Rick Morrisey – a sportswriter with the Chicago Tribune – has the following advice for John Paxson (general manager of the Chicago Bulls): Bulls should throw Joakim Noah overboard.
In this article Morrisey makes two arguments:
1. Noah has a problem fitting in with his teammates and coaches.
2. Noah is not a productive NBA player and is therefore not worth the hassle.
Abbott Responds
In response to this column, Henry Abbott – of TrueHoop – noted the following:
There’s a column in Chicago today, by Rick Morrissey of the Tribune, ripping Joakim Noah for various things. One of the criticisms is that he is not productive: “Noah’s supporters say he’s a monument to hustle. No. He’s the king of looking like he’s hustling. When you’re 6 feet 11 inches, wave your arms with exaggerated gestures and have a ponytail that bobs as you run up and down the court, you tend to look busy. He’s averaging 4.5 points, 5.9 rebounds and 18.2 minutes per game. The numbers say he’s not worth the hassle.” I think that’s a case of old-school stats deceiving. Because more sophisticated numbers show Noah’s one of the best rebounders in the NBA per minute. Noah is third in the whole NBA getting offensive boards, and his overall rebound rate is ahead of that of Shaquille O’Neal, Tim Duncan, Antonio McDyess, Kendrick Perkins, David Lee, Ben Wallace, Yao Ming and many other big names among big men. Noah looks like a bit of a spaz out there — agreed — but he is not unproductive.
The Value of Noah
So here we have two journalists commenting on the relative value of Noah. Both are looking at the same numbers. But both are coming to different conclusions.
As we often do here, let’s look a bit more closely at the numbers. We will start with a comparison of Noah, Aaron Gray, and Spencer Hawes. Each of these players was selected in the 2007 draft. Noah was selected with the 9th overall pick, Hawes went one pick later, and the Bulls took Gray in the second round. In sum, Hawes and Gray are Chicago’s alternatives to Noah. In 2007 Chicago could have selected Hawes. And today, Gray is often being chosen over Noah. Across the last 16 games, it’s Gray who has been the primary starter at center for the Bulls.
When we look at the numbers – reported in Table One – we see that Noah is actually offering more per-minute than either alternative. You wouldn’t see this, though, if you focus solely on point production. Noah certainly leaves much to be desired as a scorer. With respect to all of the scoring stats, Noah is below average. When we move past scoring, though, we see that Noah excels in everything except personal fouls. In sum, there is much to like about Noah. At least, as long as you look at more than scoring.
Table One: Evaluating Joakim Noah in January of 2009
To put Noah’s production in perspective, let’s look at the 10th player chosen in 2007. Spencer Hawes – who plays about 10 minutes more per game than Noah — is below average with respect to everything except taking shots from the field, shooting efficiency from the free throw line, blocked shots, and assists. But because Hawes takes an above average quantity of shots, some people have argued that he is a better player than Greg Oden. And I suspect many would take Hawes over Noah. The numbers, though, suggest Chicago made the right decision in the summer of 2007.
Despite the story the numbers are saying, Chicago is acting like they made the wrong choice. Noah is only averaging 18 minutes per game this season and as noted, recently Gray is starting at center for Chicago.
If we go back to Noah’s college numbers we see that he was not much of a scorer at Florida. Although Noah clearly demonstrated that he was not going to be a major producer of points in the NBA, Chicago still spent a lottery pick to acquire his services. And now, after demonstrating that his game is really not about scoring, Chicago behaves as if Noah’s not really a very productive player. Yes, as the Morrisey article indicates, Noah does have a problem getting along with coaches. But as Morrisey notes, this is not generally an issue for players who are considered productive. It looks like Noah is doing what his college performance promised, yet now Chicago seems unhappy to get what they asked for.
Finding Problems in Chicago
And when we look at Wins Produced, we see that Chicago is actually getting quite a bit from Noah. In fact – as Table Two indicates — Noah currently leads this team in both WP48 [Wins Produced per 48 minutes] and Wins Production.
Table Two: Chicago Bulls in 2008-09 after 35 games
Unfortunately for the Bulls (of the players who have played 500 minutes this year) only Aaron Gray is more productive than even an average player. And Gray plays the same position as Noah (and he’s not offering as much as Noah). At every other position, the players getting minutes are below average. In other words, every other player getting significant minutes is “not good” (at least, not so far this season).
To be fair, Ben Gordon, Derrick Rose, Drew Gooden, and Larry Hughes are not far below average (in fact, if you wanted to think of these players as essentially average, you would be fine). Still, none of these players have actually been “good” this year. And Luol Deng, Tyrus Thomas, and Andres Nocioni have clearly been below average. It’s the play of these players – players who are getting more minutes than Noah – that are primarily responsible for Chicago being a below average team in 2008-09.
And yet Morrisey want the Bulls to get rid of Noah. When we look at all the data, it’s clear that Abbott’s assessment is correct. Morrisey is deceived by “old-school stats.” By focusing only on per-game statistics, Morrisey can’t see the value of Noah. Hopefully, for the sake of Chicago fans, Paxson is not listening to Morrisey. If he is, though, another NBA team might be in position to add a player who might have trouble following the rules (like the important rule prohibiting eating in the locker room) but who certainly can help a team win more games.
- DJ
The WoW Journal Comments Policy
Our research on the NBA was summarized HERE.
The Technical Notes at wagesofwins.com provides substantially more information on the published research behind Wins Produced and Win Score
Wins Produced, Win Score, and PAWSmin are also discussed in the following posts:
Simple Models of Player Performance
What Wins Produced Says and What It Does Not Say
Introducing PAWSmin — and a Defense of Box Score Statistics
Finally, A Guide to Evaluating Models contains useful hints on how to interpret and evaluate statistical models.
60 responses so far ↓
mr. parker // January 9, 2009 at 6:53 am
Thanks for writing a Noah article. I suspect that not many gm’s are willing to add a guy like Noah considering his attitude. Its worth it to point out that if you just let him play his minutes almost every team would be better off for it. Also, its worth it to point out that his wp48 is lowered because of how terrible his team defense is.
