A few weeks ago Bill Simmons of ESPN.com referred to Mike Dunleavy – head coach and general manager of the LA Clippers – as “Dumbleavy” (see A dumbfounding night in the ‘Dumbleavy Era’). Beyond the issue of Simmons calling someone else on the planet dumb (see Speeding Up Time for Bill Simmons & I Like Bill Simmons, Really I Do for two examples of Simmons being less than brilliant) is the general idea that Dunleavy is not helping the Clippers.
Apparently the disenchantment with Dunleavy has reached a point where Isiah Thomas is considered a person who can help. Last week Chris Sheridan wrote an article detailing the contact between Isiah and the Clippers. This article included the following two paragraphs outlining where Dunleavy has failed:
Dunleavy has generally won praise for his salary cap management and his most recent personnel moves such as signing Baron Davis, acquiring Zach Randolph from New York for Tim Thomas and Cuttino Mobley, drafting Eric Gordon and acquiring roughly $2 million in cash considerations over the course of this season from teams dumping salaries.
Dunleavy’s coaching is actually the area where the most justifiable criticism could be directed. The Clippers entered Wednesday night’s game against New York 37 games under .500. He has clashed with some Clippers players, most notably Davis and Chris Kaman, although Sterling has been publicly supportive of Dunleavy and overtly critical of his players, most recently when he went on a postgame rant in the locker room after a loss to San Antonio earlier this month.
Sheridan appears to be arguing that the issue isn’t Dunleavy the general manager; but rather, the general manager has been let down by Coach Dunleavy. In other words, Dunleavy has picked the right players. The players, though, are not living up to expectations.
Two Views of the Clippers
To see if this is true, let’s compare what the Clippers have done in 2008-09 to what we could have expected if we made the very simple assumption that what we saw last year from this team’s players we would see this year.
Table One: The LA Clippers in 2008-09 After 74 Games
As Table One indicates — given what these players did last year — the Clippers could have expected to win about 31 of their first 74 games in 2008-09. Instead the team’s Wins Produced stands at 16.5. If we look at the individual players we can see that about 80% of the difference between these two views can be tied to the play of just four players: Baron Davis, Chris Kaman, Ricky Davis, and Al Thornton.
Of these, three – Baron Davis, Ricky Davis, and Kaman – have been injured. And Thornton has gone from being an unproductive rookie – yet first team All-Rookie member – to a less productive sophomore. Had the injuries not happened, and Thornton just repeated the miserable performance from his rookie season (which I might have mentioned was still good enough to get the coaches to vote him on the All-Rookie team) the Clippers might be closer to 30 wins right now. And in the Western Conference that would be good enough still miss the playoffs. Unfortunately, that’s where they are now. So even if performance stayed constant from 2007-08 to 2008-09, Dunleavy – or someone on his staff – would still be at the lottery.
Pareto Misses the Clippers
Of course the injuries haven’t just limited production. The injuries to Kaman and Baron Davis have also limited minutes. A similar story can be told about Marcus Camby. Here is a projection of what would have happened if each of these players had been available the entire season and performed as they did last year (with minutes per game from this year):
- Marcus Camby: 31.5 minutes per game, 2,331 minutes, 17.7 Wins Produced
- Chris Kaman: 30.0 minutes per game, 2,223 minutes, 10.8 Wins Produced
- Baron Davis: 34.8 minutes per game, 2,574 minutes, 9.5 Wins Produced
From these numbers we can see that the Clippers could have expected from this trio – based on last year’s numbers – about 38 wins after 74 games this season.
The Pareto Principle – as applied to basketball – tells us that about 80% of a team’s wins are produced by 20% of a team’s players. Applying this to the Clippers, if the team received 38 wins from its top three players, and the rest of the team provided 20% more wins, this team would currently have close to 47 wins.
Looking at the rest of this roster, though, it seems unlikely that would have happened. Once we get past Baron Davis, Kaman, and Camby, only Zach Randolph and DeAndre Jordan are above average. And these latter two are in the frontcourt with Kaman and Camby. When we look at the rest of the roster, we generally see players that are not only below average but well into the negative range. So although injuries have limited what the Clippers have done this year, even without injuries they would not be as good as the performance of their top players suggests.
