cannot be applied to WP based on boxscore stats because not every play is individually isolated.
Andres Alvarez says:

September 5, 2012 at 3:09 pm

Juliem,
Same warning as Kai. We want to hear what you have to say, but it’s ok to pace yourself.

I’m afraid we disagree on something fundamental. Context doesn’t matter to the stat. Getting the ball, scoring the basket. The scoreboard doesn’t care about the how. Now on defense, you’re not wrong. That is a place where assigning credit/blame needs work. A key part of the Wins Produced regression is that it fully accepts that defense matters BUT it’s not as high as many people think.

I’ll put the onus back on you. You’ve made some bold claims. Do you have numbers or stats backing up your claims? I do appreciate ‘fan expertise’ but without rigor, it just isn’t enough for me.
Charles Connaughton says:

September 5, 2012 at 3:13 pm

WP48 does *not* tell you which players are the best; it tells you which players have been most productive. It’s easy to make the jump to say ‘the best players are also the most productive’ – and the two certainly correlate strongly – but it ignores all those secondary effects that could be crucial. Comparing players filling very different roles on their teams can be misleading. For example, you could be comparing a player who is a first option on offense – drawing double teams and taking a lot of last second shots off of broken plays – to a guy that is left undefended because even then he’s a barely above average shooter. They could very well both have the same box score productivity, but which one should you pay?

Of course you could also be committing a Kevin Martin fallacy and underestimating a player that is really good at slipping defenders and getting open. So you have to tread carefully when going beyond the box score stats.
TheNBAGeek says:

September 5, 2012 at 3:17 pm

Dre,

Of course context matters. The fact that the scoreboard doesn’t care how it happened is entirely irrelevant. The scoreboard is not the actual result, the scoreboard is merely the measurement intstrument.

Julien,

Read the FAQ first. “WP Overvalues rebounds” was annoying to deal with when it was 2008. When you repeat it now, you just look ignorant.
Ty Hicks says:

September 5, 2012 at 3:22 pm

Julien,

WP actually attempts to correct for defensive rebounds that the team was going to get anyway. That’s why you see someone like Dirk(who is below avg at o-rebs and above avg at d-rebs) get valued lower than someone like Ryan Anderson(who is above avg at o-rebs and below avg at d-rebs).

I think a more interesting use of that logic would be steals and blocks. Players who get a lot of steals and blocks often do so by abandoning their defensive assignments. It’s quite possible these players could be hurting their teams more than helping them, despite putting up good numbers.
Court Vision | The Point Forward says:

September 5, 2012 at 3:37 pm

[...] Cuban is in a bit of a blog comment argument with the folks at Wages of Wins, who criticized the Chris Kaman signing and used their go-to stat, Wins Produced, to show several [...]
Julien Rodger says:

September 5, 2012 at 4:47 pm

Charles – WP measures WINS. Doesn’t that by definition, mean it’s measuring “impact” and thus who the “best” player is? PER is a stat that solely measures a player’s statistical productivity on the game, as Hollinger likes to define it. WP OTOH, by the virtue of the idea of “measuring wins”, is measuring not a player’s mass statistical productivity alone, but his impact on the team’s wins. This means factors like the difference between how much a player is actually impacting different areas of the game like efficiency and rebounding vs what his statistical production is, is extremely relevant.
Eugene Earnshaw says:

September 5, 2012 at 5:02 pm

Context doesn’t matter so much for how much a stat contributes to a win: 2 points is 2 points. But it matters a great deal for who should actually get credit for producing the stat: the person who gets the rebound isn’t necessarily the one who produced the rebound (which might, for example, produced mostly by a box-out by a big who didn’t actually grab the ball).

One of the main things that influences box score results but is not reflected in stats is shooting range, in my view. For example, by Wins Produced, Bargnani is famously bad and Ed Davis is quite good. However, Amir Johnson shot 46% when he played with Ed Davis, and 69% when he played with Bargnani. If Bargnani’s range is really contributing to this enormous difference, he deserves at least part of the credit for Amir Johnson’s shooting percentage (and therefore wins produced).
Eugene Earnshaw says:

September 5, 2012 at 5:11 pm

And to bring it back to Dirk, Jason Terry shot 10% better from 3 and 8% better overall when Dirk was on the floor as opposed to on the bench (check out NBA statscube to check out this kind of player pair info)
ClipperBlog.com Blog for the Los Angeles NBA Clippers Fans » Blog Archive » DeAndre Jordan and the ‘Developing’ Myth says:

September 5, 2012 at 5:16 pm

[...] pm While the rest of the basketball blogosphere was railbirding the discussion between Mark Cuban and the stats community on the usefulness of the individual Wins Produced stat (an interesting/surreal conversation for LAC [...]
RJ McKey says:

September 5, 2012 at 5:40 pm

I am a pretty big supporter of looking at lineup performance. Here is some simple information I find interesting:

Over the past 5 years there has been at most 18 lineups used 400+ minutes in the league (pro-rating last season). Last season had the least at 13. The average is just over 16. So with almost 4000 regular season minutes to work with, teams average barely more than 1/2 a lineup with a fairly decent sample size of 400+ minutes. Coaches are simply not providing a good dataset for the analysis of lineups. And I think this is foolish.

Not a single lineup used over 400 minutes in 2010-11 was used a proportional amount in 2011-12. This makes year to year 5 man lineup performance analysis difficult. One can resort to looking at 4 man sets but the remainder is important and unfortunate to have to blur.
Ari Caroline says:

September 5, 2012 at 5:41 pm

Clearly the lack of system dynamics in the WP model is a bit of a shortcoming. It even could account for systemically undervaluing a sub-class of players like bigs with outside range (though that is still to be proven). As Patrick and I have mentioned, these are issues that can and will be addressed in future refinements of the model. Anyone who has seen Arturo’s work should have great faith that he, for one, is never satisfied with the current version of anything.

That said, many are way too quick to throw away the baby with the bath water. The model is amazingly resilient and an amazingly strong predictor, even with its shortcomings. I’m with Dre. Until someone presents some alternative model backed by robust data and analysis, let’s acknowledge what this one can do and build on it.

For my team at Sloan-Kettering I have a rule that criticism isn’t allowed without presenting a potential alternative. I don’t get to make the rules here but I hope you’ll see the wisdom in it.
RJ McKey says:

September 5, 2012 at 6:17 pm

There were a grand total of 4 lineups used 200+ minutes in each of the last 2 seasons. 3 were great, 1 bad. That is pathetic in terms of product testing by coaches & management.
RJ McKey says:

September 5, 2012 at 7:17 pm

Ari, I think it might be interesting to do some sorting of clustering of players into 10-15 groups and look at wp and other stats within and between those groups.
Kyle Shook says:

September 5, 2012 at 7:42 pm

Arturo had a post when I first started reading the WoW network (it was on his old site or maybe Dre’s) that had a post showing how using team’s WP could be used to find the expected point-differential. He then compared that with the actual point-differential of the games and ended up showing a nearly 1:1 correlation. After reading that I’ve always had a hard time arguing against WP as it seems to have such a high predictive power.

I doubt I would be able to find the link now but it seems to be a good argument in favor of what you WoW guys are doing.
Kyle Shook says:

September 5, 2012 at 7:52 pm

Since we are all over the place in this comments section can I add that, despite the attention Henry Abbot brings to WoW, I just cannot stand his material.

His post regarding this thread was fine, and then he added this:
“I can remember only one other time when Cuban made a big bet that flew in the face of most stat geekery: When he traded Devin Harris for Jason Kidd.”

Is he talking about PER? Is everyone at ESPN legally required to kiss John Hollinger’s bottom?

He must have missed Vivek’s comment :
“You basically traded Harris (an average PG) for one of the best players in the game.”
Eugene Earnshaw says:

September 5, 2012 at 8:04 pm

Kyle: That raises exactly the gap between what Wins Produced is good for and what it isn’t: team and lineup total wins produced are super correlated with and predictive of point differential. But there is a logical leap from that to saying that therefore an individual player’s wins produced says something definitive about their worth on a basketball court. Any positive (or negative) synergy effects a player has will show up in the team and lineup WP values, but not in their individual ones. So to take individual WP seriously, you just have to ASSUME these effects are negligible.
Ben Gulker says:

September 5, 2012 at 9:10 pm

Might it also be possible that Dirk is penalized for lower rebounds and shooting % because he is misclassified as a PF (yes, I know how tall he is)? His game in recent years is much closer to that of a swing forward and I could easily see the Mavs playing him with two other bigs if they had them. Defending a young, agile SF might be an issue, though.

In my opinion, this is exactly correct. Dirk defies positional definitions, and as a result, WP struggles to tell the story of Dirk’s productivity.