Joe // January 9, 2009 at 8:50 am
“The difference between Hughes and the double-headed, inconsistent monster of Noah and Thomas is that Hughes can play a little bit.”
The Larry Hughes that perpetually produces 0 wins and always is one of the 10 most overpaid players in the NBA?
Italian Stallion // January 9, 2009 at 9:20 am
>And when we look at Wins Produced, we see that Chicago is actually getting quite a bit from Noah. In fact – as Table Two indicates — Noah currently leads this team in both WP48 [Wins Produced per 48 minutes] and Wins Production.<
I have a minor problem with making your case partly by using your own model (even though I agree with you on Noah).
Obviously, when two people disagree, it’s often because they are evaluating things in a different way. To then reference your own stats or model as evidence is not going to accomplish much. It’s more or less preaching to the choir than making the point to someone that disagrees. If others disagree, the assumption must be that they disagree with your model to begin with and won’t consider that evidence/proof of anything.
I think pointing out the potential problems with
the thinking of others is more likely to provoke a light bulb going off. However, I suppose it’s not a very good way to win friends.
Mike // January 9, 2009 at 10:56 am
I think as you note Noah is a very good offensive rebounder. He would be a useful rotation player for any team purely because of that. He also blocks shoots at a decent rate. Other than that though I don’t really see how he is much better player than Gray. With his bad attitude and the fact that the Bulls really aren’t doing well as a team regardless, I really don’t see the harm in them getting rid of him.
Rick Morrissey’s article did come across as a little malicious and ignorant, but I think his fundamental point was a good one. The Bulls suck, Noah is not a particularly accomplised player (aside from getting blocks and offensive rebounds) and if he can’t tow the line for the sake of team chemistry then he really isnt worth the hassle.
Tommy_Grand // January 9, 2009 at 11:36 am
I predict he will depart Chicago and some western team will benefit greatly from having him on their bench.
too many steves // January 9, 2009 at 11:49 am
I refuse to take the Chicago Bull’s word for it that anybody has an attitude problem, because for the past few years that team has been coached by Scott Skiles, a world-class a-hole whose attitude would conflict with anybody’s, and now by Vinnie Del Negro, who seems to be trying to follow in Skiles’ footsteps.
Really, how dumb do the Bulls look for not hiring D’Antoni when they could have? If Chris Duhon looks like an All-Star under D’Antoni, how awesome would Derrick Rose be? Chicago is filled with skilled perimeter guys and big men who can run. It’s the perfect D’Antoni team.
Anyway, I’d still take Hawes over Noah, based on what they’ve shown so far. Now, if you had Hawes and Noah on your front line, you’d have the start of a decent team.
Mike // January 9, 2009 at 2:19 pm
If you had Hawes and Noah on your front line, you’d have the start of an expansion franchise. You’d have Emeka Okafor and Primo Brezec. Except worse.
I wonder what happened to Brezec anyway. According to basketball reference he was on 3 different teams last year. Must of been traded and waived I guess.
Paul Alexander // January 9, 2009 at 4:31 pm
Primoz is in Europe playing with Brandon Jennings on some Italian club. They were good friends and Primoz wanted to make sure Jennings was looked after while over there.
Dannie // January 9, 2009 at 8:34 pm
Could you do a similar article on Samuel Dalembert. Most of the people on my blog and pretty much all Philly fans say he is the worst player they have every seen. They say he should be traded for absolutely nothing because that’s what he is worth.
I certainly don’t think Sam is great but he is not nearly as bad as people make him out to be.
Vince Gagliano // January 9, 2009 at 9:13 pm
It might just be me, but if Noah, at 6.6 Wins Produced, is your team’s M2P, that just doesn’t inspire confidence.
I’m giving Derrick Rose the benefit of the doubt. He has the least NBA experience and the highest floor of any of the next level guys.
And speaking of M2Ps, Joel Przybilla over Brandon Roy is fickle at best and laughable at worst. Again, much of his productivity owes to a 70%+ field goal percentage and 12.0+ rebounds/1.6+ blocks per 36 minutes.
If Joe can keep it up, more power to him. But I think Przybilla is just a slump away from cramping Portland’s style, and by extension, the Blazers’ record.
Granted, he’d still be outstanding productively even if he shot 50% from the floor. But if I’m Kevin Pritchard, Brandon Roy is the one getting the big fat contract.
Italian Stallion // January 9, 2009 at 10:24 pm
Danny,
IMO, many casual fans tend to underrate defensive, low scoring efficient players that can rebound.
On the flip side, IMO some people tend to overrate them because their contributions, while important, tend to be more fungible and easier to replace than some other basketball skills.
When you lose a 1o assist man or a guy that’s deadly from the 3 point line, you are unlikey to find a replacement on your bench and the slack is not going to be picked up by the rest of the team.
When you lose a 10 rebound guy that scores 8-10 points very efficiently, you might already have someone on your bench that gives you almost as much and the remainder (or close) might be picked up by the rest of the team.
The thing is, Sam is not even shooting as well this year. I would say he’s been a big disappontment so far this year.
Ryan // January 10, 2009 at 12:05 am
Noah would be much more appreciated playing some minutes next to a really good center. That crabcock of a front court in Chicago is insulting.
Richard W // January 10, 2009 at 5:12 am
I think one thing reason people believe Noah is a bad player is because he leads the league in head-scratching “what the hell was that?” plays. Like the play where Ben Gordon went up for a jumpshot and Noah immediately turned toward the basket to get the rebound just as Gordon threw the ball into his back. Lots of players make mistakes but, because of his high energy style, Noah makes every turnover spectacular. He will come along and he can be a very productive player for the Bulls if he can get some minutes.
Mike // January 10, 2009 at 6:43 am
Stallion, if casual fans underrate these kinds of players don’t you think it is possible that this website tends to overrate them?