Who is Dumb?
So which Dunleavy is to blame for the Clippers problems? After Elton Brand departed for the 76ers, Dunleavy had a problem. The Clippers last year had one player – Chris Kaman – who played significant minutes and posted a WP48 that exceeded the 0.150 mark (his WP48 was 0.233 last year and after 481 minutes this season – or before he got hurt — Kaman’s WP48 was 0.200). Obviously if Kaman is the only player significantly above average on your roster you are not going to be too successful.
Faced with this problem, Dunleavy went out and acquired Marcus Camby (for a gift certificate to the Olive Garden) and signed Baron Davis. These gave the Clippers a trio that could be the foundation of a competitive team. Unfortunately, Dunleavy the GM stopped at this point. The remainder of the roster was populated by a number of very poor players.
Putting the entire picture together, had the Clippers stayed healthy they might have contended for a lower seed in the Western Conference. But I think injuries – not the coaching of Dunleavy – caused that meager dream to be dashed. All of this means that maybe Dunleavy isn’t the most brilliant basketball mind in the world, but the word “dumb” doesn’t apply either. As for certain basketball writers, well….
- 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.
40 responses so far ↓
Dan // March 31, 2009 at 5:06 pm
Hi Mr. Berri,
Bill Simmons certainly knows his basketball. Watches games religiously, does understand many of the finer points of the game that aren’t reflected in the numbers.
My problem is when you call someone “dumb”, based solely on your debatable system. When I saw your article last week on Billups, that glosses over the fact that he somehow lost about 40% of his effectiveness over the past season, without questioning how the change of team style affects this reduction……its just another big hole in your work.
There are clearly team dynamics that are bleeding into the individual numbers. Detroit sets more effective picks, boxes out better collectively, plays better help defense. They play a different game than Denver does. None of this is accurately reflected in the numbers. The idea that a guy happens to lose 40% of his effectiveness the year he is to a different team/system/concept is kind of ridiculous on its face.
I like your work…..I really do. But you should never try to act like its so superior to actually watching the games, way too many things aren’t reflected by stats.
Anon // March 31, 2009 at 6:36 pm
I agree that DJ should probably tone down his negative attitude towards mainstream media sources. He does occasionally come across as condescending and obnoxious, which I know for a fact has turned off some friends of mine to his work.
That said, although I am a huge fan of Bill Simmons, a lot of what he says could be called into question…
Matt Walters // March 31, 2009 at 6:37 pm
Dan,
You appear to be making an error that Dr. Berri (and many other economists) counsel against: stating that a statistical result contrary to one’s own expectations is prima facie “ridiculous.” I believe there was a discussion about this in the comments for a recent post.
As I recall, Dr. Berri did offer an explanation for Billups’s decline: his age. I’m inclined to accept that explanation over your own (or at least, not accept your explanation) because Billups’s WP48 has declined gradually over the course of the season, not all-at-once as we would expect from a player suffering from a new style of play. In fact, with a little research, one can easily see that this year’s Billups began his Denver career posting a WP48 that was nearly identical to his WP48 in Detroit prior to the trade.
The other issue I would like to call attention to is the idea that statistical analysis is inferior to “actually watching the games,” or, to put it another way, that watching games and creating statistical models are in conflict as far as determining player value.
So far as I know, neither Dr. Berri nor any other serious statistical analyst has ever called on anyone to stop watching basketball games. In fact, I speculate confidently that the group of people who are interested in basketball statistics fits neatly into a larger group of enthusiastic basketball-watchers. And yet, other members of this larger group (Bill Simmons among them) repeatedly demand that these analysts stop using statistical models to make judgments about player productivity. This willful self-deprivation continues to make my head spin. If anyone is “dumb” here, it’s he who insists that we throw out the data in favor of sole reliance on our eyesight.