On offense, he plays more like a SF (almost like a slashing SG in some ways), and as a result, he is away from the basket and his ORBs are low.

On defense, he guards PFs and is thus closer to the basket, and he collects rebounds well.

It’s relatively easy to work around though. Just compare his offense to other perimeter guys and his defense to other PFs.

That said, I think it is fair to point out that Dallas without Chandler wasn’t a real contender last season.
joshweil says:

September 5, 2012 at 9:42 pm

@Kyle Shook

Arturo’s done it here: http://arturogalletti.wordpress.com/2010/12/03/game-to-game-wins-produced-correlation-for-2009-2010-season-and-a-fun-little-equation/

And I’ve done it here: http://joshweil.blogspot.com/2012/08/wins-produced-and-game-outcomes.html
Chris Yeh says:

September 5, 2012 at 9:45 pm

http://www.thenbageek.com. For the stats.
I have another simple proposal…someone should take a look at all transactions involving “stars” according to the WP definition (WP48 > 0.200). Then compare the WP figures for their old and new teams and teammates. This would be a systematic version of what I did with the Magic Johnson effect.

We could then break down WP “stars” into scorers and non-scorers to see if the behavior is any different.

For example, Ben Wallace went from the Detroit Pistons to the Chicago Bulls in 2007. His WP figure was pretty much steady.

The 2006 Pistons, led by Wallace, produced 59.1 wins. The other All-Stars on the Pistons performed as follows:

Chauncey Billups: 0.254
Rasheed Wallace: 0.095
Tayshaun Prince: 0.108
Rip Hamilton: 0.075

The following year, without Wallace, they performed as follows:

Chauncey Billups: 0.222
Rasheed Wallace: 0.050
Tayshaun Prince: 0.161
Rip Hamilton: 0.100

While the other “stars” performed the same, losing Wallace brought the Pistons down to 52.5 wins produced.
joshweil says:

September 5, 2012 at 9:55 pm

@Eugene Earnshaw

Dave Berri has analyzed the interaction effects and has found them to be small.

“4. One last point… it is assumed that a player statistics are his statistics. In other words, if you score 10 points (or get 10 rebounds), you get credit for these points (or rebounds). If you take 5 shots, you are charged the cost of these shots. It’s important to note that there is some interaction between the player statistics, but I have found this effect to be rather small. For the most part, what you see in the NBA is what you get. So it makes sense to credit a player for his statistics (by the way, I would not make the same argument for NFL players)” – http://3shadesofblue.blogspot.com/2008/02/wondering-about-wages-of-wins.html
Chris Yeh says:

September 5, 2012 at 10:02 pm

Actually, let’s make this a challenge. I challenge the proponents of other models to find a trade or free agent signing, involving a WP “star” or “superstar” (0.200 or 0.300+), where the WP model failed to predict the results better than another model. Injuries that radically alter playing time invalidate the experiment.
Ari Caroline says:

September 5, 2012 at 10:09 pm

@RJ It’d be nice if Mark gave us a platform for some experimental economics. :)

@Ben Yes, but you have to consider the very real opportunity cost of actually playing Dirk at the 4. It means that there isn’t another 4 out there collecting those missing rebounds and making up the TS% difference. It only works if he is actually played as a 3. This is actually consistent with Mark’s lineup logic.
Oliver Maclaren says:

September 5, 2012 at 11:23 pm

I don’t have much to add since I have not (yet, hopefully!) put in the time and effort required to get up to speed on basketball stats and wins produced. All I want to say it that what has most impressed me about the wages of wins folks is the mature scientific/modelling philosophy regularly espoused here. I can tell that these are people who are really immersed in quantitative analysis and have a desire to approach problems the right way.

Some of this basic philosophy of quantitative modelling is just hard to communicate, and to understand some of these ideas I think what one really needs to do is what Dre says above – move beyond, or even ignore, your “common sense”, anecdotes etc, and start making and testing models, figuring out what the right questions are and so on.