I think Vince is actually right to mention Przybilla and Roy as an example of where this model seems to glitch. If you are a low usage, front line player who gets predominantly layups or dunks, and tend to get rebounds you seem to end up with a great winscore. Whereas if you are a guard who shoots less than 50% from the field (or 33% from the 3) and have to handle the ball or pass a lot (resulting in a couple of turnovers a game) then you will be punished by the same formula.
So you end up with players like Przybilla being rated higher than players like Roy, which really doesn’t make sense. The more I think about it the more I tend to think that the other imperfections in PER aside, Hollinger’s model does at least seem to handle scoring better than winscore. If each possession is worth a point, then if a player misses two shots and makes one (33% shooting) then he has cost his team two points (assuming no offensive rebound) and made two for them. So he has broken even. This ‘break even percentage’ drops to 25% from 3. Barring a plus minus system in which a player is penalised if he takes shots from another player at his position who makes those shots at a better rate (not simple) I think these ratings are basically correct. I know Professor Berri is of the opinion that PER is flawed because a player can improve his rating by taking more shots at this percentage (it would have to be higher than this to really get the player anywhere though), but I think winscore actually over compensates by forcing players to make 50% of their shots (or 33% of their 3 pointers) just to break even. I think this, and the fact that offensive rebounds and defensive rebounds are given the same level of importance ( I am of the belief that offensive rebounds are worth more,) is the reason that big, low usage, frontline players like Przybilla tend to be rated higher than guards like Roy.
I could be wrong though.
mr. parker // January 10, 2009 at 8:21 am
Mike,
The problem with your argument is the arbitrary shooting percentages of 50% and 33%. The winscore model uses leauge averages as the baseline. This means that a player is compared to other players and not arbitrary percentages. A player is rewarded for outperforming his contemporaries who play the same position, not for outperforming an arbirtrarily chosen shooting percentage. Winscore has its flaws by definition because the correlations are not 100 percent, but it still far outperforms almost everything else out there. This paradox reminds of something I read by Bill James earlier today about the BCS process. He made the point that whenever the human polls and the computers have disagreed humans have made the input more and more like a machine equivalent to pollster thinking than as a independent measurement of team success. PER is like the newer BCS models whereas wp48 is like the older models. PER tends to agree with Stephen A. Smith, while noone who agreed 100% with winscore could ever find a job on TV.
my 2cents
Italian Stallion // January 10, 2009 at 9:09 am
Mike,
I was more or less trying to make the same point, but putting a different spin on it.
It’s not that rebounds and efficient scoring aren’t important to winning. They are wildly important.
It’s that IMO it’s easier to replace rebounds than some other skills and it’s also fairly easy to find a guy that “is capable” of scoring 8-10 points a night efficiently.
IMO, you have to ask yourself the same question, but from a different direction.
If a player has a skill that’s critical to winning, but it’s easier to come by, how valuable is he really?
IMO the reason SOME “star” players aren’t as efficient as their low usuage teamates is EXACTLY BECAUSE they have greater offensive skills and athleticism. So they are asked to do more on the court. They could easily be even more efficient if they wanted to, but it would hurt the team.
They take more shots, take tougher shots, get double teamed, shoot when the shot clock is about to run out etc…. because sometimes someone has to and they are best equippped. In those cases, their greater skill can actually get punished statistically instead of getting rewarded.
On most models (not sure about here), 10 assists more or less counts the same as 10 rebounds.
However…
IMO, if you lose a 10 assist PG, in most cases your team is totally screwed.
IMO, if you lose your PF whose primary contribution is 10 rebounds and 10 efficient points, you may not even notice he’s gone. You can usually get back most of those 10 rebounds from his replacement off the bench and the other four players will also pick up a little of the slack.
Vince Gagliano // January 10, 2009 at 9:11 am
Mr. Parker:
I do not have a problem, per se, with the fact that Przybilla and Noah are leading their teams in Wins Produced.
What concerns me, particularly with Przybilla, is that a big reason (not the only one, but still a big one) that he is ahead of Roy is contingent on 74% shooting.
Historically, and even Win Scores doesn’t have to show it, building a team around a player who just shoots the ball well never works in the long run. Never.
Because, sooner or later, those players fall into slumps, and much of their value goes with those slumps.
Thus, there are two painful truths facing Portland:
1) Sooner or later, Przybilla will revert back to normal, and Win Scores will have Roy at M2P where he belongs.
2) The Memphis Grizzlies will play Darius Miles against Cleveland and Utah, tack on $18m to Portland’s salary cap, and put a crimp on Greg Oden repeating Bill Walton’s history of winning a title in his third year.
Even a lawsuit will not stop Portland from entering the luxury tax now. Settle things on a different court, guys.
Mike // January 10, 2009 at 9:32 am
Mr Parker
The 50% and 33% figure isn’t arbitrarily assigned by me. Win score subtracts all shots from points scored (not a league average as you suggest.) The result of this is that for every 2 point field goal you can only take a maximum of two shots or you will score negative, so you must have a 50% field goal percentage. For every 3 pointer you must score one per 3 shots to break even (33%.) I agree the figures are arbitrary but they are built into the equation I am afraid. The net result is that this penalises perimeter players as they find it very difficult to break 50% field goal percentages (though not so much to break the 3 point % limit.) I think a good fix would be to only subtract missed shots as nba efficiency does.
I am not saying win score is not a good metric. I come to this site every time there is something new to read so obviously I think it is a good site and Professor Berri does good work. The more I learn however the less I believe win score to be flawless. Wins produced and WP48 however I think still seem to be pretty close.
too many steves // January 10, 2009 at 9:54 am
No system is perfect, but any system that puts The Vanilla Gorilla above Brandon Roy has some serious problems. Shooting 80% from the field doesn’t matter much if you only take 2 or 3 shots a game.
Vince Gagliano // January 10, 2009 at 11:11 am
For Przybilla the Killa, it does matter – he’s averaging 3.1 shots per game. Hey, it’s more than 2 or 3 shots a game, right?