Also, it’s obvious that pointing out the irony of Simmons calling someone dumb was a joke, you dummy.
to matt // March 31, 2009 at 10:06 pm
not that I support Bill Simmons in this argument, but to say that “And yet, other members of this larger group (Bill Simmons among them) repeatedly demand that these analysts stop using statistical models to make judgments about player productivity” is untrue at best. read his last article. you’ll see that he’s enthusiastic about statistical evaluation and nowhere does he ask any analyst to stop using their model(s) (except Hollinger – for entertainment reasons).
that said, I still view Berri’s model as a superior estimate more than any expert opinion.
and that said, I think all NBA stat models have a long long way to go before they can be used to make judgment professionally (e.g. as GM) about players’ productivity.
For example, I’m still waiting for the first reliable non-linear regression analysis metric.
it surprises me that I have not seen any non-linear formula in this context already.
mrparker // March 31, 2009 at 10:44 pm
Count me as someone who liked Simmon’s work much more before he started doing the podcasts. I dont know what Berri did to him, but he obviously has it out for him in some sort of way.
thepeaceblogger // April 1, 2009 at 6:24 am
@Dan
“My problem is when you call someone “dumb”, based solely on your debatable system.”
This quote is exactly why on the Billups/Iverson post last week I went as far as saying that DB sometime use his statistic method and probably his PhD to have a condescending attitude toward anyone who disagree with him. Bill Simmons is a great sport writer and most people online (like me) just enjoy his ESPN articles on any sports and specifically Basketball.
@Anton
“I agree that DJ should probably tone down his negative attitude towards mainstream media sources. He does occasionally come across as condescending and obnoxious, which I know for a fact has turned off some friends of mine to his work.
That said, although I am a huge fan of Bill Simmons, a lot of what he says could be called into question…”
Again I agree totally that DB should not always attack sport writer just on the fact that their fact doesn’t always match what Dr. Berri said or whatever, I mean when I read something from Simmons, I do not always agree with it but I definitely enjoy the lecture and specially the funny part of it.
Take care,
shizumaru76 // April 1, 2009 at 6:55 am
The sports guy writes some very entertaining stuff, but he can be pretty condescending too. He’s bashed Berri’s work for several very frivolous reasons, including the fact that wins produced ranked Jerome Williams highly. Win score only ranks him highly because he puts up good per minute box score stats, which is what most people use to evaluate players anyway.
And this bashing was done at the time when Simmons was habitually using selective box score data to back up his own ideas of player evaluation, such as the rule of 42, etc. The difference was that win score looks at the same data in a more careful and comprehensive fashion.
On many occasions Simmons also takes shots at other people whose ideas clash with his own, so I don’t see how it’s unfair for Simmons to take some heat when he’s inaccurate (which is a lot) or being narrow-minded (also happens a lot).
I agree WP48 deserves some criticism, but if someone blows it off with out at least giving it a good look and weighing its pros and cons carefully, do they care more about learning about basketball, or do they care more about self-promotion and hiding their lack of understanding and knowledge? If the latter is the case then I think any bashing is well deserved.
JohnG // April 1, 2009 at 9:34 am
Simmons is an entertaining writer, but the strengths of his columns are generally humor and his willingness to write about somewhat offbeat stuff. His analysis is always poorly thought out and sometimes just irrational. He also presents it in a very condescending way.
It’s funny that Simmons has now positioned himself as a “stats guy.” Through the years his writing has shown a poor understanding of what I’d call basic, simple statistics. For years he objected to “advanced” baseball stats and insisted on “intangibles” and “clutchiness.” Now, the advanced stats are “easy to prove” and “obvious,” which strikes me as somewhat condescending toward the work of people like Nate Silver. I’m all for people learning and changing opinions with new information, but he should be a lot more honest about what he used to believe and a lot more respectful of the work of people who “showed him the light.”
His writing/comments about Professor Berri have been rude and obnoxious for years. And yet, Simmons has not bothered to present a detailed, well thought out criticism of Wages of Wins. It’s hard to blame Berri for the harsh tone.
simon // April 1, 2009 at 10:35 am
I can’t believe what people are writing here. dberri didn’t say “Bill Simmons is dumb.” He was merely implying that Bill Simmons’ juvenile shot at Dunleavy (“Dumleavy”) was more justified on himself by parodying Simmons’ choice of the “dumb” word. And you guys are using that add on the on-going personal attack on dberri for being a “condescending” “arrogant” and “obnoxious” guy.