I look forward to seeing Ari’s promise of improvements to come be fulfilled! In the meantime I appreciate the power of wins produced as a good first order approximation, and the “least worst” model I’ve come across.
Eugene Earnshaw says:

September 6, 2012 at 5:45 am

@joshweil Berri offering his assurances that he has analysed these effects and found them ‘rather small’ is not helpful without some insight into the nature of the analysis. Otherwise it is just an inappropriate appeal to authority. Furthermore, even effects that are small on average might be very important in certain contexts. So I do not find that quote to be reassuring.
Arturo says:

September 6, 2012 at 7:23 am

I love this kind of thread. Intelligent discourse drives innovation.
Some point:
WP is a great model but it had/has some shortfall and limitations. This is why we continue to work on it. Going so far as to completely overhaul how we deal with assists and rebounds based on good scientific research.
There still some things to work on though. One in particular to me is the effect soMr truly exceptional players have on their teammates. Dirk and Lebron on offense, KG and Tyson on d are some of he examples we’ve come across.
I for one wil continue to work at making our understanding of the game better and I hope we can entice more of you to join in.
James Washington says:

September 6, 2012 at 7:35 am

@Eugene & everyone else saying shooters should get credit for offensive rebounds. Think about this, no one shoots with the intention of missing the shot. The basic premise of shooting for wins produced is that u need to make the shots u take at a higher rate than AVERAGE to DESERVE to take more shots. If Bargnani/Kobe (last year)/ Allen Iverson etc. do not make the shots he takes he should pass the ball to a better option. To say better options do not exists ignores the fact that these guys are professionals. Kobe taking contested fade away jumpers from behind the back board or Bargnani taking long contested 3′s is not better than an open shooter shooting. All of these players would be shown better at producing wins if they 1. Took better shots and/or made the shots they take or 2. Passed the ball to a open teammate for an assist or a Hockey assist. If a player is good enough to be double teamed for their perceived scoring prowess, if they pass the ball and help their teammates
James Washington says:

September 6, 2012 at 7:35 am

take advantage of the mismatches everyone is better off because at some point the defense will stop double teaming and the volume shooter will find themselves with better shot opportunities.
Mark Cuban says:

September 6, 2012 at 12:16 pm

let me give you some context that may or may not help define variables.
- Team chemistry is important. While it doesnt matter who scores, it does matter that all players know their role and feel good about it.
- Most teams call plays on most possessions. SO its not just about playing basketball and letting your best players do their best. We track what plays work. P&R works, hence the whole league has gone to it. Same with drive and kick.
-It doesnt matter who is in the P&R as long as it works.
-Teams go to what works until the other team takes it away.
-Same with defense. We track what is working in a game and go with it till it doesnt.
- One selfish guy can ruin a team. NO matter how efficient he is.
- a big who is below average in TS% vs centers, is often above average in TS% vs a guard. Put them in a P&R or P&Pop and they can impact the game.

in essence every non FB play is a game in and of itself, normalized by the need to protect the team elements that are important to teams
Statmandu says:

September 6, 2012 at 1:08 pm

IMO “drawing double teams” is overrated.

When you draw a double team it only has value if it leads to a much better than average shot. That’s not always the case because the double tends to come off the weaker alternative scoring options. Second, if you draw a double and do get the ball to a superior option (like Jason Terry), you get an assist. So the value of drawing a double team is sometimes captured by the player who deserves it.

What we are really talking about here is mostly hockey assists. But there aren’t all that many of them & the value should be shared between multiple passers anyway.

So if you want to say Dirk drawing doubles has value that is not being captured by the model, I’d have to agree. But IMO there’s no way it is HUGE value.

Throw him an extra win or two and be done with it.

Dirk is one of the best scorers ever given his versatility, but he’s not way above average at multiple things & is overrated because of the emphasis on scoring.
Statmandu says:

September 6, 2012 at 1:18 pm

Mark,

You are talking about another area of statistics that has value. Essentially you are trying to figure out how to use your players’ skill sets to generate as much value as possible for the team.

This model is essentially saying that most teams do a good (and similar) job of that and the resultant productivity to a large extent gets captured in their box score stats.

What you are doing is essential, but it does not measure the value of the players.
TheNBAGeek says:

September 6, 2012 at 1:30 pm

@Mark,

Again you make lots of good points. But I would argue that much of what you are saying is captured in aggregate. For instance, you say:

- a big who is below average in TS% vs centers, is often above average in TS% vs a guard. Put them in a P&R or P&Pop and they can impact the game.

But I’m unsure how this is irrelevant; you’re simply highlighting the fact that taller players should shoot better against smaller ones. But this is, of course, true for the center who is also above average in TS% vs. centers.

It’s possible you are making an argument that some centers have lower TS% because they never play in the pick and roll. This is probably also true, but as I stated before, finding out the *why* isn’t within the model’s scope.