Until then, I find it hard to nitpick about Roy. Give the Blazers credit for being in playoff contention.
Right now, though, the Lakers and the 2nd-7th place teams are in the catbird seat. If the season ends now, Portland has to play LA, and in all likelihood, gets bounced.
Oh, and the Lakers emerge as the biggest winners in the Miles soap opera.
Let’s say the Blazers are projected as a 50-win team and that Miles doesn’t wreck the salary cap. Also, we’ll assume that, for sake of discussion, Portland signs Dwayne Wade with the way their team is constructed now.
If Wade replaces Steve Blake or Nicolas Batum, suddenly, it catapults the Blazers from a 50-win team to a 65-70 win team that directly challenges the Lakers for Western Conference supremacy.
Given that Dave has projected LA as the presumptive championship favorites over the next several seasons, a Portland team signing a big-time free agent is seriously bad news.
But now, not only will the Blazers be unable to enter the LeBron sweepstakes with an $18,000,000 contract, but they have to hand out $7,250,000 in luxury tax to the other 29 league teams, which keeps Oregon’s only pro team in the “pretenders to the crown” category.
Translation: Kevin Pritchard will have to continue building up the team through the draft
Jason E // January 10, 2009 at 3:20 pm
It strikes me as strange how there continue to be arguments built heavily with opinion (case in point, the ‘IMO’ in ItalianStallion’s post regarding his *opinion* as to the ease of replacing 10 rebound vs 10 assist players) when these are empirical questions that can be tested and don’t rely on opinion.
Remember, the wins produced model speaks towards somebody’s contribution to the probability of wins. It shows that the summed probability of wins, over a sufficient sample, comes very, very close to actual wins.
If the model truly undervalues the star scorers who somehow appear inferior by this metric but are, in fact, more valuable in terms of helping a team win, the metric should fall apart and perform poorly in predicting future wins when players change teams. It should underpredict the success of a team who acquires the ‘penalized’ scorers and overpredict the performance of teams who acquire high winscore, low ‘usage’ players.
Does it fail in these cases? That’s not something that these discussions about why Kobe or Iverson or anyone isn’t more efficient relies on. It’s a matter of looking at historical data and seeing how the model has performed, where it has shown greater utility and where it has failed. Saying it overvalues based on any opinion or ‘theoretical/hypothetical’ argument is completely and totally meaningless if the model stands up the actual data, regardless of how much you think that your explanation explains why someone doesn’t shoot better or how easy you think it is to find a 10rpg front court player.
mr. parker // January 10, 2009 at 3:55 pm
Mike,
wins produced uses a position adjustment. The first calculation subtracts for shooting below 50%, however which you adjust for position this is no longer the case. It then depends on position.
Mike // January 10, 2009 at 5:02 pm
Jason E
Stallion is right. If you have a forward averaging 10 rebounds he is more easily replaced than a 10 ast point guard. Not only due to the fact that defensive rebounds are basically a given on a team level as a product of the opponents missed shots rather than the individual player (offensive rebounds are a different story) but also due to the fact that per 48 minutes, there are 68 forwards averaging at least 10 boards. If you look at guards averaging at least 10 assists per 48 minutes, there are currently only 14. So even just in terms of simple supply and demand, a 10 assist guard is harder to replace than a 10 rebound forward.
Mike // January 10, 2009 at 5:22 pm
mr parker
Sorry I don’t follow your comment exactly. However I will say that I was talking about issues I have with winscore, not wins produced. Even so I am quite suspicious of adjusting to position when it would be unnecessary to do so if shooting efficiency was weighted fairly to begin with.
Jason E // January 10, 2009 at 6:02 pm
Mike, whether or not it’s easier to find a replacement for the 10 rb player than it is for the 10 assist player still does not address the relative value of a rebound vs. an assist. That’s the issue with the value. If the values are somehow wrong *regardless* of supply and demand for the actual statistical categories, the predictions made by the model should suffer.
Do they?
I’m curious what you mean too by shooting efficiency being weighted fairly. Fairly in what context? In terms of its impact on probability of victory? Or do you have some other measure that you are trying to get at? It’s not clear when you speak of things being fair what it is that you’re trying to measure. The wins produced model (win score being a rough approximation) is trying to measure the individual contribution to the probability of victory. Measures that attempt some form of ‘fairness’ that deviate from this might be fine if you’re trying to rank players on some other scale, but if you cannot demonstrate that there is some reason for re weighting the coefficients, you’re measuring something different. Since the objective of basketball is still to win the game, I’m curious what your justification is.
simon // January 10, 2009 at 6:30 pm
Jason E:
Exactly. dberri actually addressed it from a academic professor’s view. http://dberri.wordpress.com/2006/07/ He basically said it’s not enough to call a theory with empirical “stupid.” You have to give your own evidence to the contrary or find a weakness in the paper. Also that evidence cannot just be some anecdotal story. It has to be empirical and have statistical backing. (what a novel concept eh?)
One bad thing about communicating through blog is that the comment system naturally encourages people to “yell out” simple responses. It’d be better if those who vehemently oppose dberri’s method can write their own blog article to show some real evidence to show that the WOW model isn’t valid. Just saying “you overrate rebounding” doesn’t really cut it, not to mention that it has been repeated ad-nauseum for years.
simon // January 10, 2009 at 6:32 pm
The link above should’ve been http://dberri.wordpress.com/2006/07/26/the-hogwash-argument/
mrparker // January 10, 2009 at 7:31 pm
Mike,
I’m trying to not to be long winded so I apoligize in advance if it sounds “dickish”. There has been at least 1 player rated above .3wp48 by Berri to win a championship every year with few exceptions. More of these championships have come without a player rated highly by PER, NBA EFF, etc. I think the results have spoken for themselves. If win score is not the superior model then how do you explain last year’s championship. Kobe is the “best player ever”, and he finally had good teammates. To me that was the seminal moment in this argument. A few extra nails are now in that coffin.