I have been reading this site for a long time now and I never thought dberri was obnoxious. Well, by the usual college professor standard anyway
but still, usually to impress a college prof., you usually need to provide evidence with statistical relevance and solid evidence(i.e. researched large set of data) to back it up, not just thought-experiments and anecdotal stories.
I know this blog is about those stories rather than theoretical discussions, but dberri at least has a book and papers to tell us more about how he has arrived at his current model. Do I agree with his model 100%? No way, but it certainly does have a decent amount of researched evidence behind it and seems to do a pretty decent job of predicting how a team would perform.
Italian Stallion // April 1, 2009 at 11:18 am
“You appear to be making an error that Dr. Berri (and many other economists) counsel against: stating that a statistical result contrary to one’s own expectations is prima facie “ridiculous.”
That’s a legitimate point. However, it starts with the assumption that the statistical model is correct. That is often clearly false.
People with unusually good analytic skills are often very good at measuring things using their intutive abilities. When there is a huge consensus among experts, admittedly they are sometimes wrong. But sometimes the model that says they are wrong is the real culprit.
For example:
I am 100% convinved that DBerri is correct when he states that PER overrates inefficient scorers. I think he has even laid out the case in almost indisputable terms.
However, I think his own model overrates efficiency (or underrates scoring in general).
The most highly skilled, creative, and atheletic offensive players are the ones with the ability required to generate enough offense to win a game. They take the shots that low usuage efficient scorers should not be taking and in most cases don’t even have the skill to create.
To me, all the arguments and studies that suggest otherwise are total nonsense even though I lack the skill to produce the statistical study that proves it.
All any player has to do is go on the court and try to up his PPG by “X” and he’ll immediately find out that he has to take some shots that will lower his eFG% in order to do so.
You can’t get all layups and dunks. In some situations the idea is to find the player that is the least bad alternative.
You have to account for these things somehow.
IMO, none of the models I have seen to date is weighing the usage/scoring/ efficiency issue correctly yet even though most are attempting to do so.
Granted there are players like Allen Iverson and others that get overrated by UNsophisticted fans and sports writers, but the smartest coaches and fans can tell both intuitively and by watching games who the better players are even though none of the models has it quite right yet.
mrparker // April 1, 2009 at 12:15 pm
Stallion,
I like the phrase “overrates efficiency”. You can’t have 5 guys that average 10 rebounds a game on one team. You can’t have 5 guys that can only score from 10 feet in. I think we can all agree on that. What we can’t agree on is the compromise.
My personal opinion is that it is better to be a guy who can make the lower percentage shots but doesn’t want to take them if he doesn’t have to. I think thats what win score represents. There is a reason why a guy like Lebron James who “can’t shoot” has such a high shooting percentage.
When building a team a general manager must keep the law of diminishing returns or some basketball bastardization of it in mind. My belief if its not so much about trying to build an entire team of wow all stars. Instead I think the optimum team is built one piece and one decision at a time. You gonna give me billups for Iverson? I’ll take it. You want to make every team start over from scratch and draft all available players? I’m not so sure wow could help you build a successful team.
JohnG // April 1, 2009 at 1:31 pm
I also don’t understand the anger directed toward Berri. It’s fine to take issue with his model, but the personal attacks are way over the top, especially considering that the guy provides a free service for your entertainment.
ilikeflowers // April 1, 2009 at 2:57 pm
All these brain-based constantly changing models (the source of most opinions) aren’t of any use unless you’re going to put them to the test. Predictions please.
ilikeflowers // April 1, 2009 at 3:00 pm
No details needed, just black-box it and give us some predictions so that we can judge your opinions appropriately.
dustin // April 1, 2009 at 3:13 pm
Are you going to post anyting about the WSJ article about assist inflation? I saw you were quoted in it.
My first reaction is that the pace in the 60’s was significantly faster than now. They are talking about adjusting Oscar robertson’s numbers for the 5% change in assisted percentage, but there is no mention of the 125.5 possessions/game in the 60s compared to the 91.7 possesions/game right now.
from the article:
Adjust Mr. Robertson’s numbers to Mr. Paul’s era, and Mr. Robertson moves to second place all-time in total assists behind John Stockton of the Utah Jazz, who played from 1984 to 2003.