Lastly, if the system is so important, we should see large variations in performance both from year-to-year (as rosters change) and when players change teams. Yet we don’t (with the possible exception of players moving to and from the Spurs :-)
James Washington says:

September 6, 2012 at 1:52 pm

@Mark

We understand that context matters but by the context you just provided almost everything that happens on a basketball should be reflected in the box score somehow. If a center (C) has a high TS% vs a guard (G) and teams keep going to plays that work then that center should have a high TS%. If the C does not something is going on there and we can look at shot distribution to figure out what. If you are a 7 fter or a C and your name is not Dirk or Kevin (Durant & Garnett) you probably should not be taking 19-22 ft jumpers very often. Also if ur team’s system is geared toward making mismatches of the C/G nature than it is to the C’s advantage to be a close to the basket when this mismatch occurs unless they are a very good 3 pt shooter both % and otherwise.

While WP doesn’t measure everything it does a great job of showing who does what & why certain players are good/great. This is very valuable information.
James Washington says:

September 6, 2012 at 2:02 pm

If a player on your team has a low TS% vs average than you can look deeper to figure out why.

It is to the your advantage to look at as many things as possible to find what works. If you find something that no one else has don’t tell anyone and keep using it (hat tip to the podcast by Dre). Wow is here to help out & if you find that what we say is helpful keep using it because it is currently being provided to you for FREE. This is every traders wet dream. A free product most people are ignoring that is very effective in explaining not just what has happened but why and who is responsible. I hope during the season there are more articles that pique your interests and more topics you can use to make your team and the game of basketball better. One thing you have that none of us have is the ability to affect change. I hope we can help you do it in a way that positively effects basketball (which will also bring your team more continued playoff berths and championships if used correctly).
Arturo says:

September 6, 2012 at 3:20 pm

@Mark,
To expand a bit on Statmandu and TheNBAGeek’s responses, We done quite a bit of research on year to year correlation. A lot of it is summarized here (http://wagesofwins.com/2012/07/25/age-is-just-a-varible/).

The key finding is that the year to year correlation is fairly strong for all players (close to 80%). To your point, the correlation of players changing teams fluctuates more. The implication is that while the majority of teams do optimize their resources by putting players in a position to succeed there are players out there that are being mismanaged or miscast in the wrong role. Which is the point being made on Kaman. I will again caveat that these are rare instances.

You mentioned the idea of players having longer career arcs. My work has seen some of that as well. I haven’t published it here as of yet but the key finding isn’t really that players’ productivity vs age curve is changing but rather that the pool of older player that can actually suit is broader (owning to medicine and nutrition most like). This means that the quality of older talent is actually going up with time.
Jon Nichols says:

September 6, 2012 at 3:31 pm

Just a quick question:

Most proponents of WP cite it’s strong ability to explain wins, such as here:

http://arturogalletti.wordpress.com/2010/07/20/predictive-stats-bad-metrics-correlation-in-the-nba/

According to that study, it explains 94.9% of wins, a wonderfully high number. However, has anyone done research on the correlation of WP to team wins without the team defense adjustment? Specifically, in the “how to calculate Wins Produced” page:

“Step Four: Incorporate team defense and calculate adjusted P48.”

I’d be very curious to see how WP correlates with wins if left only with the same individual defensive statistics that a metric such as PER uses (steals, rebounds, blocks, fouls, etc.). Otherwise, using team stats seems like a poor way of allocating value to individuals, which is what the goal of any *player* rating should do. From my experience playing with different stats (PER, Win Shares, WARP, etc.), adding a team factor can always put my R^2 values over 90%.
Arturo says:

September 6, 2012 at 3:41 pm

Jon,
From the http://wagesofwins.com/wins-produced/how-to-calculate-wins-produced/ page:

“The average value, in absolute terms, of DEFTM48 is 0.010, so incorporting team defense results in a very small adjustment to an individual player’s per 48 minute performance. Furthermore, the correlation coefficient between P48 (adjusted for defensive rebounds and assists) and Adj. P48 in 2010-11 was 0.996.”

So Productivity before the defense adjustment has a 99.6% correlation with Productivity after and the correlation to wins remains at the 94% level. I’ve also done similar work with simple Win score (that’s at about 90%).