Vince Gagliano // January 10, 2009 at 7:42 pm
Here’s a nuance that nobody has covered about Win Scores, and is actually a point that I brought up with Dave:
A made three-pointer accounts for 2 points in Win Scores (3 points – 1 field goal attempt). However, the three-point play is only good enough for 1.5 (3 points – 1 field goal attempt – 1*.5 free throw attempt)
A made two-pointer gives you 1 point in Win Scores (2 points – 1 field goal attempt) But making two free throws after missing a field goal gets 0 (2 points – 1 field goal attempt – 2*.5 free throw attempts)
And only getting 1 of 2 at the line actually results in a -0.5 Win Score.
Okay, I apologize if the next point seems anecdotal, but bear with me:
Theoretically, this measure actually works against NBA superstars such as LeBron James and Kobe Bryant.
Frequently, these stars are more likely to head to the free throw line as a result of the so-called “veteran’s whistle” These plays often arise from penetrating to the hoop instead of outside jumpshots or perimeter shooting. Because their field goal percentage would probably be lower on contested shots than shots that aren’t as contested or are uncontested.
Because they are prime candidates for three-point plays, their Win Scores suffer.
Going back to statistical evidence, consider some stars from last season, their WP48’s, and their free-throw stats.
Chris Paul, M2P:
.406 WP48, 4.9 FTA per game
Jason Kidd:
.352 WP48, 1.9 FTA per game
LeBron:
.321 WP48, 10.3 FTA per game
Steve Nash:
.289 WP48, 3.0 FTA per game
David Lee:
.259 WP48, 2.9 FTA per game
Kobe:
.247 WP48, 9.1 FTA per game
In this ranking is a twinge of irony. That half point lost in the three-point play never disappeared. It just went over as a personal foul to the defender who committed it in the first place.
However, even with the half-point cushion, there are still some missing Win Score points if the player misses the field goal attempt – even if they make both free throws.
Just a counterpoint I wanted to make.
mrparker // January 10, 2009 at 7:49 pm
Vince,
I can’t speak for Dave because I’m not Dave but I’m pretty sure he’s written about how he stopped tinkering with those type of nuances because they didn’t add any significant changes in wins produced.
Vince Gagliano // January 10, 2009 at 7:59 pm
I dunno.
It might be a nuance, but it’s still one to be aware of.
Here’s another one, or couple:
Offensive rebounds are valued exactly the same as defensive rebounds in Win Scores.
If a possession is kept alive by an offensive rebound, it still counts as one possession possession.
But if a possession is ended by a defensive rebound, it ends. End of story.
Long story short: Players who tend to crash the offensive glass are less likely to affect their teams overall efficiency differential than players who are outstanding defensive rebounders, who positively affect defensive efficiency every time they get a defensive rebound.
The change in efficiency can alter the expectations for wins. Changing the wins can change the total overall production of wins, which in turn affects individual productivity.
Enough of these nuances can result in enough of a change in wins produced to throw the system into serious doubt. I’m just saying.
Vince Gagliano // January 10, 2009 at 8:18 pm
And for my first point, here are some other names this season who can provide a guiding line:
Dwyane Wade:
.273 WP48 (est.), 10.1 FTA/game
LeBron (as PF):
.320 WP48, 8.7 FTA/game
Joel Przybilla:
.356 WP48, 1.9 FTA/game
Marcus Camby:
.461 WP48 (!??), 3.7 FTA/game
I don’t care what Win Scores say, as a GM, I am NOT passing up either of the first two in favor of the last two.
Vince Gagliano // January 10, 2009 at 8:19 pm
Addendum: Last statement should be spelled out:
As a GM, I don’t care what Win Scores say, but I am NOT passing up either of the first two in favor of the last two.
Jason E // January 10, 2009 at 10:33 pm
Vince wrote:
“A made two-pointer gives you 1 point in Win Scores (2 points – 1 field goal attempt) But making two free throws after missing a field goal gets 0 (2 points – 1 field goal attempt – 2*.5 free throw attempts) [...] And only getting 1 of 2 at the line actually results in a -0.5 Win Score.”
and also wrote:
“However, even with the half-point cushion, there are still some missing Win Score points if the player misses the field goal attempt – even if they make both free throws.”
Vince, you are wrong here. You are incorrect about the case of 2 made FTs. If you are fouled on a FGA, no missed shot is recorded, so none is subtracted. If the FTs were shooting FTs, the shot is not recorded. Consequently, your math falls apart.
Two made FTs after being fouled on a shot attempt produces exactly the same win score as the made two point shot without a foul.
Mike // January 11, 2009 at 8:43 am
mr parker
you wrote “If win score is not the superior model then how do you explain last year’s championship. Kobe is the “best player ever”, and he finally had good teammates. To me that was the seminal moment in this argument.”
I don’t remember ever seeing a metric which has Kobe listed as the best player ever. Which model are you refering to? I suspect you are actually just talking about hyperbole from sportswriter’s and the media, although even they were hardly refering to Kobe as the ‘best ever.’
You also have to bear in mind that when this blog makes predictions off of ‘wins produced’ it is actually making predictions off of efficiency differential. Efficiency differentials are like the benchmark of team performance, so of course making a prediction on the outcome of a playoff series based on this is likely to be successful.
If you can simultaneously market your model along with these predictions which are based on a measure which can exist independent of it, then when or if it comes true you can convince people such as yourself of the soundness of the model. These people may then refer to ’seminal moments’ which convinced them, unaware they have really just been shown the usefulness of efficiency differentials in predicting the strength of team play.
Back to your point about Kobe being considered the ‘best ever’, if sportswriters say that the Bobcats will win this years title and wins produced says they wont, does that mean wins produced is proved a great model when they don’t? Would you consider that a seminal moment?