“These guys want to promote the game and make it about new stats and new records,” Mr. Robertson says. “They don’t want yesterday’s stars to be today’s stars.”
kevin // April 1, 2009 at 7:31 pm
Simmons has a huge problem with interpreting statistics. His baseball analysis is laughably naive.
This is the thing with Simmons: he watches basketball games and makes certain assumptions. Then he bases the legitimacy of statistical analyses on how closely they align with his assumptions. So of course he’s going to have a whole host of batshit opinions. He never holds them up to the light of critical judgement.
andert // April 1, 2009 at 10:58 pm
I agree that dberri is very condescending and that can be a turn off.
My question to you, dberri, is where does on the ball defense come into play, if at all?Say Steve Nash, one who has a reputation as an awful defender, and Devin Harris, one who has a reputation as a great on-ball defender, have the same WP48. Do you believe the players are equal and produce the same wins toward the team?
I always wonder how much you believe in your system. Not necessarily believe, but how exactly accurate is. Do you believe nothing else matters? If you don’t believe nothing else matters, how much does the other stuff matter? Is there a certain +- that I can add to any players WP where you would not be comfortable saying that any player is more productive than the other? (For example, if somebody has WP of .250, should I assume that he is roughly equal to anybody within .1 of his? .2?
I am just interested in how exact of a science you view this system.
Mike G // April 2, 2009 at 4:24 am
I had Dustin’s question, too. Big O benefitted from many more possessions per game. Shouldn’t his assists be adjusted DOWN?
kevin // April 2, 2009 at 7:01 am
“I also don’t understand the anger directed toward Berri.”
Expertise really annoys some people. Listen to Fox news sometimes about how they feel about academia.
dustinc // April 2, 2009 at 8:27 am
Andert,
Individual defense is not addressed in wp48, something that db has always acknowledged. However, if you are a good defender in a team system, (any cavs player or boston player), that will be reflected.
There is a post a while back about using adjusted? plus/minus to measure an individuals defensive impact, and dberri found it only changed a players wp48 by 10% or so in either direction I believe…
Tim // April 2, 2009 at 8:57 am
The worst thing Simmons does to Prof. Berri is refuse to mention him or his work by name, even when he is criticizing it. Thank heavens for TrueHoop. Maybe Prof. Berri has decided to try to provoke Simmons into acknowledging his existence.
It’s too bad that people who do want to analyze basketball with statistics seem to argue with each other as much as they do with people like Simmons. It reminds me of the financial mess we are in, although with less serious results — the experts often argue with each other, while the journalists cherry pick the expert-sounding opinions which seem to support their gut feelings or ideology because they don’t understand what the experts are arguing about.
Tom Mandel // April 2, 2009 at 9:03 am
If a statistical method that evaluates the number of wins a player produces correlates to a high degree with the number of games the team actually wins, then it is prima facie a winner. That’s what WP does.
Frankly, if you don’t agree with that, it’s hard to know where to take a conversation with you. Either you think that the stats don’t distribute responsibility correctly, which is hard to claim, or you think that contribution to wins isn’t a measure of how good a player is. Now, admittedly, basketball is both competition and entertainment. There are guys whose productivity *looks* higher than it is because of how entertaining they are: e.g. extraordinarily athletic or fast or aesthetic.
As to the claim that it’s possible to “overrate efficiency,” that made my jaw drop!
Oren // April 2, 2009 at 9:34 am
“or you think that contribution to wins isn’t a measure of how good a player is”
To a certain degree this is true. For example, people are arguing that Granger should be considered a shooting guard. However, since Indiana doesn’t have any better options, Granger frequently plays small or power forward. As such, Granger contributes fewer wins to Indiana then he would if he played shooting guard. Although, it seems that Indiana thinks that his playing at PF maximizes the amount of possible total wins given their roster.
Likewise, if a team decided to start Camby at PG, I suspect his WP would decrease. But it wouldn’t mean that he’s become a less effective player only that he wasn’t used effectively.
I always had the impression that this model was supposed to explain the “hows” and not the “whys”. I’m not sure that this has been made fully clear as illustrated by some of the questions above.