The defensive adjustment by the way is straight from the boxscore as well (just from the team & league totals).
Jon Nichols says:

September 6, 2012 at 3:47 pm

Arturo,

When you say, “and the correlation to wins remains at the 94% level. I’ve also done similar work with simple Win score (that’s at about 90%),” do you mean that WP without the team adjustment correlates with wins at the 94% level?
Arturo says:

September 6, 2012 at 3:51 pm

Jon,
Yes. The defense adjustment is not that significant to the overall calculation. Most of the defensive value is captured in the boxscore (reb,blocks,steals) stats.
Jon Nichols says:

September 6, 2012 at 3:58 pm

Arturo,

Thanks for the clarification. I wasn’t aware that the correlation between non-adjusted WP and team wins had been tested before.
Jon Nichols says:

September 6, 2012 at 4:04 pm

Also, since this debate has somewhat shifted to the pro’s and con’s of WP, have you seen this article?

http://bbmetrics.wordpress.com/reality_check/
Arturo says:

September 6, 2012 at 4:15 pm

Jon,
No. I wouldn’t consider that a fair test given how the model is built. You need to account for the opposition. You’d have to use the full lineup data (team and opponent) and project out margin using ADJP48 (Raw-non position adjusted productivity, size matters :-) ). That would be an interesting test. Particularly for looking for non boxscore effects.

If I were working on this kind of regression I’d also look at throwing in Homecourt, Altitude and Rest and ,if I had the data, refs.

Purely hypothetically of course :-)
hansoo says:

September 6, 2012 at 4:19 pm

Is there value in applying Organizational Behavior theories to NBA teams? (I’m new here and there seem to be a bunch of crazy smart people here so this might have already been discussed/researched.)

To elaborate: it seems like some of the arguments that Cuban is making are more around the “softer” skills that are essential for building any good organization:
- having good judgement in the moment (Dirk making the right pass at the right time)
- being a good “co-worker” (JKidd helping teammates develop as players)
- having players understand their roles and buy into them (Jet coming off the bench)
- having management create a system that fits the talent, and/or find talent that fits the system (the Mavs building around the unique talents of Dirk)

To re-phrase: are WP, TS% like having good operating metrics, in that they are critical for running an organization, but are they enough to build a team with?

Thanks for the discussion.
Eugene Earnshaw says:

September 6, 2012 at 6:41 pm

On further thought, the diagrams should have been mpg instead of minutes played so as to be biased less by injuries (e.g. Varejao, Andersen, Bargnani).

@JamesWashington: I wasn’t saying shooters should get credit for offensive rebounds off their misses. I was saying they should get credit when their floor-spacing causes their teammates to have higher shooting percentages. As a general note, your comment assumes that better shots are available than the ones that are actually taken. While sometimes there may be, there is no guarantee. It’s not like someone like Ed Davis passes up the dunks and close shots that give him a relatively good shooting percentage: the problem is he doesn’t have a skill set to allow him to up his usage rate efficiently.
Chris Yeh says:

September 6, 2012 at 8:38 pm

@hansoo,

I strongly believe that organizational behavior and applied psychology have a place in constructing a team. These can help you get the most out of your team.

Bear in mind, however, that all the best behavior interventions in the world are unlikely to turn Al Harrington into Kevin Durant.
Ari Caroline says:

September 6, 2012 at 8:53 pm

Eugene, You’re correct, of course, about mpg (with a minimum threshold for total minutes). You’ll see it in the next charts, Gd Willing.
joshweil says:

September 6, 2012 at 9:41 pm

@Eugene

It’s entirely appropriate to defer to the work of Dave Berri! He’s the sports economist with 20+ journal publications who built the model and wrote the book. If you have data to the contrary please present it.
Eugene Earnshaw says:

September 6, 2012 at 10:56 pm

No, actually, it doesn’t matter what Dave Berri’s credentials are without an argument. Don’t get me wrong: I’m sure he did an analysis that he found convincing. If I see it, maybe I’ll find it convincing too. But just referencing an interview on a blog where Berri says interaction effects are minor is not evidence that anyone should accept.