I feel like I have been lead further and further into criticism of wow by these arguments when it wasn’t my original intention to do so. The more I think about it however, and the more people here react negatively to any independent thought (Jason Es reaction to Stallions comment for example) the more I realise that there are good reasons this has remained on the fringes within the sports statistics community after more than 2 years activity. It’s almost akin to the ‘natural health community.’ Everything that conventional wisdom knows about medicine is wrong. Natural homeopathic remedies are the best, and any arguments to the contrary are merely proof of the depth of the ignorance of everyone else.
Italian Stallion // January 11, 2009 at 9:14 am
On the suppy and demand issue, I was clear to point out that IMO rebounds are a very important contribution to winning.
My point was that because rebounds are easier to replace than assists, some thought should be given to that fungibility when evaluating the value of players. It’s one thing to say that player “X” contributed “Y” wins because of his 10 rebounds. It’s another to say he’s better or more valuable than the PG that gives you 10 assists every night.
On the Koby/Iverson type of debate people often have, I think you have to dig deeper than the type of generalization I made and also have to look past the stats a bit.
IMHO, Iverson clearly takes a lot of unnecessary poor shots and is not a great 3 point shooter to begin with. So his inefficient high usage hurts his team at times. He’s probably a little better than this model indicates, but not nearly as good as models that give him a lot of credit for all his inefficient scoring.
On the flip side, IMO Koby does tend to get underrated here. I think he could easily increase his efficiency while modestly lowering his scoring and rate better here, but it would hurt his team. The demands placed on Koby at times are huge (even with all the talent he has around him these days).
mrparker // January 11, 2009 at 9:21 am
Mike,
I’m criticizing the notion that scoring as many points as possible as long as you have a fg% above 33 is positive. Maybe we got off track after that point.
I’m a bartender and the Kobe is better than Michael argument pops up at least once a week. There are lots of Laker fans in DC. That is the point that I always start from.
There is really no way to ever settle this argument without blowing up the nba and starting a new league from scratch comple with a draft and team’s named after their owner’s philosophies. I’ve degenerated to just making bets when I disagree with someone and winning most of them based on efficiency based thinking. In that way the proof is in the pudding.
Finally, I encourage independent thinking. What I spend most of my time on this site doing is arguing about using college wp40s as a basis to draft players. I don’t think there as a high correlation between it and nba success at all.
However, I also think that beating your head against the wall trying to prove that there is something wrong with wp48 when there is a very high correlation between its measurements and actual team wins is pointless. If that’s the point of analyzing sports performance than I’m going to get started on my article about how Tom Brady’s 40 time could be better.
Mike // January 11, 2009 at 9:57 am
Sorry mr parker I agree that we got off track, I also agree that it is ridiculous that people believe Kobe Bryant (or Lebron James, or anyone else for that matter) is better than MJ was.
However despite what you seem to think the correlation between wp48 and team wins is actually not that great (I believe about 75%) unless you make a ‘team adjustment’ which brings it in line with efficiency differentials (giving it it’s 95% accuracy rate). In fact there are arguments that many of the other metrics such as nba efficiency would be a better predictor of future wins than wins produced if it included the same adjustments.
Once again this site is very informative, and I find it very useful. But people should not take it as gospel. Especially not some of the more outlandish assertions. Statistical fudges aside, people should use be able their common sense if they find something dubious, and should not be blindly criticised when they do so.
Italian Stallion // January 11, 2009 at 10:54 am
I believe that even Professor Berri would probably agree that statistics and statistical models are just one tool in a larger toolbox that can be used for evaluating players and teams.
Many debates occur when people view statistics as infallible or their shortcomings as discrediting their usefulness.
I think the trick is to identify the “exceptions” to the stats and use other insights to compliment them.
Personally, I think it’s obvious that Roy is both better and more valuable Joel Przybilla. I would feel that way even if I didn’t also believe his shooting is eventually going to revert to the mean.
mrparker // January 11, 2009 at 11:18 am
I agree with you about the blind criticism. However, I don’t think all the criticism on this comment thread has been blind. What goes on most of the time on this board is
1. blog updated with a point against “conventional wisdom”
2. it gets read through a link on another page
3. ANON,
DAVID LEE HAS NOT MID RANGE GAME OR A JUMPER HOW IS HE BETTER THAN SPENCER HAWES????
Lastly, if team defense adds to the accuracy of wp48 correlation whats the beef?
Mike // January 11, 2009 at 12:10 pm
Okay just to finish this up, if adjustments add to the predictive accuracy of wins produced there is no beef except the possibility that the adjustment leads to the high level of accuracy rather than the model itself. Like I said, the wins produced model is reasonably accurate at predicting future wins without the adjustment but with it very good. However if you apply the same adjustments to nba efficiency, or even points per game, you get an equal or greater level of predictive success. So to summarise there is no beef at all. If you agree with the wow method then fine as with the adjustments it predicts wins well. However if you disagree (ie maybe you think low usage big men are overrated in the system), you will know that given the same adjustments other metrics work just as well.
dberri // January 11, 2009 at 12:24 pm
Mike,
What you are saying about the “team adjustment” and NBA Efficiency and other models is not true. You have to be aware that some people have decided to re-label the residual from a model “the team adjustment”. That is not what is meant by “team adjustment” in Wins Produced.
I realize you are new to this website and there is much to read. The amount of material posted here is equivalent to several books. That being said, before you make pronouncements about what Wins Produced does or does not do, you should make sure you understand the model.
Mike // January 11, 2009 at 2:03 pm
Sorry professor, I realise I am new to this and you are clearly much smarter than I am (no sarcasm) but I believe what I have actually said is basically true.
You have stated yourself that there is a 0.99 correlation between PAWS/min and WP48 (which is the reason you can convert the former to the latter as I have been doing.) However as you have told me and as I found out you cannot predict wins accurately using WP48 gained by this method (it seems to overestimate.)
However when you calculate wins produced the ‘long way’ there are adjustments relating to the field goals made by the opponent, team rebounds, assists, personal fouls, and the number of unforced turnovers of the opponent. This is what I called the ‘team adjustment’ and after this is done your result predicts wins very accurately.
I have seen data that the average error between wins produced with the adjustment and actual wins for the 03-04 season as low as 1.67, with a correlation of 0.982. However without this adjustment the average error was 7.1 wins, a correlation of 0.721 ( http://spreadsheets.google.com/pub?key=pLWcAQTLnESuYe23RjQEFUg ) which seems to reflect my experience with wp based on paws/min.
I have also seen an exercise that shows that wins produced with the position and team adjustments actually predicts future wins at a slightly less accurate rate (0.7947) than points per game (0.7981) and nba efficiency (0.8070) with these adjustments.
I am far from an expert but I believe this to be strong indication that the adjustments lead to the accuracy of the model, rather than then weightings assigned to individual box score statistics. This to me means that while still useful, wins produced is not markedly superior to other metrics such as nba efficency when looking at player productivity, as long as you include the adjustments when looking at the team as a whole. Therefore (and this is just an example) people who wish to argue that Joe Przybilla is not actually as productive as Brandon Roy for example, are perfectly within their rights to do so. But you’re right I am new, and not an expert.
Vince Gagliano // January 11, 2009 at 2:44 pm
My apologies about the missed FG confusion.
I always assumed that if it was jacked up towards the basket and it bounced off the rim, it was 0-1.
Hmm, then if (keyword: if) two made free throws count the same as a made two-pointer in Win Scores, might it be possible that the play is more productive given the half-point cushion?
Phil // January 11, 2009 at 7:03 pm
Vince, that would make sense. But believe the logic is that, while and-1 is indeed a boon, most teams do not draw and-1s at disproportionate enough rates to merit their inclusion in WP (or any other metrics I know of). What percentage of FTs come from and-1s or fouls on 3-point shots, which gives a similar bonus? I would imagine probably around 10-20%.
The same goes for offensive fouls – those that draw them do not get credited. (They are counted as both a turnover and a foul for the offender, however, so charges should be worth 1.5 possessions, if my thinking is correct). Drawing more than .3/game is exceptionally high.
I find this omission disappointing. dberri has mentioned on a few occasions that the inclusion of either statistic doesn’t make a significant difference in terms of predicting team wins/losses.
Both are pretty rare occurrences, and fairly equal for most teams. Can you provide stats that show a team that draws significantly more and-1s or charges? Because if it’s the same for everyone, it does not matter in terms of predicting team record.
It does still mean that certain players are more valuable, however, and imo thus merits inclusion. Just because a stat is relatively uniform across teams, does not mean it should be omitted, as stats are about judging players as well. Put it this way: if all teams rebounded the same, would rebounding not be important?
Vince Gagliano // January 11, 2009 at 7:13 pm
But, even here, we hit a crag.
Not all lineups are created equal. Mike D’Antoni used smaller lineups in Phoenix that relied on perimeter shooting in lieu of an outside game, while the Lakers have a frontcourt of two true 7-footers.
Also, different players at different positions rebound at different rates. In his prime, Jason Kidd could grab boards like a small forward, while Larry Bird used outstanding court awareness to compensate for lack of size at power forward.
And, as I mentioned, offensive and defensive rebounding can have differing effects on efficiency differential.
Correct me if I’m wrong, rebounds are valued the same as made two-pointers because of the logic that, on defense, denying an opponent the chance at two points has roughly the same effect as scoring two of your own.
In reality, it’s a little bit less, because merely grabbing an offensive rebound is no guarantee that the basket will count. I might be mistaken, but the rebounding is overvalued to make it roughly the same given this disrepancy.
Jason E // January 11, 2009 at 7:51 pm
“I might be mistaken, but the rebounding is overvalued to make it roughly the same given this disrepancy.”
You are mistaken.
In the WP model, the value of a rebound was determined empirically by regressing rebounds against victories. That the absolute value of a rebound is very close to the absolute value of a point and this is very close to the absolute value of a missed shot makes sense given what they mean in terms of a possession and given that on average, teams tend to score about one point per possession. But it was not through this logic that the values were determined.
Vince Gagliano // January 11, 2009 at 8:45 pm
I should really stop flapping my gums now.
David // January 11, 2009 at 10:09 pm
In the spirit of Vince’s arguments, I’ll rehash a similar problem I pointed out in the past. An assist plus an assisted 2-point field goal should contribute the same value as an unassisted 2-point field goal, as they contribute equally towards a team’s winning chances. This can be accounted for making unassisted field goals more valuable than assisted ones. But WP does not do this. Hence, players that create their own shot are undervalued.
Jason E // January 11, 2009 at 11:04 pm
“An assist plus an assisted 2-point field goal should contribute the same value as an unassisted 2-point field goal, as they contribute equally towards a team’s winning chances.”
…and in the WP model, this actually happens. I would really suggest that you look at the actual model and the steps involved. It does not appear you’ve done so, David.
There is an adjustment that, in effect, takes a portion of the made basket and allocates it to the player making the assist. It is outlined in Berri’s explanation of the computations.
see step two:
“http://www.wagesofwins.com/CalculatingWinsProduced.html”
It is a means of adjusting for the fact that a point is a point (and a rebound is a rebound in the case of a blocked shot followed by defensive rebound) at the team level, but in terms of individual contribution, the player dishing the assist/blocking the shot did contribute to the particular case.
Win Score is an approximation. It winds up being very, very close, but it’s an approximation nonetheless. You appear to be confusing win-score with wins produced, critiquing the second by referencing issues in the former. The approximation (win-score) does not include this adjustment. Critiques of win-score would do better to remember that it’s an approximation of a larger, more complete model.
David // January 12, 2009 at 6:52 pm
Jason E,
You are correct in that WP allocates part of the credit of a made basket to teammates’ assists. But it does not differentiate between assisted FG’s and unassisted FG’s. Only assisted FG’s should transfer credit over to assists. And that is the problem I am pointing out. At an aggregated team level, it all evens out, but at an individual level, credit is unfairly stolen from players whose made FG’s are unassisted more frequently than their team’s average; i.e., from players who create their own shots.
Jason E // January 12, 2009 at 8:56 pm
Once again, I find myself saying the same thing. This model is empirical. If not differentiating between assisted and unassisted FGs is a large problem, then the predictions made by the model should be off.
Are they?
mrparker // January 13, 2009 at 10:14 am
Mike,
I could be wrong but I think in your last argument you are confusing the words predict and reflect.
Italian Stallion // January 13, 2009 at 10:28 am
Jason,
I understand the issue you are debating with David (an interesting observation I might add) , but if I understand him correctly, I don’t think he’s suggesting that the model is off in aggregate.
I think he’s suggesting that the assist adjustments are being done in aggregate. That might take some credit away from some individual players and transfer it to others for a net that is OK, but with issues at the individual level.
Since I am not familiar with the calculations, I can’t say what is happening, but something like that could also explain some of the differences between perceptions and statistics when it comes to a handful of players.
I also wish that someone more familar with modeling basketball would comment on the fungibility of some stats issue that I brought up.
IMHO, supply and demand is a very big issue when it comes to player value.
Determining the contribution of an assist, rebound, blocked shot etc… to wins is one thing, but if some player skills are easier to replace than others, then IMO those players that possess the rarer talents are worth more money and are more valuable.
I have always felt that great PG’s are underrrated because a great assist man is so hard to replace. The loss to a team can be meaningful even though the value of an assist may not be as great as some other stats that are easier to replace.
I could make the same casae about efficient scorers.
Efficient 10 point (or less) scorers are a dime a dozen. In fact, I think a lot of guys that are less efficient than that could easily become efficient 10 point scorers if they were more selective with their shots and scored less.
I would also argue that the extra 10 points an efficient 20 point scorer contribues is more valuable that the the first 10 because efficient 20 point scorers are so rare and teams need scorers.
Jason E // January 13, 2009 at 11:17 am
It’s important to remember what data input the WP model has: box score data, and nothing more. It’s widely collected and easy to parse. With these data, there are limits to what can and cannot be incorporated into the model.
It very, very often sounds like severe critics of the model want to disregard it because it doesn’t make a distinction here or there that *might* (but also might not) have significant impact on the overall results, and want it to include data that are not available to the simple model. If you have the composite stats for the NBA (easily downloaded in spreadsheet format) you can do the computations in excel, with very little complication.
It’s been said before in this forum and elsewhere: “the perfect is the enemy of the good”.
Could a model that incorporated a distinction between assisted and un-assisted FGs, only taking partial credit for the basket away from those FGs that were the result of an assist? Perhaps. It’s entirely possible. But it’s not *necessarily* so. And deriving criticism based largely on how the model differs from popular perception and a series of hypothetical explanations as to why values are wrong that brings it more in line with popular perception is very, very different from finding real fault with the method.
I think it is important to note that methods like adjusted plus-minus, methods that require far, far, far more voluminous data input. You cannot port it into excel and get an answer quickly. It takes parsing of every play to get raw data and successive computation to wind up with results that appear to low consistency over time and have confidence intervals around many of the results large enough to encompass far, far, far too many possibilities.
More data does not always result in a clearer picture, especially when we are not entirely clear on what interactions are at work. In such cases, more data can (and often does) lead to models where the range of possible outcomes is so vast as to render the results meaningless.
So what we’re left with to critique *this model* is whether or not on aggregate, it works and when it doesn’t, in particular, to examine why. The criticism about assisted FGs does not address whether or not the model works, nor does it address a particular when it doesn’t. It addresses a hypothetical cases where it might not work based on the perception of players. This is not the same thing as finding fault with the model. It is a critique that seems to suggest that the model should replicate perception. Such a model is not needed, as it adds no utility at all, but it seems to me that this is the heart of 99% of the ‘critiques’, that it doesn’t show what we already know must be true.
So once again, this is an empirical question: do the predictions made by the model hold?
David // January 13, 2009 at 12:02 pm
Jason E,
I understand that my proposition is not feasible due to lack of data. So I’m not blaming anyone for that.
As for your empirical question about the predictive power of the WP model, I can’t answer that with too much confidence. There are several shining data points (the Iverson-Miller swap especially), but I don’t really know how to measure the predictive power of the model systematically.
The obvious approach of comparing the team-sum WP against the team’s wins seems flawed because of the team adjustments built into the model. I think others have shown that performing team adjustments can make any reasonable model into a very good one according to this metric. So a proper approach would have to focus on cases when players change teams or when players get injured, but I haven’t seen a comprehensive statistical analysis of this sort, just a bunch of isolated examples. Maybe I’ve just missed it.
dberri // January 13, 2009 at 12:51 pm
David,
It is not true that a “team adjustment” can make any model look good. The research you are referencing used re-labeled the residual from an econometric model the “team adjustment” and then proceeded to forecast wins. This is not how we evaluate models. Any collection of independent variables plus a residual will predict the dependent variable with 100% accuracy. You learn this in the first few weeks of undergraduate econometrics.
The “team adjustment” used in Wins Produced is simply team defense. If you add this to NBA Efficiency you do not explain wins as well as Wins Produced.
mrparker // January 14, 2009 at 8:24 am
This argument brings up a problem humans have with “computers”. We are perferctly willing to agree with their outputs as long as the output is along the lines of our preconceived notions.
Bill James wrote an excellent article on this subject concerning the BCS. I can’t remember the exact website but it was linked from footballoutsiders.com
Italian Stallion // January 14, 2009 at 10:53 am
I have no desire to change any model or its inputs so that the results match the general perceptions. I am saying that IMO sometimes it appears that perceptions could be more accurate than the model when you look at the details more carefully.
The Ship Sets Sail: Live Thoughts and Early Impressions of Gallinari « The Ship Be Sinking // January 19, 2009 at 7:11 pm
[...] I’m mostly not a fan of Dave Berri’s analysis for a host of reasons, but he’s absolutely right about Joakim Noah. Even in a loss he was probably the best player on the court today (+4, only Bulls starter with a [...]