The model shows that a player like Varajao hits most of his shots. That shows us how he performs. Knowing that he hits most of his shots because he can mostly shots layups is why he performs the way he does. I would think that you need to use both the how and the why to understand how good a player is.
It seems to me that some people want to primarily focus on the how while others primarily focus on the why.
Ken // April 2, 2009 at 9:39 am
The bones picked with WP48 always follow one of two paths:
1. player “X” is really talented, can create his own shot, and is a nightmare to guard, so he is really good… Why isn’t all this “usage” reflected in berri’s model? (Kobe Bryant/Allen Iverson argument).
1. player “X” only rebounds/blocks shots, and can’t do anything else (cough *score* cough)… why does Berri consider him a great player? (Rodman/Jerome Williams/Marcus Camby argument).
Want to know why dberri never responds to these critiques? I would imagine it is not because he doesn’t question his own system. It is because he is tired of answering the same questions over and over again. Regular readers tend to have the same response: Please read the book…
Arturo // April 2, 2009 at 10:30 am
The biggest problem that simmons seems to have is that he claims the stat models based on the boxscore like wp48 can’t account for all things on the boxscore. This is a common misunderstanding of people who don’t get statistics. The basis of modeling is based on correlating the data we have readily available (the box score) in the best possible manner to the desired result (in this case wins). We do this by building a series of models and evaluating the correlation (i.e. how well the predicted outcomes line up with the actual outcome) or accuracy of the models and picking the best one. The professor has earned my admiration because he has published detailed notes on how he built his model (others have not *cough* Hollinger *cough*) and subjected himself to independent review on the internet. There is no fudging involved & no interpretation of defensive coeeficients ( for example hollinger again). What I think bothers laypeople is that they think that you have to measure every factor to accurately model a result. You don’t. The pareto principle crearly establishes that not all data is created equal, some factors (the box scorestats) account for a majority of the variation so even if we measure only say 40% of the possible stats it is entirely possible to predict >95% of the result. Again as with any good model the proof is in the predicted results. How good are the predictions made using the model?
Prof. Berri posts his predictions in a public forum and stakes his reputation as an economist on the accuracy of a well built statistical model. The model has not shown itself to be deficient and in fact if you review the results over time way more often than not has been shown to ge fantastically accurate at lining up wins to stats.
Are there opportunities? Sure, you could make a convincing argument that the biggest problem would be something called multicollinearity. This is a phenomenon by which two variables (or stats) are highly correlated and we end up wrongly assigning value to one stat (say a measured one like off fg%) where it should be assigned to another stat ( say a non-measured one like quality defensive possesion). By this argument, the weakness of the model lies not in predicting the number of wins for a team but in accurately diving those wins among the players and assignin an accurate and transferrable value to those players. It would be fascinating to have the professor analize the predictive strength of the model for wins for those assets (players) that are transferred from one team to another. You would have to account for age,position changes and injury but it would provide a clearer picture of the how big of a factor multicollinearity is . The more accurately future value can be predicted for those assets the more likely we have the right variables (stats) being measured.
mrparker // April 2, 2009 at 10:41 am
aturo,
Please stop making sense if you are going to comment in this thread.
Arturo // April 2, 2009 at 1:33 pm
Apologies.
I think the biggest opportunity with the wow model is just what you said :
“You want to make every team start over from scratch and draft all available players? I’m not so sure wow could help you build a successful team.”
I’m not sure either and that’s the main difference between this and something like pecota. The wow model, i think, has been proven empirically to properly estimate the value in wins of a roster as it is currently configured but I don’t think it has been shown (*****yet*****) to do the same for the individual value of a player. If I was a graduate student in sports econimics that would be my thesis proposal. Can you use the model to predict the transferrable value in wins of the individual components of the roster? Also can you make predictions on the value of re-assigning minutes between the assets? My hyphothesis is that yes you could make a reasonable assesment of value (but probably not as accurate as the value assesment of the roster) but some adjustments (which could be calculated) would have to made for position shifts, sample size and confounded stats.
This really would be the tweak that takes the model to next level in my opinion. The value proposition for teams isn’t really in assesing the current value of their roster but in evaluating the cost benefit of possible roster adjustments (playing time,trades) and the real economic values of player contracs (which would require building a sliding model for player performance by age and position).