And, not to make a big deal of this, but the burden of proof is on the Wages of Wins people here. For them to draw conclusions about the value of an individual basketball player from the correlation between team wins and team wins produced is not valid without evidence that players are almost never responsible for each others’ stats. I haven’t seen any such arguments: if they’re out there, I would genuinely like to see them. Evidence such as I discussed earlier, although not at all conclusive (not that it was intended to be), at least indicates why one might suspect that interaction effects are important.
hansoo says:

September 6, 2012 at 11:28 pm

@ChrisYeh – totally agree. no degree of Magic Johnson Effect is going to raise Al Harrington’s game to Durant’s.

just wondering if there are some other models/theories that can improve the thinking around NBA team building and that hasn’t been explored as deeply.

look forward to your Magic Johnson post.
Andres Alvarez says:

September 6, 2012 at 11:46 pm

Eugene, Sorry I’ve been busy for a bit. I am curious. How much of Dave’s academic work have you read? He does love a challenge and answering things. He does have over 40 published papers and has said a lot on basketball. What exactly hasn’t been answered to your liking in them? (citations please.)
Eugene Earnshaw says:

September 7, 2012 at 5:54 am

I’m no expert here: I have read an unsystematic (but fairly extensive) selection of material from various websites, including the various foundational ‘here’s what wages of wins is’ posts, and the arguments for the superiority of wages of wins over the rivals: I have read everything I could find that attempts to justify the methodology behind wages of wins. I have seen a few comments defending the assumption that interaction effects are minor, but I have seen no evidence. As I said, I’d be genuinely pleased to be directed to the evidence if it’s available (even if it’s an academic paper rather than a website post). Any suggestions as to where to look?
Kai Xiong says:

September 7, 2012 at 9:29 am

I think the overarching criticism that I’ve seen is that people have an unsettled intuitive/gut sense that WP48 isn’t capturing something on the individual level as often times contradicts what we fundamentally believe about the game of basketball on many levels. WP48 on the Team Aggregate level without a doubt in my mind is the best predictor of Team success and that’s while those team values correlate so greatly. Combine everyone’s stats and attributed WP values and the box scores tell the story. However, looking at it on a more micro level, it seems to be lacking in some regards. WP goes under the assumption that WHY and HOW doesn’t really matter (other than the marginal effect on dreb). But after scouring around the site, I’ve yet to find something like what Chris Yeh suggested. Something showing WP before Superstar and after Superstar, or WP before Spurs/Triangle/Not George Carl, etc. This would go a long way to silencing critics.
Kai Xiong says:

September 7, 2012 at 9:50 am

I got bored and ran a regression on the effect of changes in playing time from 10-11 to 11-12 to all players in both seasons and it seems that MPG Delta has a decent correlation to WP48 Delta. Y=0.0065x+0.0081 R2=0.0927. I have a spreadsheet but nowhere to post it. Now the question I had was the chicken and egg conundrum. Did these guys get more playing time because they were win producers or were they win producers because they had more minutes. Looking at 10-11, there was a net negative correlation between MPG Delta and WP48 interestingly enough, worse players actually got more playing time and better players got less (I’d say injuries mattered), but increased playing time on avg increased WP48 by 0.0365 and less playing time by 0.01665.
Andres Alvarez says:

September 7, 2012 at 11:11 am

Eugene,
That’s fine but not quite acceptable. You are criticizing Wins Produced in terms of “standing up to rigor”, when it is one of the few metrics that has made it through peer reviewed processes. Again, you haven’t addressed those. To say “Wins Produced needs to defend itself” is fine. The point is, it has, in academic work. I reiterate: what problems with the academic works do you have? (citations, please)
A Sneak Preview from the Wages of Wins Labs. | The Wages of Wins Journal says:

September 7, 2012 at 12:53 pm

[...] has been working tirelessly, inspired by the great work of Ari Caroline. We’re not quite ready for a full demo but for those of you lucky ducks still around this [...]
Eugene Earnshaw says:

September 7, 2012 at 12:53 pm

Where did I use the phrase “standing up to rigor”? That’s not even grammatical.

Andres, I am not the one making claims about being able to evaluate the value of players. I am reporting the fact that nobody has produced for me the evidence that would be required to make such claims convincing. I have been assured the work has been done, but nobody will tell me where to find it.