But then again,I’m only posting for laughs beacuse I’m bored at work.
kevin // April 2, 2009 at 4:51 pm
I like arturo’s responses too. Nice job, kid.
kevin // April 2, 2009 at 5:01 pm
FWIW, I ripped Simmons a new one in an email I sent him June last year when he was lauding how good the Lakers were and how unstoppable Kobe was and how they were going to beat the Celtics in 6. He must have gotten lots of those because he mentioned all the vitriolic responses to his column a couple weeks later.
Rob O'Malley // April 2, 2009 at 5:52 pm
Yeah I think Berri doesn’t often reply on this forum for a couple of reasons.
1) He has a career and a family I’m guessing and that most likely takes much more time and priority than this blog.
2) There is no reason to respond to these seemingly first time readers that demand answers to their dumb questions when Berri has thoroughly explained almost any question people have had at some point in the history of this blog. It seems like these persons main objectives are just to criticize Berri instead of being in search of actual answers because answers are readily available in previous entries.
3) He probably knows his regular readers are already fully aware of #2 and will be more than able to explain these things. (See Arturo.)
Side Note: “The pareto principle crearly establishes that not all data is created equal, some factors (the box scorestats) account for a majority of the variation so even if we measure only say 40% of the possible stats it is entirely possible to predict >95% of the result.” This was a great point made by Arturo. Malcolm Gladwell in his book Blink described this idea very well. He explained how a doctor in Chicago who had the responsibility of making his intensive care unit more efficient. So he implemented this formula that predicts heart attacks with 95% accuracy. It’s based on only four factors, I can’t remember what they were right now. But it ignored things like age and previous illnesses and stuff and other doctors thought that was crazy. But it worked because it predicted so accurately and made that part of the hospital much more efficient. I can’t really do it justice here. But if you want an example of what Arturo was talking about I recommend reading Malcolm Gladwells Blink.
dberri // April 2, 2009 at 5:57 pm
For what’s its worth…. I do read all the comments. And Arturo made a very good point about modeling today.
Rob is also correct. Responding to everything in the comments would be both a) time-consuming and b) very repetitive.
Matt Walters // April 2, 2009 at 6:28 pm
“The wow model, i think, has been proven empirically to properly estimate the value in wins of a roster as it is currently configured but I don’t think it has been shown (*****yet*****) to do the same for the individual value of a player.”
Arturo, where are you getting this idea? If this was all WoW did, Dr. Berri could just post each team’s efficiency differential and call it a day. The whole point of WP is that NBA player production is so consistent from season to season that predictions about “the transferrable value in wins of the individual components” and “the value of re-assigning minutes between the assets” are easy. That’s why Dr. Berri makes such predictions so often on this blog, and why he is so often correct.
I think your point about multicollinearity would be a good one if traded players and players with a jump in minutes suddenly started posting wildly different WP48. But….they don’t.
Arturo // April 2, 2009 at 7:09 pm
First,
To Prof. Berri , thank you. I just can identify with how hard it can be to explain to people that building a useful predictive model isn’t about getting all the available data (which can be ridiculously expensive and time consuming) but finding the most statistically significant points of information.
To Matt
My point is that there is an argument that can be made about the actual breakdown of the wins to the players. Let me give an example, I could argue that someone like Kevin Garnett has a significant impact on the win scores of the players around him because of the impact he has on his team’s defense. Therefore he would be undervalued by the metric because the net impact of removing him from a Team would be greater than predicted by the model. Now i’m not saying that you would expect wildly different numbers from KG or his teammates if he got traded but there would be some impact. The fascinating part would be quantifying this impact (The flipside to this might be allen iverson
See, I think the model provides a very accurate measurement tool of your team as it is currently constituted but ,and this is just my gut feel from my own work experience , a slightly less accurate picture of how your team might be constituted in the future. What would be interesting would be to use the existing examples of roster changes to determine if a correction factor is needed and what it would be (Call it the good teammate/bad teammate postulate) . Somebody like Jason Kidd or James Posey could be a good test case. Prof.Berri or someone else might have already done the work and I just haven’t seen it (maybe it’s in the next book)
mrparker // April 2, 2009 at 8:23 pm
Arturo,
what you’re a looking for is definitely out there and easy to quantify. I’m actually shocked that noone on this blog has picked up on it yet. Its what I use to predict pro success according college numbers and it has been very accurate though I lack the statistical skill to properly quantify the result of “very accurate”. I’m hesitant to throw anything out there because I’m enjoying being the only one.
Jason E. // April 2, 2009 at 8:34 pm
Arturo, great comments.
Since he didn’t seem to say it himself, I’ll repeat something that Dave has said himself. WP tells us who was more productive (and I’ll add that it tells where that production came from) but it doesn’t tell us why. If someone is misused by his coach, this could be a part of the ‘why’.
Evan // April 2, 2009 at 10:08 pm
Arturo, are you applying to be Prof. Berri’s research assistant?
You really said alot of what I was thinking, but a million times more eloquently.
I think it’s important to note that the model doesn’t explain 100%. Could you tweak the model to get it closer? I seem to recall Prof Berri insinuating that he thought it was possible, but because it would not add very much, simplicity is better. And he’s probably right, for his purposes.
That said, if I was a GM (sidenote: NBA owners, feel free to contact me) I’d be tinkering to see if I could improve on 95%…although it’s completely unnecessary right now because based on Berri’s model, the NBA is still completely inefficient.
Matt Walters // April 3, 2009 at 5:29 am
Arturo,
Funny you should mention Jason Kidd and James Posey- their WP48s were virtually unchanged after switching teams. Garnett’s WP48 went up during his first season in Boston (slightly- certainly within the expected year-to-year variation) but his former teammate didn’t seem to lose anything after leaving the team (and his hypothetical sphere of influence). Furthermore, both Minnesota’s and Dallas’s untraded players exhibited very little change in productivity, and on both teams some players offered slightly more while others offered slightly less.
I point this out because 1) I don’t think you are going to find a significant or consistent enough difference to make any corrections or “tweaks” to the model based on your gut feelings about multicollinearity, and 2) I think you just haven’t read enough of this blog to realize some of your questions have already been settled. This is not to say Dr. Berri’s models are perfect (what models are?) but rather to encourage you to become more familiar with the established extent of their predictive power.
Arturo // April 3, 2009 at 9:43 am
Evan,
If I wasn’t getting paid to do analysis as a fulltime job and was going to do a doctorate? I would love to be the professor’s assistant (I do this kind of thing to unwind).
Matt,
I’m going to cherry-pick data a little
http://www.wagesofwins.com/Boston0708.html
Wins Produced for the Celtics 07-08:
Projected based on 06-07: 51.9
Actual: 68.3
Change:16.4 (of which KG is responsible for 3.3)
His teammates as a whole improved from a projection of 36.6 wins to 50.4 wins produced (a 38% improvement). I’m going to say that probably falls outside of expected variation.
As for Kidd there is a before and after difference for the Mavs as well (as a group they are worse than projected)
http://www.wagesofwins.com/Dallas410809.html
Wins Produced for the Mavericks 08-09 (Mid-season):
Projected:66.2 Actual:44.8 Delta: -21.4
Kidd is responsible for -2.6
His teammates as a group: 42.8 (Proj.) 23.9 (Actual) (or a 45% decline as a group )
Again outside what you would expect as normally occuring variation. Seems like Kidd was a bad fit for that team after all.
Matt Walters // April 3, 2009 at 11:38 am
Arturo,
http://dberri.wordpress.com/2009/03/22/aging-billups-and-telling-stories/
http://dberri.wordpress.com/2008/09/24/jason-kidd-really-did-help-the-mavericks/
I encourage you to read more posts.
mrparker // April 5, 2009 at 9:19 am
On the Mavs,
It seems that Josh howard has been injured all season. Im not sure because I haven’t watched them play, but he has only played 47 games thus far and at nowhere near his previous level. Knowing how one guy can hurt team defense and knowing that he was their third best defender last year(according to defensive rating at bball ref) maybe the decline can be almost entirely tied to him not being the defender he once was when he is on the court.