I’ll tell you what. You tell me in which of Berri’s papers he addresses the strength of interaction effects, I’ll read the paper(s) and see if I’m convinced. I’ll be super pleased if it’s a good argument. If you don’t know where this analysis is to be found, perhaps you should yourself consider what reason you have to dismiss the importance of interaction effects.
Andres Alvarez says:

September 7, 2012 at 1:06 pm

Eugene,
Your point was “Wins Produced needs to defend itself” Your claim is you’ve read a lot of stuff on the net. My response is that it has “defended itself” and you have not cited any of that. It is pretty easy to say “I found websites that disagree with you.” Regardless, the point that you’re ending with “I’ll read the papers” (many of which are mentioned in the books that motivated this site.) means you haven’t checked the defense yet. I’ll enjoy what you have to say then. Until then, I’m afraid your requests/demands etc. are pretty empty.
Jon Nichols says:

September 10, 2012 at 9:29 am

@Arturo

Your claim that the team adjustment doesn’t have a large effect on the explanatory power of WP seems to conflict with Berri and Bradbury (2010):

“One should note, though, that in calculating Wins Produced, a team defensive
adjustment is used, which takes into account the number of points scored
by the opponent, opponent’s field goals made, opponent’s free throws made,
opponent’s turnovers (that are not steals), and team rebounds that change possession.
If you add this team-defensive adjustment to Equation 5, you can increase explanatory
power for NBA Efficiency to 58%. For PERs and Game Score, explanatory
power rises to 56% and 60%, respectively.”

Given that the explanatory power of PER was 33% prior to the adjustment, the team adjustment holds considerable weight. You’re saying the same effect does not happen with WP? Given that both WP and PER are based on the same set of sets, I would think it would. If I’m wrong, do you have a citation for WP without the team adj.?
Jon Nichols says:

September 10, 2012 at 9:30 am

WP and PER are based on the same set of stats*

Sorry for the typo.
Arturo says:

September 10, 2012 at 10:10 am

Jon,
I’m actually going to put this together as a post this week. I would have done it over the weekend but my internet access was limited. It’s a fair question and I really should get the data out there.
Are Coaches Playing the Right Players? | The Wages of Wins Journal says:

September 12, 2012 at 12:20 am

[...] you recall this work started examining how teams misplayed centers. Astute readers made the point that minutes per-game was a better measurement than total minutes. [...]
Court Vision « says:

October 16, 2012 at 4:12 am

[...] Cuban is in a bit of a blog comment argument with the folks at Wages of Wins, who criticized the Chris Kaman signing and used their go-to stat, Wins Produced, to show several [...]
Should Jason Kidd Have His Number Retired in Dallas? | TheSportsNewsHub.com says:

November 21, 2012 at 4:27 am

[...] Smith, a mustachioed Bill Simmons, Brooklyn Nets’ owner Mikhail Prokhorov, the Wages of Wins blog commenters, the SEC, and just a few days ago, [...]
Court Vision | The Point Forward - SI.com says:

January 24, 2013 at 5:50 am

[...] Cuban is in a bit of a blog comment argument with the folks at Wages of Wins, who criticized the Chris Kaman signing and used their go-to stat, Wins Produced, to show several [...]
Craig says:

April 3, 2013 at 9:01 am

Curious to read this after its mention on Twitter. There are arguments to be made I suppose and since Kaman has become a fan scapegoat in Dallas, it’s curious to see the arguments.

Wins produced is a curious stat for guys who have rarely played on a winning team. How would they be for Elton Brand? How about for Kyrie Irving? Statistics largely have to be put in context.

IMHO, the failure to make good use of a player who has significant strengths and weaknesses is largely about two things: available personnel and coaching decisions. With a guy like Kaman, having him on a team surrounded by good defenders would provide better cover for him, as it has for Dirk for many years. Would he flourish in OKC? On the other hand, his greatest success was playing alongside Brand in L.A. and Brand himself is a power forward, not a center. Why has Carlisle never put them on the floor together?

Kaman perhaps doesn’t pass enough and should try to work more in the low post where he’s very effective, although he still has a good midrange jump shot and shoots over 50%. If he needs to alter his game, is it not Carlisle’s job to tell him to do so? I am puzzled as to why they have not insisted he work on reestablishing his low post game and pairing him with Brand or even Bernard James at PF. The lack of perimeter defense and subpar playmaking isn’t helpful either but surely things could be improved.
Someone Else Telling the Story about the Overpaid in the NBA says:

April 8, 2013 at 11:10 pm

[...] News. The Dallas Morning News also commented on Chris Kaman’s name on the list. By the way, you might recall that we criticized the signing of Kaman before the season started. And Mark Cuban – Kaman’s boss – objected to our analysis (see the comments on Ari [...]

September 2012
M	T	W	T	F	S	S
« Aug				Oct »
					1	2
3	4	5	6	7	8	9
10	11	12	13	14	15	16
17	18	19	20	21	22	23
24	25	26	27	28	29	30

Share this: