The sports media has declared LeBron James as the Most Valuable Player in the league. If we define MVP as Most Productive (which seems reasonable), and we define Most Productive in terms of Wins Produced (which seems reasonable), then the sports media is correct. That is, the media is correct if we are looking at just the Eastern Conference. If we expand our view to the entire league, though, then an argument can be made for Chris Paul.
The MVP of Each Team
At least, that was the subject of the last post. For this post I want to look at the MVP of each team. Again we are going to define MVP in terms of productivity. And again productivity is defined in terms of Wins Produced. The results of this analysis are reported in Table One.
Table One: The Most Productive Player on Each Team in 2008-09
There are a number of stories one can tell from Table One. Here are a few (in no particular order).
- The average top player on each team produced 14.3 wins, or 36% of each team’s total.
- The one factor that dominates perceptions of performance is scoring. Of the 30 top players listed in Table One, twelve were the leading scorer on their team.
- Of the players who received some consideration for the MVP award, only Tim Duncan and Chauncey Billups was not the leading scorer on their respective teams. Both Duncan and Billups were the second leading scorers on their teams.
- There is a 0.60 correlation (the correlation coefficient is r) between a team’s Wins Produced and the Wins Produced of the team’s top players. If we look at R2 we see that 36% of a team’s Wins Produced can be explained by the productivity of their top player. So the top player isn’t everything, but it’s something.
- Another way of looking at the same issue. Of the fifteen below average teams, only three had a Wins Produced leader that produced more than 14.3 wins (Troy Murphy of the Pacers, Gerald Wallace of the Bobcats, and David Lee of the Knicks). So if you don’t have an above average leader you are not likely to be an above average team. This is an important lesson to learn about building a winner. It’s possible to build a dominant team without one dominant performer, but it’s not as easy.
Kobe and Flash
Kobe Bryant fans will note that Kobe is not the leader in Wins Produced on the Lakers. Pau Gasol was actually a bit more productive. Gasol had his best season of his career in 2008-09, although the actual difference between what Gasol did this past season and what he did in 2006-07 (his last full season in Memphis) is not very big. Gasol posted a 0.240 WP48 [Wins Produced per 48 minutes] two years ago. Had he maintained this production this year his Wins Produced would have been about 15.0 (a mark that’s still good enough to lead the Lakers in 2008-09).
Although Kobe was not the most productive on the Lakers, he was still quite good (he ranked 11th in the league). Quite good, though, was not nearly good enough to match the production of LeBron. In fact – as I noted in my last post – Kobe was not as productive as Dwyane Wade.
Actually, that doesn’t seem possible. “Everyone” knows that rebounds are all that matters to Wins Produced. And Kobe was a better rebounder in 2008-09 than Wade. Therefore, Kobe must have been more productive.
At least, that’s the story I have heard. But when we look at a comparison of all the box score numbers – as reported in Table One – we see that Wade comes out ahead. It should be noted that in addition to an edge on the boards, Kobe is a better free throw shooter than Wade. And Kobe is less likely to commit turnovers. But Wade is a more efficient scorer from the field and does get more assists, blocked shots, and steals. As a consequence, Wade is the more productive player.
Table Two: Comparing Kobe Bryant and Dwyane Wade
One should note that this outcome is not confined to 2008-09. Since Wade entered the league in 2003-04 he has posted a 0.241 WP48 while Kobe has posted a 0.223 mark. And Wade’s career total was reduced by his rookie season [0.128 WP48] and his injury plagued 2007-08 campaign [0.146 WP48]. If we ignore these two seasons, Wade’s career mark is 0.280. In Kobe’s entire career he has never posted a WP48 that reached the 0.280 mark.
In sum, Wade is more productive than Kobe. This is true this season and it’s true if you look at Wade’s entire career. And yet, to the best of my knowledge, Wade has never been named the “best player in the game”. Of course, if we focus on productivity and we look at all players, Wade is not the “best” player in the game. But he is more productive than Kobe (who is often called the “best” player in the game).
– 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.
Mark Wylie
May 10, 2009
For years now I have thought Wade is a better player than Kobe (and far more exciting to watch) so loved this article. Kobe has some of the most passionate fans around proclaiming he is the best in the world, it would be interesting to read their responses to this article
Zach
May 10, 2009
I tried showing my friend this post. He is a big bball fan, but not a big fan of these “new age” stats. When he saw Kidd ranked ahead of Dirk, he refused to even continue reading. Here is what he said he would learn if he started reading this stuff…how would you guys respond to this?
“i would learn how to completely ignore the impact of a guy attracting 2-3 defenders every time, and think a guy is better just cuz hes hitting wide open 3s and gets a lot of different stats. ask 100 coaches, executives, etc. who is better, kidd or dirk, u will not get ONE kidd vote. thats how inane this is. you are not like the moneyball movement and all right, youre like a movement of dumbasses just obsessed with the stats movement that youre blindly ignoring common sense in favor of stats in a situation stats dont tell the story and grossly misrepresent what happened”
dberri
May 10, 2009
Zach,
The basis message of The Wages of Wins — with respect to the evaluation of basketball players — is that the “experts” in the NBA get it wrong. We can see this quite clearly in the free agent data (and we don’t need Wins Produced to tell this story). Given this research, telling us that the “experts” disagree is not very convincing. It is also not surprising. I would add that the “common sense” argument simply comes down to “this is what everyone thinks therefore it must be true.” Again, that is not very convincing either. Hope that helps in your discussion.
Rob O'Malley
May 10, 2009
Isiah Thomas, Kevin McHale, Michael Jordan, Larry Bird, and Elgin Baylor are Hall of Fame players and are considered “Experts” by just about anyone. Now do they know anything about identifying quality players and assembling a team? It doesn’t seem like they have the slightest clue. Even the ones who seem to know the most get a lot of stuff wrong (Joe Dumars).
PJ
May 10, 2009
I think a fair number of people thought Wade was better than Kobe this year. As for their whole careers: Kobe came right out of high school and has stayed healthy, unlike Wade. And he’s older, of course. Add this all up, and he’s played much, much more than Wade has, and at a consistently high level. So even if you grant that Wade, when healthy, is better than Kobe, you’re stuck with the fact that Kobe has had the much more impressive career at this point, I think. And I can’t stand Kobe (and I like Wade).
PJ
May 10, 2009
Oh, and the one name that really surprised me on this list was Mike Miller. Has he always came out well using Wins Produced? If so, I had no idea.
I was a little surprised by Mike Conley, too… might have thought Marc Gasol had edged him out — but Memphis isn’t exactly full of productive (neither is Minnesota, of course).
Ray
May 10, 2009
Ok, I’m waffling back and forth with Wins Produced. Someone convince me.
I’m skeptical about WP because I see that some players are above certain players that I’ve always perceived as better (David Lee over Kobe.) However, when I look at the summation of Wins Produced for every player, they are always within a win or 2 of the team’s actual win total. That’s a pretty convincing number.
With that said, with the position adjustments taken into account and the statisitcal weighting, does David Lee play a position that is more condusive to winning? Do things that PF’s (or C’s) do lead to more wins that a typical SG? If that’s the case, then WP tells us that Lee isn’t necessarily a better player than Kobe, but that Lee efficiently plays a position that leads to more wins than Kobe. Am I correct in this assessment. If so, I feel like I can look through WP through a proper framework, in which I can appreciate the data in front of me, but also I can safely say that Kobe (or Roy or Pierce) is the better individual player.
Marcus
May 10, 2009
The one thing Wins Produced is definitely not about is determining who is the “best player.” At least not in terms of talent or skill at doing difficult things. What it does say is which players most efficiently and consistently do things that actually win basketball games. Whether being a better player means being better at doing difficult things or it means making your team more likely to win games is a personal decision. Wins Produced will not inform you as to the former, but is at least a generally good measure of the latter. (On ball defense, leading to low percentage shots instead of steals or blocks, and awarding rebounds to interior rebounders, is the one obvious facet of the game not captured by any traditional stat, and thus is not included in Wins Produced.)
Ankur
May 10, 2009
Question on Table 1 – when you talk about the team’s WP “if top player became average,” is that average for his position, or average for an NBA player (ie .100)? Seems as though that would make a rather big difference in how useful that number is, particularly given the prevalence of big men on this list…
dberri
May 10, 2009
Ankur,
Average for a position is 0.100. Average across the NBA is 0.100. So average is 0.100 for everyone.
Rob O'Malley
May 10, 2009
The two things that stuck out to me the most are Mike Miller being more productive than Kevin Love. I guess that’s due to Love playing like <30 minutes. And the second thing I noticed was Dirk producing so much less this year. That definitely surprised me.
Ray
May 11, 2009
I was wondering if someone could answer my previous question from my last post and I also had another one. What are the team adjustments made for WP? I was wondering because a team’s total WP is always strikingly close to its actual win total, yet WP can’t factor in team defense. A team can have players with good WPs, which would lead to a higher predicted win total, but happen to also be bad defenders. Is it possible for a team to have a WP of 55, but because they played such poor team defense, only wind up with 45 wins? Or is there a team adjustment for such a problem?
Peter
May 11, 2009
Well, Ray, part of this can be explained in efficiency differential.
Unlike WP, ED considers the efficiency of a team as a whole. Often, however, teams overperform or underperform their ED’s for a reason.
If a team gets fewer wins than its WP would indicate, then it wins games by a larger margin of points than it loses them. They tend to have more close losses and blowout wins.
But if it overperforms, the team has more blowout losses and more close wins. This may be indicative that the team has a strong clutch performer.
Kevin
May 11, 2009
Troy Murphy was on my fantasy team. Everyone in my league kept trying to send garbage my way to get him in a trade. I should send them the WP-MVP per team table.
Horsecow
May 11, 2009
I wouldn’t say “everyone” knows that WP is all about rebounds, but there does seem to be a certain type of player that does well in WP, and maybe, just maybe, “everyone” is not insane for noticing that. To be specific, the type is a player who shoots very few FGA and gets lots of rebounds. Players who fit this type include: Andris Biedrins (8.6 FGA/g), Joel Pryzbilla (3.3), Jason Kidd (7.6), Joakim Noah (4.7).
I think the logic of the critique here goes like this: these players benefit from _not_ taking field goal attempts and instead devoting themselves to rebounding. They don’t bear the risk of shooting difficult shots, which is shifted to their teammates, and consequently they get an increase in WP (by rebounding teammates’ missed shots and, frequently, making layups off of offensive rebounds) while their teammates risk a loss of WP (by shooting difficult shots and not having as many chances at rebounds or easy put-backs). You could make a case that Dirk Nowtizki’s decreased productivity is due to the Mavs’ increased reliance on him to take shots with the departure of Devin Harris (and his replacement by a reluctant shooter like Kidd) and the injury to Josh Howard. Asking a player to take more shots usually means asking him to take more and more difficult shots, with diminishing returns. Nowitzki took 20 FGA/g this year (a career high) as opposed to 17 FGA/g in 2006-07 and 2007-08.
Which is all to say that it may not be appropriate to compare players who take lots of shots (e.g. Brandon Roy, 16.9 FGA/g) with players who take very few shots (e.g. Przybilla). Comparing Pryzbilla and Noah, however, or Roy and Kobe Bryant (20.9 FGA/g), by WP makes sense.
Westy
May 11, 2009
I agree with the above in being surprised at Mike Miller leading the Wolves.
I know he’s always been above average in WP, but would have suspected Jefferson still may have led, despite missing half the year. Or Love by the end of the year.
They might be an interesting team to analyze. How many wins could they have gotten had Jefferson stayed healthy? Once they have a new GM, it might be interesting to do a post on them and their future. With 3 1st round picks this year, I’m hopeful as a Wolves fan that they can maybe luck into one more good player.
Bryan
May 11, 2009
Dr. Berri,
I’m the “friend” Zach referenced. I just want to make sure he’s not misrepresenting your viewpoint to me. He claims that “it’s not about why, it’s about how.” He says that this stat explains clearly that Kidd was “more productive” and “more efficient” in his minutes than Dirk. He says it doesn’t say why this happened, but the stat says Kidd “was more productive.” Is that your viewpoint as well?
I sure hope not. I mean, I’m fully aware experts don’t know much. In the MLB I’m a huge proponent of the stats movement and think all the “experts”, John Kruk, Joe Morgan, etc. are morons–but in the NBA, I think Hubie Brown, Jeff Van Gundy, the Czar, etc. are pretty darn good. If you said to any of them Jason Kidd was better than Dirk, they would fall over laughing.
And they’d be right because it’s just not true. Dirk gets 2-3 guys on him every time he touches the ball. Defenders play Kidd 3 feet off behind the line.
Box score stats in basketball tell almost nothing. Daryl Morey understood this. It looks to me like your stat completely ignores what he realized and puts blind faith in stats that don’t tell anywhere close to the whole story.
I’m listening, so you have a chance to convince me, but if my friend is really accurately representing your viewpoint that Kidd > Dirk, I don’t know how much I can believe if any.
Ray
May 11, 2009
Does an individual player’s WP get affected by his team’s defensive performance? Say Player X has a WP48 of .250 and Player Y has the same rating, .250. Player X’s team, however, is the best defensive team in the league, while Player Y’s team is last. Is Player X’s WP48 inflated because his team plays better defense than Player Y’s?
mrparker
May 11, 2009
The four factors break down as follows according to Dean Oliver:
40% fg efficiency
25% turnovers
20% rebounding
15% free throws
That is for both sides of the floor. Now, a player can only shoot from one side of the floor. So a gunner who does little else can only contribute to 20% of the game. A good rebounder
is usually good on both sides of the floor and therefore is also contributing to 20% of the game. Gunners are usually shorter and shorter teams tend to give up higher shooting percentages(http://ww.basketballprospectus.com/article.php?articleid=82). Therefore I would also argue that good rebounders have a bigger effect on overall efg%.
We can all blame Michael Jordan for our biases because never before had a player that short at that position been considered the greatest. We forget that a shooting guard shouldn’t be able to have that sort of effect on the game. Here’s an exercise to try. Only consider all retired shooting guards. Who would you consider the second or third greatest shooting guard ever? Would you put that guy in the top 10 or 20 players of all time?
mrparker
May 11, 2009
And just for kicks this is my all time top 10. I made it according to who I think played the best against top competition(I.e long histories of performing well in the playoffs). There are only 3 guys on it under 6’9.
1. Michael Jordan
2. Bill Russell
3. Wilt Chamberlain
4. Kareem Abdul-Jabbar
5. Magic Johnson
6. Shaquille O’Neal
7. Tim Duncan
8. Larry Bird
9. Julius Erving
10. Scottie Pippen
mrparker
May 11, 2009
Ray,
That would depend on how much each player had to do with the teams defense being great. If you ever get a chance to read Dean Oliver book, you would have to read about how he calculated a players defensive rating.
mrparker
May 11, 2009
bryan,
I think this argument always boils down what you consider makes a good player. You are only bringing up shooting but there is always other work to be done on a basketball court. While Dirk is busy shooting, Kidd is busy doing “dirty work”. What wins produce does put a value on “dirty work”. This isn’t to say that if Kidd were a better shooter that wouldn’t help Dirk be more productive. Its to say that if Kidd were a better shooter he would be a better player and Dirk might be more productive as well.
Ray
May 11, 2009
Am I supposed to be reading Oliver’s book or get Wages of Wins? Or both? And thanks mrparker. So the WP and WP48 that I look at are influenced by team defense? Specifically, is Delonte West’s, for example, WP higher because the Cavs were great defensively, or would West’s WP stay the same if he played for the Kings (awful defense; assuming he maintained his current production)?
dberri
May 11, 2009
Bryan,
Your argument seems to rely heavily on the idea that the “experts” know who is best. But when we objectively look at the evaluations of these “experts” (such as free agent salaries, voting for awards, minutes played), what we see is that scoring totals dominate the evaluations. Factors like shooting efficiency, rebounds, and turnovers are not considered very important. And we see that in your comments also. You are primarily impressed by Nowitzki’s scoring.
By the way, there is a book that details the argument I am making. Might be worth reading.
Christopher
May 11, 2009
Let me try: Everything is in the boxscore, just look at how “glue guys” are evaluated (the poster boy is S Battier). [And read more blog entries here too, it really is!] I think the kneejerk negativity here is because what WS says does not conform to “conventional wisdom”. But this is what data mining is all about: take a large dataset, look for patterns/trends that are robust and meaningful and see where that takes you. Where this takes WS is that efficiency is king and that REB matter more than many would think. That is really all.
I think where people have this “no way” reaction is when comparisons go across usage rates, like horsecow said above. Also, do not expect perfection –there are uncertainty bars on everything here, maybe C Paul and L James are statistically indistinguishable for all we know. And realize that much of WoW is based on per minute production and is not predictive per se. By the latter comment I mean that you can’t take WS and expect it to track thru trades (look at D Harris, S Marion from last year, or even J Oneal from this year). WS is a diagnostic scoring of each player that permits W-L records over longer time horizons to be partitioned to individual players (which is incredibly useful). It is not forward looking. And it is not situational.
Nick
May 11, 2009
Professor Berri,
I’m not sure if I get this point:
“Another way of looking at the same issue. Of the fifteen below average teams, only three had a Wins Produced leader that produced more than 14.3 wins (Troy Murphy of the Pacers, Gerald Wallace of the Bobcats, and David Lee of the Knicks). So if you don’t have an above average leader you are not likely to be an above average team. This is an important lesson to learn about building a winner. It’s possible to build a dominant team without one dominant performer, but it’s not as easy. ”
This strikes me as the wrong set to look at. If you want to know if you need a dominant performer to have a dominant team, wouldn’t you look at the above average teams to see how many of them have a player with 14.3 WP or more? In this case, only 9 of 15 do, so I would guess that it’s not that hard to be an above average team without a dominant performer.
Am I just misunderstanding your point?
Nick
May 11, 2009
Christopher,
I think one problem WS/WP has is that while it has certainly done well to highlight undervalued or overvalued statistics, which is a tremendous value, to me it has never been very persuasive in establishing direct causation between a player and his statistics.
I mean, it’s one thing to correlate the value of rebounds to wins. But it’s quite another to then attribute 100% of that value to the player who grabs that rebound when you haven’t at all established that he is 100% responsible for its production.
That’s where WS/WP has always been fishy to me. There’s no way to share attribution for individual box score statistics, which to be fair makes it the same as every other box score metric, but nonetheless makes it questionable when applied to individual players since it seems to me facially wrong that every player is 100% responsible for 100% of his box score statistics.
mrparker
May 11, 2009
ray,
Thats a loaded question. D. West had his best offensive rating and defensive rating of his career. The defensive rating is definitely because of the team he played on as his defensive rebounding and blocks were the lowest of his career(steals were the highest but I dont think that makes a 3 point difference from his career). On the other hand his offensive rating was off the charts for him. Looking a little bit closer its easy to see that his offensive rating has been highest when his fga/3pa has been highest. He seems best suited to hang out by the 3 pt line and that is his role in cleveland.
Ray
May 11, 2009
I guess my question is, mrparker, where does that play into WP? WP is strikingly accurate. As you know, you can add up the WP of every player on a team and it’ll come within a win or two of the actual win total. I’m wondering how a model can be so accurate in determining wins if it didn’t factor in a team’s defensive performance (not arguing the model, just trying to understand.) Basically, in a player’s WP48, is his team’s defensive efficiency factored in? Is a player who plays 35 minutes a game going to have a higher WP48 because his team was exceptional defensively?
mrparker
May 11, 2009
Yes, the team’s defensive efficiency is factored in. You’ll have to read Oliver’s book to see how exactly thats figured out. That will answer your questions. Wages of Wins is a great read as well. However, I don’t think it delves into the defensive team adjustment. Maybe the new book will.
Ray
May 11, 2009
Thank you very much, mrparker. Does Oliver use the same system as WoW, or are there slight differences?
brgulker
May 11, 2009
@ Ryan:
Not to be rude, but you’re not getting it.
This model isn’t about who’s “better.” It’s about who’s more productive.
Let’s take the Lakers, not the Mavs, so we can be removed a bit as an example:
Yes, Kobe is “better” in that he has the ability to make amazing plays — make difficult passes, make difficult shots, etc. But, taking those risks on a consistent basis (which he does) lowers his overall productivity (he shoots a lower percentage than he otherwise would, and he commits more turnovers than he otherwise would).
OTOH, look at Gasol’s stat line. He rebounds very well, blocks shots, and shoots a high percentage.
These numbers don’t tell us anything about “talent” or “being a better player” with respect to traditional player evaluation, i.e., scoring. But they do indicate that Gasol’s numbers with respect to efficiency on offense and defense are very, very good.
These numbers also don’t tell us anything about shot creation and/or how the opposing team chooses to defend players. Yes, the Lakers often throw the ball to Kobe with 3-4 seconds on the shot clock, and he is thus forced to take a difficult shot, a lower percentage shot, and yes, those do add up. OTOH, these numbers also don’t reveal anything about how many bad shots Kobe willingly takes when he’s forcing the issue — and anyone who watches him knows he does that quite often.
The same could be true of Dirk. Yes, Howard was out this season. Yes, he draws double teams. Yes, he took more shots than ever. And yes, he’s more “talented” with respect to traditional evaluation.
But is he better than Kidd? That all depends on how you’re defining “better.” And if you’re interested in learning about a new definition, spend some time reading the archives here and purchasing Dr. Berri’s book. If you’re at all open-minded, it will affect the way you think about the NBA in the future. And for a quick intro, I’d recommend searching the site for any posts related to Allen Iverson — you’ll see the power of this model at work to its fullest.
mrparker
May 11, 2009
Oliver uses offensive rating and defensive rating to estimate player contribution. Dberri took that a few steps further(not sure how much he actually studied Oliver vs. how much they just came up with the similar components) and combined both of those elements to get an overall per minute production.
mrparker
May 11, 2009
Sorry fellas I get a little carried away with extra thoughts when the comment threads get lively.
I was trying to figure out how we could have seen cleveland’s jump coming(20+) wins. So, I was looking at how the team is built. Here’s what I came up with.
Offensively, two good assist guys…3 pt shooters all around…big men who are good offensive rebounders….and lebron james.
Defensively, always 2 good defensive bigs and lebron james.
I was thinking about what New Orleans would have to do to take advantage of how good Paul is.
Following Cleveland’s model they have a ways to go.
They’d have to get rid of West(not a good defensive big or off rebounder). They need Chandler to get back to his old production levels. They need a two guard to help Paul distribute. Man, they have a lot of work to do.
Nick
May 11, 2009
@brgulker
Even though he didn’t state it (and perhaps he doesn’t know it), the core of Bryan’s complaint is that because WS/WP is a metric based on box score statistics, it can only be as valid as those statistics are.
Things like double coverage or more generally attribution don’t translate into box score statistics so they don’t translate into WS/WP.
For all of this individual player analysis operates on an assumption that players are wholly responsible for all of their box score statistics because that’s the underlying premise of box score statistics.
Why else would you keep them?
But if that premise isn’t correct (or more likely just overstated), then WS/WP suffers as a model when reduced to individual performance.
It may still be the best model of its kind (I think it is) but that doesn’t cure the causation flaw that seems inherent to me. And this flaw seems to be at the core of many of the complaints that pop up here even if they don’t drill down that far.
mrparker
May 11, 2009
don’t other stats get inflated when players are double teamed more like fta and assists?
Nick
May 11, 2009
I have no idea. Even if it did, that would still be a proxy for what should probably have a separate statistic.
And that’s kind of the point: WS/WP may be the best way to look at box score statistics but it doesn’t make those statistics themselves any better.
mrparker
May 11, 2009
It always been my opinion that once a guy reaches .3 level it means the other team can try whatever it wants but they aren’t going to stop that guy from killing them in every way possible
Michael
May 11, 2009
Sorry but I fail to really understand the point of this post. It is triumphantly announced that D Wade is more productive than Kobe as if that is some kind of convention defying revelation, when the reality is…everyone already knows!
Take a look at the followins stats for Kobe and Wade this year;
Kobe’s P.E.R score 24.3
Wade’s P.E.R score 30.4
Kobe’s Win Share score 13.3
Wade’s Win Share score 14.4
Kobe’s Statistical +/- score 5.89
Wade’s Statistical +/- score 9.51
Kobe’s NBA Efficiency score 24.18
Wade’s NBA Efficiency score 29.29
Wade scores higher on ever single one. I cannot actually find a single metric which disagrees with this consensus!
Then you also come out with “In sum, Wade is more productive than Kobe. This is true this season and it’s true if you look at Wade’s entire career. And yet, to the best of my knowledge, Wade has never been named the ‘best player in the game.’”
Seriously? Never heard Wade called the ‘best player in the game?’
Here are two articles calling him the BEST PLAYER EVER!
http://www.palmbeachpost.com/heat/content/sports/epaper/2006/06/21/z17c_tom_bar_0621.html
http://www.cbssports.com/nba/gamecenter/recap/NBA_20060620_MIA@DAL
Seriously I understand the paradigm is to debunk conventional wisdom and so forth, but this is just getting silly.
ilikeflowers
May 11, 2009
If you’re going to assume that Bryant’s (and other primary scorers’) shooting efficiency is depressed by [1] greater defensive attention and [2] taking bad shots due to lack of time, doesn’t it also make sense to assume that Bryant’s efficiency is also inflated by [1] his team trying to get him good shots and [2] his teammates deferring to him offensively (i.e. giving him the one ball) allowing him to be less negatively effected by the presence of good teammates as are most other players? If you’re going to assume that specialized circumstances are significantly suppressing his numbers why ignore the specialized circumstances that might be significantly inflating his numbers?
This is one of the main oversights of the individual rebounding is overrated vs scoring crowd as well. Shouldn’t the team get credit for setting picks and running plays designed to get their designated scorer easier/favorite shots?
By the way, I’m pretty sure that there is a post around here somewhere where dberri experimented with reducing individual rebounding credit, if memory serves it didn’t make that much of a difference.
Michael
May 11, 2009
Why is my comment held for moderation?
Michael
May 11, 2009
Sorry but I fail to really understand the point of this post. It is triumphantly announced that D Wade is more productive than Kobe as if that is some kind of convention defying revelation, when the reality is…everyone already knows!
Take a look at the followins stats for Kobe and Wade this year;
Kobe’s P.E.R score 24.3
Wade’s P.E.R score 30.4
Kobe’s Win Share score 13.3
Wade’s Win Share score 14.4
Kobe’s Statistical +/- score 5.89
Wade’s Statistical +/- score 9.51
Kobe’s NBA Efficiency score 24.18
Wade’s NBA Efficiency score 29.29
Wade scores higher on ever single one. I cannot actually find a single metric which disagrees with this consensus!
Then you also come out with “In sum, Wade is more productive than Kobe. This is true this season and it’s true if you look at Wade’s entire career. And yet, to the best of my knowledge, Wade has never been named the ‘best player in the game.’”
Seriously? Never heard Wade called the ‘best player in the game?’
Here are two articles calling him the BEST PLAYER EVER!
http://www.palmbeachpost.com/heat/content/sports/epaper/2006/06/21/z17c_tom_bar_0621.html
http://www.cbssports.com/nba/gamecenter/recap/NBA_20060620_MIA@DAL
Seriously I understand the paradigm is to debunk conventional wisdom and so forth, but come on.
Michael
May 11, 2009
Bah I give up.
mrparker
May 11, 2009
Michael,
Maybe the reference is being made to television coverage where Kobe is the best player in the world is a comment that doesn’t get whomever said it laughed off of television. Instead it is taken as fact. I’m tired of the talking head commentators repeating it as well. It makes the chances of having an intelligent basketball conversation much lower than they should be. Instead every conversation degenerates to a dissertation on how Kobe has the best jumper, fadeaway, three point shot, passer, and on the ball defense and its everyone else’s fault but his that LA ever loses a game.
Jason E.
May 11, 2009
brgulker wrote: “Yes, the Lakers often throw the ball to Kobe with 3-4 seconds on the shot clock, and he is thus forced to take a difficult shot, a lower percentage shot, and yes, those do add up.”
They don’t really add up, not much more than they add up for any other player. The notion that Kobe gets stuck taking a disproportional number of late in the clock shots and that surpresses his fg% relative to other players is overstated to the point of being false. 12% of his shots come in the final 4 seconds of a 24 second clock. That figure is real, real, real close to the 11% for the Lakers overall. I suspect the difference isn’t statistically significantly different, but it certainly doesn’t suggest that they hand him the ball in bad situations late in the clock significantly more than his higher shooting volume alone would predict he’d get. It happens to him 1% more of the time than it happens to the team in general. Further, while this year his ‘late in the clock’ fg% dropped relative to shots that happened before (discounting fast break % where everyone does well), the same was not true a year ago, when he was more or less as efficient late in the clock as he was earlier. The differences don’t look like much more than a little stochastic noise around the true mean of his fg efficiency.
It really would be nice if people looked into questions that can be answered empirically before trying to answer them with speculation.
Ray
May 11, 2009
I’m a big fan of Al Jefferson’s game so I was wondering about his WP. Jefferson has been a 13-14 WP guy, which is really good but it isn’t great. I compared Jefferson’s numbers with Tim Duncan’s.
Jefferson: 23.1 ppg 11.0 rpg 1.6 apg 49.7% FG 73.8%FT 1.7 bpg 0.8 spg 1.8 TOs 2.8 PF 36.7 mpg
Duncan: 19.3 ppg 10.7 rpg 3.5 apg 50.4% FG 69.2%FT 1.7 bpg 0.5 spg 1.4 TOs 2.2 PF 33.6 mpg
Jefferson’s numbers are very similar to Duncan’s. Two other factors are in play, though. Minnesota averaged about 3 more possessions per game than San Antonio. Also, I’m never really sure where Jefferson or Duncan are technically playing, so positional differences could dictate the numbers. With that said, given how similar Jefferson and Duncan are on a per minute basis, Duncan finished the year with 16 WP and around .285 WP48. Jefferson finished with a .215 and was on pace to produce 13.4 wins. How did Duncan jump ahead of Jefferson so much in WP48? Is there a factor missing outside of the data I provided, or did Duncan just slowly outplay Jefferson on a per-minute basis? And also, could you build a foundation as a contender with a .220-.230 WP48 type of player like Jefferson being a franchise guy?
Michael
May 11, 2009
Mr Parker I don’t know that seems like a generalisation to me. I think in general most elite players get their due in the media, and there seems to be a degree of hyperbole applied to most of them at some point or another. I suspect if the Lakers sucked and the Blazers ruled the west you’d be constantly hearing about how amazing Brandon Roy is. That just the way it is.
Ray you wrote, “As you know, you can add up the WP of every player on a team and it’ll come within a win or two of the actual win total. I’m wondering how a model can be so accurate in determining wins if it didn’t factor in a team’s defensive performance (not arguing the model, just trying to understand.) Basically, in a player’s WP48, is his team’s defensive efficiency factored in? ”
The short answer is yes. When calculated the ‘long way’ WP is derived from a teams efficiency differential, which is the number of points scored per 100 possessions minus the number of points conceded per 100 possessions (or offensive efficiency minus defensive efficiency.) When you know a teams efficiency differential you can “forecast” the number of wins the team will have with remarkable accuracy. (I put forecast in inverted comma’s because it isn’t really a true projection as you already have the actual real life win data to hand.) Following this the ‘credit’ for these wins is apportioned amongst the individual players. For this reason I question the efficacy of using these win totals as a proof for the validity of the method by which they are apportioned to the individual since the wins are not projected from the individuals onto the team, but derived from the team onto the individual (at least using the long way.) As I have noted before you can use a ‘short way’ of calculating WP from postion adjusted win score which does go from the ‘player up’ but results in far less accurate predictions for wins (around 70% as opposed to 95%+.)
Also on your point about Jefferson I think most championship teams have a wp48 guy in excess of 0.300 leading them, but I could be wrong.
Nick
May 11, 2009
@Jason E.
“It really would be nice if people looked into questions that can be answered empirically before trying to answer them with speculation.”
Well, considering that the Lakers as a whole had 52% of their baskets assisted with 3-4 seconds left in the shot clock, but Kobe only had 16% of his baskets assisted, I would say that the gist of brgulker’s statement is probably true.
The Lakers rely on Kobe to take a disproportionate amount of the difficult shots. Not a single one of the rotation players had less than 50% of their baskets assisted with 3-4 seconds left. Kobe only had 16%, and while he generally always receives fewer assists, the drop when the shot clock is going down is startlingly sharp i.e. the 36% difference is much larger than any other difference for Kobe (25% is the second largest).
So if you had looked at 82games a little longer, I think you would see that while brgulker’s point was a little off-base the gist of it seems correct.
Kobe takes the difficult shots. Now why he takes them is a different question, but it seems very likely that he does.
Ray
May 11, 2009
Michael, you’ve shaken up my faith in WP with your comments. I’ve been under the impression that you take the player’s WP and tallied it up and it always came out remarkably close to the actual win total. I didn’t know that the toal wins was dispersed to the players. Do the team adjustments fudge the numbers quite a bit?
I was starting to use WP as my exclusive source for data on NBA players, citing the “WP adds up to actual wins” belief. Now that I know that that isn’t quite the way it works, what is the best selling point of Wins Produced? Is this the best statisitcal data for evaluating player production and why?
Michael
May 11, 2009
I don’t know about ‘fudging the numbers’ but they do influence the accuracy of the win totals without a doubt. I don’t think you should disregard wp because of that, just be aware that despite the ingenuity of the adjustments wp is really just a linear weights system in the same vein as PER (although PER discounts the use of defensive adjustments as Hollinger considers the approach “incredibly crude”.) To me that means that WP is a worthwhile metric to look at in conjunction with the many others floating around on BBR and 82games etc, but is not markedly superior and therefore its more controversial claims should not just be accepted without thought.
Ray
May 11, 2009
So, a guy like Delonte West, who’s WP48 is like .175 or so, is his WP48 higher because he plays on a very very good Cavalier team? Would he still have a .175 if he played for the Kings? What should be concluded about Delonte (or any player really, I used him for the heck of it) when you look at WP48? How much of the number is relative to the success of your team and the players around you?
dberri
May 11, 2009
Ray,
Delonte West has been an above average player for much of his career. One thing about the NBA to keep in mind. Relative to what we see in other sports, NBA players are consistent over time. This means that NBA performance does not depend on his teammates as much as people believe.
Michael
May 11, 2009
Like I said you can calculate WP from PAWS/min and that will correlate 99% with WP derived from E.D. If you want to see Delonte Wests WP48 seperate from the effect of his teams defence then you could calculate it that way. Go here to learn how to do that https://dberri.wordpress.com/2007/02/25/introducing-pawsmin-–-and-a-defense-of-box-score-statistics/ although the individual score would be basically the same. If he played for the Kings I have no idea what his score would be as that would depend on how he was used in their system, although if he fulfilled a similar role he would most likely have a similar score allowing for variations in things like shooting efficiency which may fluctuate. As for what you should conclude about him (or anyone) based upon his score is up to you. Is he the kind of player likely to be overrated by a stat ignoring shot creation and usage in its evaluation? Or is he likely to be underrated because he is outstanding defensively? These and other questions are the reason that watching the games and using a variety of sources for your assessment is important. That said in terms of this model a player with a .175 wp48 can be said to be significantly better than the average player, so you could conclude he is solid in that regard.
Ray
May 11, 2009
Prof. Berri,
What about the team adjustment then? Say Player X produced 15 wins on a 30 win team, and was traded to a team that won 45 games, for Player Y who produced only 5 wins on the 45 win team. Would you expect that Player X would produce 10 more wins by himself, which would increase the team’s wins to 55 games (assuming everyone else’s production stayed the same)? Or do the team adjustments create too much flexibility in the predictive values of Player X’s contributions to his new team?
Ray
May 11, 2009
“Is he the kind of player likely to be overrated by a stat ignoring shot creation and usage in its evaluation?”
Michael, could you explain a little more what that means? Is there something relatively important that you feel that WP leaves out of its computations? What exactly is shot creation and usage?
Michael
May 11, 2009
Ray I just mean that because the weights applied to individual player stats were regressed from the team level, shot creation is not a factor since at the team level it is considered a given. Usage refers to the use of possessions, another factor with is a given at the team level.
Ruben J. Durham
May 11, 2009
By tieing WP to team efficiency differential, of course its going to match team wins. In effect, WP is rigged to add up to team wins as predicted by efficiency differential. So really what we have is a system that is simply rigged to ‘predict’ team wins in the current year, and is apparently not designed to actually predict future wins according to Christopher’s comment. And this is the future of basketball statistics? Maybe it’s me, but I dont get the hype about WoW.
dberri
May 11, 2009
Ray,
The team defensive adjustment is explained in the book (as well as at the website for the book). Back in 2006 (which much of what is being discussed today was discussed originally) I put up a post explaining how much the adjustment means for the evaluation of the individual player. If we focus on per-minute performance, we see the effect is quite small. There is a 0.99 correlation between PAWSmin and WP48.
Ray
May 11, 2009
Michael, I’m trying not to hassel you, but I’m new to this site, and you’ve been very helpful. I don’t know what shot creation actually means. And do you think scoring is properly weighted in Berri’s system? I’m not trying to be a fan hell-bent on player scoring, but are guys who can’t put the ball in the basket (even if they shoot 55%) overvalued?
Anon
May 11, 2009
As I understood is the value of each stat was derived using efficiency differential, and that once they had the value of each stat they just plug everyone’s numbers in and calculate their WP48. The reason that WP48 totals tend to equal team wins is because on a team level the values given to each of the stats correlates highly to efficiency differential and wins.
Most criticism of WP48 essentially says that what is valuable on a team level does not necessarily translate to an individual level. For example, when a team gets a defensive rebound, that means they got a stop so this is seen as very valuable to the team. But does this mean that the player who got the defensive rebound deserves most of the credit for the stop because he got the rebound? That’s not necessarily the case and it’s one big criticism people have.
The other problem people have is the context of players’ shots, and the fact that WP48 doesn’t take into account all the difficult shots that primary scorers supposedly take a bunch of. It basically just says “oh he missed a shot that’s bad,” while some people argue that “creating shots” is valuable and that therefore WP48 underrates these players
Overall I’d say it clearly has a few flaws, and isn’t perfect, but it’s as good or better than any other model I’ve seen for evaluating players, especially compared to something like PER where Hollinger just basically pulled a bunch of random numbers out of his arse until the ratings players got matched up with how good he thought they were.
Rob O'Malley
May 11, 2009
What does “Creating shots” even mean? Im so tired of hearing that nonsense. If a shot doesn’t go in then it is worthless to your team. If you get an assist then you are credited with one. Other than making a basket or getting an assist, I can’t see how “creating shots” is not measured. If it doesn’t go in it means nothing to your team regardless of how difficult or fancy the shot was.
Nick
May 11, 2009
@Rob
Said like that, it probably is nonsense. But I would assume that “creating shots” means the ability for a player to create for himself or another player a relatively better, though not necessarily good, shot in a situation when the team offense has otherwise stagnated. Alternatively, it could just refer to a player’s ability to convert difficult shots at a higher rate than most players.
The important, uncaptured nuance here is that the idea of “creating shots” seems to rest on the not too absurd premise that not all shots are created equally. In a given game, there will be open threes, break away dunks, step back twenty footers with a hand in your face, and floaters in traffic. All of these shots are of varying difficulty.
Now if the difficulty of a given shot is entirely the product of the shooting player’s poor shot selection, then “creating shots” is nonsense, because the player did not have to take the difficult shot. But if there are systemic causes for these difficult shots (i.e. poor offensive game planning, excellent defense, poor spacing, underskilled teammates, etc.) that are not the sole responsibility of the shooter, then that player’s ability to mitigate those shots’ potential difficulty either by “creating” a less difficult shot or by simply being better at making those difficult shots is an excellent and useful skill that is not picked up by a box score. The idea is that in the absence of that player the team would perform worse in the already bad circumstances.
Now I don’t know if this is true, but if “creating shots” is real that’s what it would be. I tend to think it is real though, because, I mean, bad offensive possessions happen all the time but box scores can only capture the final results of those possessions i.e. a missed shots, turnovers, or unlikely baskets as though those results were somehow distinct or severable from the rest of the possession. To me, that’s facially silly.
Players do sometimes, of course, take bad shots for no reason, but many other times there’s been a systemic failure on the part of the offense and to assign all of the blame for that failure to its final result will always be misleading.
stephanie
May 11, 2009
I think the worst player type that WP evaluates are players who get enough rebounds and score a lot to get a nice WP score but who suck at defense. Like Amare, Big Al, Lee. They practically give up as much as they give. That’s why adjusted +/- is interesting to me. It says yeah, these guys are awesome at offense but they cut themselves off at the knee by letting furniture score on them. Ditto Nash. It’s hard to win a playoff game when the opposing PG is regularly dropping 30-10.
Michael
May 12, 2009
“Overall I’d say it clearly has a few flaws, and isn’t perfect, but it’s as good or better than any other model I’ve seen for evaluating players, especially compared to something like PER where Hollinger just basically pulled a bunch of random numbers out of his arse until the ratings players got matched up with how good he thought they were.”
Sorry but I think this is hogwash. P.E.R is a linear weights metric just like wins produced except without the ‘adjustments.’ Hollinger has been doing this for years (he published his first pbp in like 2002) and I don’t think he ever claimed that P.E.R is perfect, which is why if you read his work he uses a plethora of other approaches when making his cases. You may not agree with some of his weightings but that doesn’t mean the metric was ‘pulled out of his arse’ does it. I could disagree with the efficacy of regressing team stats to the individual level, but that doesn’t mean I should be insulting the prof for coming up with the idea does it. At the end of the day WP still deserves consideration despite its flaws and the same goes for P.E.R.
Ray shot creation just means the ability to create shots. This aspect of the game can be measured by the percentage of a players shots which are assisted, or the percentage of a teams shots/possessions used by a player. This relates to wp because by regressing from a team level, usage is treated as a given since every team uses all of its possessions and therefore all that matters is efficiency. However in reality this is not the case since as Nick pointed out all shots are not created equal, and I would suggest you simply need to watch the games to see the benefit of an offensive player who can draw the defences attention, splitting double teams, creating open looks etc.
Stephanie lol @ ‘letting furniture score on them’
Michael
May 12, 2009
BTW whilst on the subject of PER, has anyone else seen Lebron has a score in the playoffs so far of 44.7! Unbelievable! His win share score is over 3 already! I don’t have my spreadsheets with me but a quick calculation indicates his projected wp48 is 0.670!! That is just incredible, can anyone check I have this correct?
Michael
May 12, 2009
Okay I have to say ‘I don’t have my spreadsheets with me’ is officially the geekiest thing I have ever written :-p
Nick Gelso
May 12, 2009
Very interesting. This new way of rating players based on stats is not something I consider myself a professional at.
I think LeBron was the correct choice for MVP this year. Chaucey was there too w/ Kobe and D-Wade.
mrparker
May 12, 2009
Michael,
My beef with the shot creation argument is that a player who is truly uber-talented shouldn’t have to resort to taking difficult shots. That player should be able to get whatever shot he wants and when he wants it. Or help his team get better shots before the clock is expiring.
As far as Kobe is concerned, the Lakers have a plethora of above average offensive players. There is no excuse for Kobe to be taking impossible shots with under 3 seconds. If my coworkers were just throwing there unfinished projects on my desk the day before the deadline I might get a little sick of it after a few times. I might start helping them a week or two before the deadline to make sure the work gets done as best as possible.
Nick
May 12, 2009
@mrparker
“My beef with the shot creation argument is that a player who is truly uber-talented shouldn’t have to resort to taking difficult shots. That player should be able to get whatever shot he wants and when he wants it. Or help his team get better shots before the clock is expiring.”
I would say this is a comic book version of the best players in the NBA. The best players in the NBA are doing fantastically if they make half their shots. And most of them don’t make half. Difficult shots are simply part of the game, as shooting is generally not easy.
Moreover, in a case where an “uber-talented” player was playing one on one, I might agree with you. But when there are in nine other people on the floor in addition to the uber-talented player (and not all of his teammates will be uber-talented and some of the defenders might be) even Uberman will have to settle for low percentage fadeaways every now and then. It may be that he chose a bad shot or it may be that he’s making the best (or at least better) of a bad possession.
Because sometimes teammates don’t set effective screens; sometimes they miss the open man; sometimes the defender is just really good. There’s a thousand reasons why good players don’t always get ideal shots and many (if not most) have nothing to do with the quality of the player taking those shots.
If it were like you said, where bad shots were simply always optional, then we’d just have Lebron James dunking the ball at will on one end, while Chris Paul does finger rolls on the other for a final score of 342-340. It seems to me the game is more varied than that.
mrparker
May 12, 2009
Nick,
Thats the literal interpretation of what I said. However thinking in terms of what a good shot is, its a shot that a player can reasonably expect to make 50% of the time.
Michael
May 12, 2009
Hi Mr Parker, I do understand your point from a theoretical standpoint but in a practical sense I think shot creation (usage) is important. Using your Kobe Bryant example here is a link to an old article (2005) by Kevin Pelton which lends credence to that idea (that high usage offensive players can help your team significantly.) http://www.82games.com/pelton13.htm
mrparker
May 12, 2009
Dwyane Wade
.516 3.5 12.2 7.8 40.3 3.0 2.8 11.6 36.2 115
Kobe Bryant
.502 3.5 12.8 8.2 23.8 2.1 1.0 9.7 32.2 115
There is only one real difference in these numbers. Its the ast% and the usage. I suspect that wade getting his teammates involved is what makes all these metrics value him higher
Nick
May 12, 2009
I would say a good shot is contextual. If you get a shot you can only make 50% of the time after a 3 on 1 break that’s not very good.
If you’re Eric Gordon facing up Ron Artest with 4 seconds to go on the shot clock, getting a shot you can make 40% of the time is probably pretty good. If you change that to 1.6 seconds left in the game, it’s probably pretty remarkable.
The spectrum of shots available at any given moment in the game varies, meaning that shots you tend to make 50% of the time can either be good, bad, average, or implausible at any given moment.
So “creating shots” (i.e. creating good shots), if it means anything, just means a player’s ability to improve that available spectrum by either adding higher percentage shots to it or by executing those shots more efficiently.
Under no interpretation, literal or otherwise, would this be the same as the player being “able to get whatever shot he wants and when he wants it.”
A better version would say that the uber-talented player can get a better shot than an average player could when he wants it.
Caleb
May 12, 2009
“However, when I look at the summation of Wins Produced for every player, they are always within a win or 2 of the team’s actual win total. ”
I know very little about statistical analysis so I’m not qualified to say one way or another… BUT, I’ve read and heard from many other folks who do know their stats that this means nothing.
One could conceivably invent a metric that came to the exact opposite conclusions that WP does and still have the numbers match the team wins. How does this work? I have no clue… but this is what I’ve heard from others in the field.
dberri
May 12, 2009
Caleb,
The statistical analysis you heard about was riddled with problems. So what you heard is not true. It is also not true that PERs with a team adjustment explains wins as well as Wins Produced.
By the way, all of this will be briefly discussed in the next book. In fact, much of what is being discussed in this comment thread will be touched upon.
ilikeflowers
May 12, 2009
mrparker,
Wade’s a more efficient scorer as well as a better passer. Using nba.com’s stats I have Kobe with 1.07 points per shot and Wade with 1.18.
ilikeflowers
May 12, 2009
Here’s your Kobe hyperbole of the day, ‘…which is a really weird trait for a guy[Kobe] who — along with LeBron James — is a once-a-decade basketball talent.’
– Tim Keown.
mrparker
May 12, 2009
I like flowers,
Somehow they both come out with a 115 offensive rating. Until you get to steals, assists, and blocks these two have very similar component stats. I just noticed that, I thought it was interesting. Also I have Kobe’s true shooting percentage at .56 and Wade at .57. Wade is better but they are still pretty close.
ilikeflowers
May 12, 2009
Wow, I didn’t even look at assists, blocks and steals. Wade’s killing Kobe there. Kobe does have a 10% edge in net possessions. But Wade’s stats are clearly superior overall. It even looks like one could reasonably conclude that Wade is a more effective defender (within his team role) since he has significantly more steals and blocks with fewer fouls.
/ minutes played are very similar (within 3%)
jbrett
May 12, 2009
Professor Berri,
It seems to me your blog could benefit from posting, at the beginning of each Comments section, a list of time-saving conventions for the new or unindustrious poster. I only found it a few months ago; I spent a long time reading the older articles, and eventually I bought the book. This seemed the sensible approach, though, judging from the tone of many of the comments left, not the favored one. For the benefit of the many posters who consider this site homework-optional, I submit the following list of generic positions that NEED NOT EVER BE ELABORATED UPON EVEN ONE MORE TIME:
A. I have little or no training in statistics (me, for one)
B. Obviously, any metric that says Player A (let’s say, oh, Jermaine O’Neal) is not as good as Player B (how about, um, David Lee) is clearly flawed
C. Anyone who’s ever watched a game can see that Superstar A (Allen Iverson, anyone?) is ten times the player that Serviceable Role Player B (Chauncey Billups, maybe–or how about Andre Miller?) will ever be
D. Superstar A and his ilk cannot be quantified in the same way as mortal players can; they only shoot 42 percent from the field and 28 percent from 3-point range because
jbrett
May 12, 2009
Oops, ran out of time.
D (Continued) their teammates DEMAND they do so, by leaving them with the tough shots at the end of the 24-second clock
(See how much space that one will save, when all you have to type is ‘D’?)
E. My friend/ brother-in-law’s boss/ opinionated alter-ego hasn’t read or studied your work, but I told him the results say Mike Miller is way better the Richard Jefferson OR Rip Hamilton, and he says you’re clearly deluded
I’ll stop there–but, obviously, as other arguments become hackneyed, they can be assigned the next letter. Think how much easier it will be to find the genuinely interesting discussion when the endless repetitive jabber is distilled to a handful of letters one can note and skip past. It seems like an idea whose time has come. Any thoughts?
mrparker
May 12, 2009
football outsiders has a fill in the blank comment form for when fox posters come over…thats always pretty funny
Oren
May 12, 2009
“Well, considering that the Lakers as a whole had 52% of their baskets assisted with 3-4 seconds left in the shot clock, but Kobe only had 16% of his baskets assisted, I would say that the gist of brgulker’s statement is probably true.”
How about we take out every single shot that Kobe shot with less then four seconds on the shot clock and see what impact that has on his effective Field Goal Percentage?
By my calculations, doing this would mean that Kobe shot .516(EFG%) as opposed to .502(EFG%). Given that he shoots 21 FGAs per game this would mean that he would score about .5 more points per game. If he plays 36 minutes per game, then this would mean his WP48 would increase by about .014 pts. Presuming he played 3000 minutes that would mean that his Wins Produced would increase by .875.
Of course, if we do this for him, we’d have to do this for everyone else. And I suspect this would make the “average” player better which would lessen the bonus. So, figure that if Kobe didn’t shoot the ball with less then 4 seconds, he would improve his Wins Produced by half a game? Three fourths of a game at most, I would think.
It may be slightly more then noise, but I’m skeptical it has a huge impact.
Michael
May 12, 2009
Oren, putting aside discussions of the number of seconds left when Kobe shoots (although I would point out that your expected improvement in the ‘average’ player would not be equal to the improvement you would see in a player like Kobe who takes a larger percentage of those shots relative to possessions used) I actually think the main issue in this case would be, if given the number of possessions which Kobe uses, could you expect his team mates to match his level of efficiency? In other words, absent Kobe from the Lakers team, would the other players on the be better, or worse offensively?
Ruben J. Durham
May 12, 2009
Gee, jbrett, what a great idea! Then we could all exist in a fairytale world where everyone agrees and noone has the indecency to come along and poke holes in Dr. Berri’s very flawed system! Although I believe every good despot in history has taken steps to silence his critics. Is that what youre wanting to see here?
Michael
May 12, 2009
Robert I can to an extent understand your frustration but don’t you think that inflammatory comments are a little counter productive? I think that the professor tolerates a good deal of dissent in these comments sections and that leads to some informative and engaging (at least to me) discussions. Besides the majority of jbretts comment categories are strawman to the extent that they need little attention, except to note that for someone who is totally convinced by what they find here, any contrary opinion would no doubt be met with the same frustration as you feel when being met with the close mindedness which sometimes follows such expressions.
Okay anyway…to follow on with this point on shot creation/usage, here are some quick numbers from 82 games to support this argument as it pertains to Kobe and the Lakers this year.
With Kobe the team scores 117.2 points per 100 possessions, and concedes 105.8 points per 100 possessions.
Without him the team scores 103.9 p/100poss and concedes 104.5 p/100poss. That means they have a net gain of 1.4 on defense, but a loss of 13.3 on offense without him.
The teams net points (points allowed minus points scored) are at +658 with Kobe on the court, and -30 without him. Also the effective fg% rises from 48.7%, to 52.1% (a gain of 3.4%efg) when he is on the court.
So you can see that a player like this does seem to have myriad benefits for his team, regardless of whether people consider shot creation to be real consideration, or that usage can indeed affect efficiency. If anyone would like to look at a much more in depth analysis of this, check the link I posted above to the Pelton article, although the article is referring to 2005 it seems the net result is similar to (if I gather less pronounced than) this year.
Michael
May 12, 2009
*Ruben. Sorry it’s quite late here.
Phil
May 12, 2009
Oren, wonderful analysis.
Mark
May 12, 2009
Rockets GM rips on win score
“Keri: On the statistical front, are you, or people in the organization, searching for a catch-all measure of player value, like VORP in baseball? Is John Hollinger’s PER that stat? Or is basketball too context-dependent for a catch-all measure of value?
Morey: There’s certainly value in a catch-all stat, especially from a public or media perspective. I look at it this way: I don’t think people who minimize it are right, or the people who trumpet it. Being able to rank players that way is a way to simplify for mass audiences. But in terms of decision-making, it really is almost useless. Everything becomes more contextual. How will this player fit in our system, with the plays we run, with our defensive scheme? Those are the questions that rule the day, not whether a player is No. 7 or No. 6. ”
http://sports.espn.go.com/nba/playoffs/2009/columns/story?columnist=keri_jonah&page=Morey-090512
jbrett
May 12, 2009
Ruben,
I don’t think anyone here is looking to silence anyone else; why come here at all, for such a vain pursuit? All I would like to do is foreshorten the process. I will venture that the five positions I described comprise nearly half of the comments posted on this site. It’s enough to make one think that nearly every poster is reading his first article. I’m not saying anyone should be silent; I’m suggesting that, when hundreds of people have already made those particular points, perhaps one could be brief, and allow some new points to be made. I don’t sign on to wax rhapsodic about the virtues of Kobe in the clutch; it’s been done, ad nauseum. I’ll just type ‘D’ and we can try something fresh. Is it really too much to ask?
Ruben J. Durham
May 12, 2009
Those points continue to get harped on because they really never get resolved. Sometimes theres no resolution, with Kobe fanboys, and thats justified. But legitimate concerns with the structural flaws endemic to wins produced also see the same treatment and go unresolved. That is what I have a problem with. Complaints about the results, ie, what order players are ranked in, are not important. But complaints abiut the process and the convolutions of the statistical path from raw #s to wins produced need to be thoroughly resolved, and they havent been yet.
Nick
May 12, 2009
@Oren
“So, figure that if Kobe didn’t shoot the ball with less then 4 seconds, he would improve his Wins Produced by half a game? Three fourths of a game at most, I would think.”
I wasn’t making the point that Kobe’s taking difficult shots more frequently would be statistically significant in terms of WS/WP. I was responding to Jason E’s assertion that Kobe did not take a disproportionate amount of difficult shots when the shot clock was running down.
So I simply stated that I thought that as a matter of empirical fact, it seems at least likely true that when the shot clock is low, Kobe takes a disproportionate amount of difficult shots, because his shots were disproportionately (even for him) unassisted.
That was my point and my only point, because something can be empirically true without being statistically significant.
Nick
May 12, 2009
Also, I don’t know (and I mean I honestly don’t know) that WS/WP is even the correct metric to determine the affect taking and making difficult shots has on a player’s value.
I mean, WS/WP is based on box score statistics which do not distinguish between difficult and easy shots, right? So that means WS/WP implicity assumes that the ability to make difficult shots is no more or less valuable than the ability to make easy shots.
But since it is without question empirically true that there are such things as difficult shots that are distinguishable from easy shots (though the precise line between the two may not be clear), the only way you’d know the value of making each is if you had a raw tally of each to correlate separately to wins, right?
So it couldn’t be the case that if you started back at square one with a more nuanced box score that more accurately corresponded with the empirical reality of basketball all or at least some of these WS/WP outcomes might look significantly different?
It just seems like we’re starting with a box of crayons (the box score) and saying it’s a sufficient proxy for the spectrum of visible light (basketball). And I’d guess that like half the problems people have with WoW are actually not with the metric itself but with the inherent flaws in the box score that they never had to notice before.
jbrett
May 12, 2009
Michael,
How is it possible that the Lakers get better defensively when Kobe is out of the game? We all know he’s one of the greatest defenders of all time–at a minimum, the best on the court today. We KNOW this; we’ve seen it with our own eyes. If the numbers say otherwise, the numbers have to be lying, don’t they? You must have arrived at them erroneously, or else you’ve manipulated them in order to make a point not supported by all the visual and anecdotal evidence to the contrary. I don’t have the skill to find the flaws in your conclusions, but I know you’re wrong because, well…I just know, OK?
(I can’t keep that up. Sorry, Michael, for using your numbers as a foil for a different venue.)
I’m a Laker fan. I love Kobe. The better he is, the happier I am–but it pains me to think he has succeeded AI as the poster boy for all that ails Win Score and WP48. I don’t think anyone is arguing that these metrics are a standalone measure of who is better; everything has a context. Marcus Camby, for example, gets a lot of grief about where WP48 ranks him–more accurately, WP48 is often criticized because of where he ranks–and I can’t say I’ve ever paid much attention to his game. However, I accept that the numbers measure something of value, and if what I need is a 4-5 type, I accept that he has great value, at this moment–not 5 years from now, not for 42 minutes a night or 78 games a year, and not to run the point. Would I take him over Kobe? As with every other question of that type, it depends. If I add Camby, could I trade Gas0l or Odom for Brandon Roy, and give Kobe minutes at SF? Would he be better there than my other options? Can Camby play enough minutes to make the move worthwhile? There are things Professor Berri’s work cannot tell me–but what it CAN tell me has value. I will add that, if you disagree, you will have difficulty demonstrating why. If you DON’T accept that something valuable is being measured, why would you choose this, of all places, for your soapbox?
I have been accused, directly and indirectly, as accepting WOW as gospel; I must admit that, when understood in the proper context, that isn’t far off. For the devout (Christian, anyway), the Bible may be viewed as the controlling moral and philosophical authority; however, if one wants to build a swimming pool, the story of Jesus walking on water is far less valuable than a step-by-step instruction manual for building a swimming pool. WOW is not the end of the discussion, nor is it a complete how-to guide for assembling an NBA dynasty–but “I can tell just by looking that WOW is woefully lacking” is not even a discussion. (Talk about faith versus empiricism.) THOSE are the comments I could do without; time is precious–and I’ve given this particular battle enough of mine. Have fun.
Ruben J. Durham
May 12, 2009
I was thinking and I believe I finally came to a realization about wins produced and the problem with applying team logic to individuals. Bare with me while I talk thru it – OK, WP is not saying how much a player is inherently worth in wins. Its not talking about talent, its not talking about value in some sense of any environment. Its describing the worth of the player’s numbers versus his speicifc team’s statline minus his numbers. Dr. Berri has worked out the relationship between team statsb and winning percentage. Technically one measure of value would be to look at the predicted wins without those numbers, the difference being WP. It doesn’t know or care about the difference in raw basketball ability between a high-talent player like Kobe who is able to adapt and produce in a variety of teams, situations and roles, and a low-talent hardworking big man who rebounds well and only shoots high % layups. It cares about the difference in numbers and how they were worth to the individual teams. Thats why its not predictive, someone even said it wasnt supposed to be. It looks at numbers and sees how many wins those numbers – but not the player-interaction elements of talent and adaptability – would add to the team he played for. Is this a better understanding on my part? I dont agree with this approach, but now I at least think I understand what WP is aiming at.
jbrett
May 12, 2009
Well, I thought I could let it go, but I didn’t even make it off the page. Let’s add:
F. I haven’t read THE WAGES OF WINS, nor am I likely to, and as a result I will begin by gainsaying basic tenets of the book
(Perhaps this should have been ‘A’ and the rest moved down. Oh, well.) How else could anyone comment that the boxscore is inadequate to the purpose at hand? Nick, you might as well go before the Supreme Court and argue there’s no real basis for a written code of laws. In some universe, somewhere, you might be right–but you’re in the wrong place to even make the argument. If you HAVE read the book, and still want to talk about the metrics AND
G.
jbrett
May 12, 2009
Oops again. If you want to discuss the metrics while simultaneously saying the boxscore is insufficient, let’s throw in:
G. I read your book, and I say “Nunh-unh.”
I’m going now.
Nick
May 12, 2009
Is it really a radical statement to say that the box score inadequately captures basketball? Really?
Fair enough.
I read the book and said sound but invalid.
jbrett
May 12, 2009
Nick,
I wanted to find the right quote, rather than paraphrase Dr. Berri, but I’ll do my best: The boxscore is a highly useful resource for measuring player productivity, provided that each statistic is weighted properly for its correlation to wins; to matter, points, rebounds, etc. must be defined in terms of how they aid in winning games.
To say the box score fails to capture the essence of basketball is irrefutable, esoteric, and wholly irrelevant to the discussion. (I confess, I don’t have a clue what “sound but invalid” means.) There may be a forum for debating Win Score or WP48 by positing that the raw data from which they are derived is inadequate, but I continue to say that the value of doing so on the website of the metric’s creator is nil. WP48 is not intended to be an oil painting, Kirlian photograph, or psychic reading of the essence of the game; it is a measure of the production of discrete, quantifiable acts which lead to the winning of games. If you wish to challenge the methodology applied, that’s one thing–but I rarely see that kind of post. What I overwhelmingly see here boils down to the idea that we can’t measure everything, so the unmeasurables I perceive invalidate all the things we CAN measure, or the importance we assign to them. If you wish to tell the builder he used the wrong tools, or used them incorrectly, or left out some parts, that can be debated. What, however, is gained by telling the builder he should not have built to begin with? or that what he built is useless because he did not allow for factors he could not assign concrete value? The usefulness of the product may be restricted, but it does not cease to exist; the builder has used all of the materials and skill to which he has access, and may be justifiably annoyed at the conceit that the absence of perfection renders the construction without purpose.
The only reason I ever suggested we shorthand some of the repetitive arguments is this: Dr. Berri seems quite longsuffering on the topic, but the substance of his posted responses is largely to refer to the book, the website, or posts the initiator of the dialogue could have read with a minimal effort. Wouldn’t it be nice if that resource–direct correspondence with the author–could consist of more than, “I’ve already talked about that”?
RsquaredCollins
May 12, 2009
The book doesn’t address the fallacy of regressing wins on team statistics and assuming that the relationship at the team level holds at the individual level, and neither do the author’s responses. It actually doesn’t even seem to occur to him. So are we allowed to broach that subject?
jbrett
May 12, 2009
I could not and would not stop you, as I’ve said before. Your point above does not fall within the seven worn-out positions I have mentioned; it sounds like an interesting discussion, and one I would follow intently, even if (or perhaps because) I lack the knowledge or skill to participate. Further, even if over the coming weeks or months that debate should get beaten to death, I still would not expect the subject to be verboten; I would say we assign it the next letter–currently ‘H’–and everyone could then express the position with the utmost brevity, while moving on to something fresh. I swear, I’m really not that hard to satisfy. :)
Nick
May 12, 2009
“the builder has used all of the materials and skill to which he has access, and may be justifiably annoyed at the conceit that the absence of perfection renders the construction without purpose.”
I didn’t ever state that WS/WP was without purpose. In fact, I said it was the best of its kind of metric. I did point out the flaws in the box score for two other reasons:
1) To note that most of the problems people have with WS/WP actually have nothing to do with WS/WP. Instead, they have to do with the inherent flaws in the box score. People just never noticed these flaws until WoW essentially magnified them (not worsened just made them more obvious) by essentially reducing wins to their production. But this isn’t a flaw in the WS/WP model; it’s a flaw in the data set upon which that model is based and its been there since the NBA began. WS/WP has if anything improved the use of this data, but its still flawed data.
2) I do criticize people who use WS/WP to refute the value of things that WS/WP does not measure. A commenter had essentially tried to dismiss the effect on a player’s value that taking (and making) a disproportionate amount of difficult, late shot clock shots may have by pointing out that if the player had not taken those shots, his WS/WP would hardly be affected because the player’s eFG% would hardly change.
This is a bogus argument because it assumes (as WS/WP does because box score statistics do) that the only difference between easy shots and difficult shots is that difficult shots are made at a lower percentage.
And because it makes this assumption it does not address the at least possible outcome that making difficult shots may be a more valuable basketball act than making easy shots (i.e. one would think that ability rare thus valuable). Alternatively, that may not be the case. Importantly, however, WS/WP cannot answer this question because it never (because box scores do not) correlated difficult shot-making to wins separately from easy shot-making.
And I’m not saying this makes WS/WP “without purpose” for not making these measurements; that would be a ridiculously difficult undertaking. But I am saying if you
a) Know that two distinct actions exist
b) But never tally their occurrences separately and instead just classify them as the same
c) and then only correlate that single classification to wins
Then you can’t posit or refute the fact that one or both of the conflated actions may have a different weight than the combined if it was correlated to wins separately. Your model doesn’t cover that issue.
So I think its worthy of criticism when someone states that difficult shot-making does not affect a players value anymore than normal shot-making by pointing out that WS/WP, a model that simply assumes the acts have equivalent value, does not reflect a difference in effect.
Once again though, I’m not saying WS/WP is without purpose. I’m just saying it has limitations and this is one of them.
Nick
May 12, 2009
@rsquared
You can broach it, but you won’t get much of a response. My guess is since box scores essentially do the same thing (assume that its valid to assign statistics players in the same way you assign them to teams), WS/WP is just, intentionally or not, leveraging that overlooked flaw to come up with unconventional wisdom.
But I’ve never gotten a response on that issue or any other of that type.
Maybe if you phrase it as:
“The book doesn’t address the fallacy of regressing wins on team statistics and assuming that the relationship at the team level holds at the individual level, and neither do the author’s responses. It actually doesn’t even seem to occur to him. Therefore, Allen Iverson is better than Andre Miller because he is clutch.”
That might trick folks into addressing it.
jbrett
May 12, 2009
Nick,
Your points are well-taken. As a Laker fan, I have a more-than-academic interest in quantifying ‘clutch,’ even if I have my doubts about standardizing the definition.
I would make the simple argument that better shooters make a higher percentage of their shots. If that’s too simple, I’ll say that better PLAYERS find a way to take more high-percentage shots. Perhaps, if Kobe stayed within the offense a little longer, he could trade shooting an admittedly-impressive 40 percent on 180-degree-pivot, double-clutch with-a-hand-in-the-face 22-footers for 55 percent on wide-open 18-footers and 2 more assists per game. To borrow your phrase, that may not be the case, and we may never know. My explanation, admittedly difficult to quantify, is that the quality you’re seeking–let’s call it ‘shot-making’ ability, the talent for putting in tough shots under duress–is a PHYSICAL skill; it is distinct from the BASKETBALL skill of getting, or creating (for yourself or a teammate) a higher-percentage shot. I find myself wondering if it isn’t the Iverson dichotomy removed one step, to form a general principle about ‘taking the tough shots.’ AI has done that everywhere he’s played–and been replaced by someone who lacked that same reputation or bravado, who only led the team to more wins. I’m suggesting that getting BETTER shots is a possibly more measurable, and definitely more valuable, skill than making more tough shots than might be expected–and I say that in spite of relishing every single one of Kobe’s circus acts. In the back of my head I’m thinking Magic would have either made an easier shot, or made the pass for a layup. I look forward to the day we have enough data to do more than guess; until then, I’m inclined to believe that the beat-the-shot-clock heroics are more about highlight reels than basketball IQ.
THIS dialogue is anything but worn out; now we’re having some fun!
Nick
May 12, 2009
JBrett,
“I’m suggesting that getting BETTER shots is a possibly more measurable, and definitely more valuable, skill than making more tough shots than might be expected–and I say that in spite of relishing every single one of Kobe’s circus acts.”
Well, sure. I guess the presumption about difficult shot-making was that in some scenarios an offense can do no better than a difficult shot (because of poor team execution or superior defense) and so in those scenarios people who can make those shots at a higher (though not necessarily high) rate may have a distinguishable value from normal shot-making.
But if its like you say and Kobe’s (or Iverson’s) just opting for bad shots, then that’s different.
What I was really talking about was that if it is inevitable that team offenses will stagnate for reasons not assignable to any one player (superior defense or poor team execution) then any one player’s ability to mitigate that stagnation by excelling at the difficult shots it leads to or by creating better than normal shots in adverse circumstances could very well have a value distinct from the value of easy shot-making i.e. teams that can execute difficult shots or get better shots in difficult circumstances win more than teams that don’t.
The thing is you would have to do a massive amount of tape-watching to even start to accumulate the data for this. But let’s say that with some super-hypothetical computer you could tally all the relevant data up (difficult shots made, difficult shots opted for, difficult shots required, improved shots relative to circumstances) then you reset the WS/WP regression to include these new variables.
You may find that each of these new types of acts has a unique value that would affect WS/WP across the board. Alternatively, it might not make a difference. Thing is, you can’t know unless you can do it but you probably can’t do it.
Note also that none of this is meant to imply that guys like Iverson are underrated because they take difficult shots; rather it says all players ratings are incomplete because the data isn’t accurate. With more accurate data, you may find that Kobe and Iverson are worse because they tend to create unnecessary difficult shots.
But yeah, there’s no regression to solve these real problems. There’s just video tape and patience.
jbrett
May 13, 2009
Nick,
I can’t disprove anything you’ve said there, obviously. However, we do have some anecdotal evidence which could be viewed to weaken the case for the someone-has-to-take-the-tough-shots faction, and it comes back to Iverson again. (Sorry.) Maybe there are ‘conventional’ pundits who would rank Billups above AI–he’s got a ring, after all–but Billups reputation is for taking BIG shots, not tough shots. Without AI filling that make-the-best-of-it role, shouldn’t Denver’s offense have fallen off, when someone else had to take over?
Two possibilities. One, Billups took those end-of-clock shots, but the degree of difficulty was not perceived to be Iverson-level acrobatics. Two, no one player filled that role; instead, the team continued to run the offense until a decent shot presented itself.
If you consider Billups on par with AI, let’s switch the comparison to Andre Miller. Same fate for the teams involved, so Philly doesn’t seem to have missed their go-to guy very much. Does this prove anything? No. But it does cast some doubt on the value of a guy who’s willing to take any shot, if he perceives it necessary, or if he can hang on to the ball until the clock runs down so low he cannot be faulted for whatever shot he ends up getting, no matter how tough.
Good talking with you; off to bed for now.
ilikeflowers
May 13, 2009
RsquaredCollins, Nick
[1] There is no ‘fallacy of regressing wins on team statistics’. It’s a perfectly logical position to take. It may or may not be correct but to call it a fallacy is itself a fallacy.
[2] ‘…assuming that the relationship at the team level holds at the individual level’. This is certainly an assumption of the model. One way way to reasonably test the assumption’s validity is by seeing how well a teams’ wins are predicted when individual players leave/join teams or play more minutes. Ultimately, this is the key point. If you can accurately predict team wins under these circumstances as well as or better than other eyeball/statistical models then you have an assumption that is likely a reasonable approximation to reality.
To simplify,
[1]G
[2]F
dberri
May 13, 2009
One thing to remember is that basketball players –relative to what we see in other sports — are quite consistent over time. This suggests that the numbers assigned to the player in the box score are mostly about the player.
By the way, this is something I have said before.
So I guess we probably do need a list of letter codes.
One last note… in the next book this issue will be touched on again with even more evidence.
Nick
May 13, 2009
@ilikeflowers
Just generally:
The fallacy Rsquard stated was not “regressing wins on team statistics. ” Instead it was “the fallacy of regressing wins on team statistics and assuming that the relationship at the team level holds at the individual level.”
I’m just sayin. Disagree as you will, but disagree with what he actually said.
@dberri
You don’t need a list of letter codes, since people would have to learn the codes and that more or less moots the point. You need a FAQ.
Nick
May 13, 2009
Also, statistics staying consistent doesn’t really speak much to causation by itself. I mean, that could just as easily suggest that what’s actually consistent is a player’s ability to remain equivalently dependent on an offense or teammates of a particular level of competence i.e. guys get stuck in similar roles where ever they go.
The only way you’d sort that out is to work back from instances in which players did not remain “quite consistent” and try to empirically determine what caused the change. That would be a lot more illuminating on this causation issue.
Yee haw, Book 2. We want answers.
ilikeflowers
May 13, 2009
Nick, that sentence can be reasonably interpreted either way. My points are valid under either interpretation, only the quote from point [1] would require modification. I’m just sayin :0).
jbrett
May 13, 2009
FAQ would work. Another idea would be to make the links to the 38 months of archives really, really large, in brighter colors, possibly flashing.
R^2Collins
May 13, 2009
Flowers, the fallacy is in applying team regression coefficients to individuals and assuming the relationship continues to hold true. That’s the core of Dr. Berri’s method, and he has yet to prove that it’s an appropriate assumption to make. According to Erich Doerr’s spreadsheet, WP doesn’t predict future team wins as well as something like P.E.R. or win shares. Someone even said yesterday that WP wasn’t even meant to be predictive. So if that’s your “key point”, I’m afraid WP has failed to impress so far.
R^2Collins
May 13, 2009
What’s the “code” for that line of criticism, by the way?
jbrett
May 13, 2009
RsquaredCollins,
I’ll make a suggestion for an experiment one might try; it’s beyond my skills, time available, or commitment to the premise, but I’m thinking you might be the one to pull it off.
Take someone else’s formula–NBA Efficiency, PER–if you’re industrious enough, try every formula you’ve ever seen. Using the individual player scores from each for last season, predict the number of wins for each team; I don’t know how you get from one to the other, but it sounds like you might. Then look at the correlations, and compare them to WS/WP. I think you will be hard pressed to hang with 0.94, since the different weightings of each statistic in each formula seem likely to push the correlation either up or down, and there’s very little room for up, but that’s just me.
And I don’t know if this is on-point enough for you, but Dr. Berri made an observation–I’m paraphrasing again, unfortunately–that, since NBA player numbers have a higher year-to-year correlation than most other sports, regardless of team, teammates, or coaches, there is ample reason to conclude that the numbers in the boxscore tell us something useful about the players to whom they are attributed. Don’t know if I’m rendering it accurately; I think it’s from a post from late 2006 or early 2007. And somewhere there’s a link to the actual peer-reviewed articles behind the intentionally-non-technical book; that one, I think, is from earlier in 2006. Perhaps the issues you are raising were discussed two or three years ago.
Todd
May 13, 2009
“One could conceivably invent a metric that came to the exact opposite conclusions that WP does and still have the numbers match the team wins.”
Nice discussion guys. I like it. We could use a metric that simply
1. counts the number of points a player scores per 100 possessions to determine his offensive efficiency,
2. gives all players 1/5 of his teams defensive rebounds per 100 posessions while he is on the court on the assumption the rebounder didn’t play all of the defense, and
3. adds up the steals he gets per 100 posessions.
I’m kind of thinking out loud and there would have to be some other adjustments, but I think I came up with a, simpler, equally predictive model that makes AI the best player of his generation. Using this clearly flawed model he is also very consistent.
I have read WOW and read this blog a lot, and I have found that the most common defense to the problem of attributing stats that are predictive on a team basis to individuals is that individual performance remains consistent over time. Using the model I outlined above, we will have correlation to actual wins and consistency of production over time that is quite similar to WS even though many of the wins will be attributed to completely different players.
Todd
May 13, 2009
To make it clear, I think WS is far superior to the model I described above, but the model I showed does demonstrate the flaw with the way win score is determined and the biases involves in it’s use.
ilikeflowers
May 13, 2009
R^2Collins,
I’m not aware of this spreadsheet that you reference, although I am aware of other systematic comparisons (pot calls kettle black, etc.). Regardless, the assumption of the model is not illogical which is my understanding of the word ‘fallacy’. It’s entirely reasonable to think that “regressing wins on team statistics and assuming that the relationship at the team level holds at the individual level” might yield a reasonable representation of reality. This would be a fallacy if there was something illogical about assuming that individual stats for a 5-man team sport might reasonably capture reality and that there’s something ridiculous about claiming a relationship between wins and the measurable activities of a team trying to win. If this spreadsheet that you mention is definitive (and error-free) against w48’s predictive ability (compared to some other model) then maybe there’s a better approach, but there’s nothing illogical about this approach.
Thank you for providing mention at least of some more concrete criticism, other than ‘this is a fallacy and dberri’s ignoring it’.
ilikeflowers
May 13, 2009
The code depends upon the quality/conclusions of the spreadsheet that you site. It could be,
[H] Some other model X is clearly better than wow according to inconclusive and/or crappy analysis Y.
or
[NULL] There is no letter needed for the case in which model X is clearly better according to a definitive analysis Y, since if it comes along most here will move on to model X anyway.
mrparker
May 13, 2009
dberri,
Would you state that big guys are more consistent over time than little guys? Thats my hypothesis.
R^2Collins
May 14, 2009
Flowers, the faulty logic is that the application of team coefficients to players implicitly assumes 5-on-5 basketball is the same as 5 concurrent games of 1-on-1. And I don’t think it’s especially outrageous to claim that it’s illogical to describe a 5-on-5 game in that manner.
jbrett
May 14, 2009
R,
The unwavering tone and character or your posts has at this point provided enough of a sample that I feel comfortable saying that one of the following four statements is true:
1. You have never read one single word Dr. Berri has written (unlikely, but can’t be ruled out)
2. You have forgotten most or all of what you have read by Dr. Berri (possible, I guess)
3. You are in the habit of passing off other people’s thoughts–Berri’s, in this case; ‘five concurrent games of 1-on-1’ might be a direct quote from WoW–as your own (this is called plagiarism, I believe); or
4. It has never occurred to you that all of the ideas that come into your head are not being formulated for the very first time in the history of the universe, and THAT is why you can see no purpose to determining if the questions you raise have already been addressed (obviously not, since YOU only just thought of them).
I can tell you’re better versed in the field than I am, and that makes you someone who might offer something interesting to the discussion–but I assure you that you have yet to do so. You’re just miring us down in the newest layer of pedantic self-absorption that comes from not doing the homework.
Would it really be that arduous a task? If you walked into the middle of a dinner party conversation, with no idea what concepts had been previously batted around, even you might feel a bit sheepish about being second to tread a path; you might even recognize the rolling of eyes as meaning the same as the inevitable ‘been there, done that’ response you would provoke–but you could always console yourself with ‘How could I have known; I wasn’t here then.’
THIS dinner party, however, has transcripts, and you COULD have known, with only a modicum of effort; the tack you’ve taken here is the dinner-party equivalent of ‘Sorry I’m so late; everyone stop and catch me up.’ Do better, dude; make the effort.
R^2Collins
May 14, 2009
Getting a bit defensive, are we now? It’s interesting you should say that I need to catch up, since I’ve read the book and have been following this discussion since the book was published. It’s true that I am bringing up sentiments that are not new, but they are sentiments which have been addressed with either: “You are not well-versed in statistics enough to challenge me and I cannot teach you econometrics in a blog” (wrong), “This has been through peer-review and therefore is more valid than any other current statistical method” (wrong – perhaps the ‘peers’ you need to submit to are at APBR), “This will be addressed in the second book” (we’re waiting), or some obfuscation/non-answer that sidesteps the issue of misapplying regression at the individual level and making a team adjustment that would allow literally any method to claim the vaunted 94% explanatory power. I am merely the latest in a long line of critics who are not fan-boys to be dismissed out of hand, critics who see the structural flaws of WP and are still waiting for a satisfactory response. I’m not the first, I won’t be the last. But it’s someone’s duty to continue to hammer away at the same flaws until someone bothers to concede their existence and make an effort to correct them. Otherwise it would just be an echo chamber for the misguided and uninformed.
Nick
May 14, 2009
Jbrett,
I would say a similar problem to assuming that “that all of the ideas that come into your head are not being formulated for the very first time in the history of the universe” is assuming that other thoughts, especially those which are not particularly unique, cannot occur in other people’s heads independently, especially if they’re all thinking about the same problems.
I would put 5 concurrent games of 1 on 1 into that category. So we’re not reall y talking about plagiarism; we’re talking about tardiness.
Also: If you don’t see that the fact that you can sum individual players WP (which are based on the weights from the team statistics) to reach a .94 correlation with team wins does not all answer the question of whether or not those WP are properly apportioned between the players of a team, then I would say that you’re also a little late to the dinner party, except this dinner party is the much larger sound argument dinner party.
So you should make some effort to think for a second why that doesn’t work.
@ilikeflowers
I generally think the phrase “logical fallacy” means illogical; that’s why they have a separate phrase for it. I just take fallacy by itself to imply something that is misleading or false, and logical assumptions can be both of those things.
So I don’t think its fallacy to state that the “quite consistent” production of players over time is insufficient evidence, by itself, to justify the assumption that a regression of team statistics will hold true for individual players.
Because all the “quite consistent” production over time actually would suggest is that whatever causes a player to be productive over time is consistent. If you want to then assume that that cause is his personal responsibility for all of his statistics, fine, though it could just as easily be that same player’s consistent ability to fit into interdependent systems equally over time. But sure, say it’s personal responsibility, but that’s a second assumption and the thing about assumptions is that you assume them because either you don’t have sufficient evidence to prove them or no one disagrees with them (which is clearly not the case here).
dberri
May 14, 2009
Let me repeat:
On the topic of applying what we know from the team to the individual: “One thing to remember is that basketball players –relative to what we see in other sports — are quite consistent over time. This suggests that the numbers assigned to the player in the box score are mostly about the player. ”
On the topic of a team adjustment (and I repeat): The only way you can add a “team adjustment” to anything and make it predict as well as WP is to define team adjustment as the residual from your model. This is not what I did and would not be a correct approach (and it is not a good sign that the APBR community can’t understand this point).
It is important to learn what is meant by the “team defensive adjustment.” It is in the book and it has been explained in this forum (and it is not adding a residual to everything).
And again, this is something I have said before.
Nick
May 14, 2009
I’ll repeat:
“All the ‘quite consistent’ production over time actually would suggest is that whatever causes a player to be productive over time is consistent. If you want to then assume that that cause is his personal responsibility for all of his statistics, fine, though it could just as easily be that same player’s consistent ability to fit into interdependent systems equally over time. “
dberri
May 14, 2009
Nick,
Your story suggests that players who are “unproductive” are simply “unproductive” because their coach isn’t asking them to do productive things. In other words, Eddy Curry commits turnovers and fails to rebound because his coach tells him to do this. That doesn’t seem to be a very reasonable explanation.
Christopher
May 14, 2009
Regarding making a different model that shows AI is the “best” and the like.
Of course one could do this (and I urge you to DO IT as it’s much harder than the mental sandboxing that has occurred in this thread). But this is the _real_ issue here. WS is actually a grand data mining exercise, i.e., a model based on actual observations. The problem with, say, PER, is that it is not. If you contrast the derivation of PER with WS the machinations of the former really have no basis. It’s quite subjective. And with subjectivity you can obviously go where you want to.
ilikeflowers
May 14, 2009
R^2Collins,
‘the faulty logic is that the application of team coefficients to players implicitly assumes 5-on-5 basketball is the same as 5 concurrent games of 1-on-1.’
Or it could be assuming that the 5 positions and overall strategies are well-defined and reasonably consistent so that over a large number of observations the analysis can start with analyzing them separately. Even if you think that it’s treating the 5-player sport as 5 games of 1-on-1, it’s still a perfectly reasonable starting point given the small number of players and the well-defined roles. You then examine team effects from this starting point. Or you can assume the reverse is true and examine individual effects from that starting point – or anywhere in between. There’s no faulty logic here. Whether the assumptions turn out to be correct or not they certainly do not represent a fallacy.
Nick,
I didn’t mention anything about player consistency and regardless that is just one small portion of the evidence for wow’s approach. Also, both definitions of fallacy still work with my posts. In fact to just use it as meaning something misleading or false against wow implies that the assumptions that wow makes are misleading or false. This seems easy to establish with evidence against. The evidence for is in the book (peer-reviewed) and on this site, perhaps the evidence demonstrating that wow is so flawed as to be false will be presented for analysis.
Nick
May 14, 2009
Dberri,
I agree that’s not a reasonable explanation. It’s also not my explanation.
I’m saying the coach (among other factors) asks players to be productive in varying ways that give them varying levels of control over their production. And that as a player moves around he more or less stays in the same role thus unless there’s a more than marginal change in his teammates influence, he should be about the same unless he declines in that role.
But if we’re going to throw out caricatures, let’s just look at a teams like the 06-07 and 07-08 Bulls, where, as you state, almost all of the rotation players declined in productivity because of a decline in shooting efficiency.
In your story then all of these players just independently got worse at shooting rather than their poor performance being to some degree tied together. For though you always say that your numbers don’t address the “why,” in fact according to you they to a substantial degree do.
Because whatever the “why” is, you say it mostly has to do with the player, which means astonishingly that Luol Deng, Ben Gordon, Kirk Hinrich, Tyrus Thomas, Andres Nocioni, Ben Wallace, and Chris Duhon all discretely declined in shooting efficiency while still playing together. Otherwise their productivity wouldn’t mostly be about the players (unless by mostly you just mean like 51% in which case I’d agree with you but then say WP by your own admission is horribly flawed).
But I would say that’s not a reasonable explanation.
Nick
May 14, 2009
@ilikeflowers
I would say you have facial evidence that your assumptions are wrong if you assume that positions are a sufficient proxy for roles. Kobe Bryant and Dahntay Jones are not fulfilling the same “SG role” out there. If anything Kobe would be more similar to Tim Duncan and Jones more similar to Kendrick Perkins.
In the minimum (and this isn’t really sufficient) couldn’t you throw some usage into the mix and have you know like two axes i.e. a positional/usage adjustment.
R^2Collins
May 14, 2009
NBA decisionmakers – the ones derided as irrational here – tend to evaluate players and lump them into player types/roles that do not necessarily hinge on rigid positional definitions. And those roles tend to stay constant across teams, for instance if a scoring guard for Team A moves to Team B, he would most likely remain a scoring guard.
Contrary to what Dr. Berri suggests, those roles dictate to a large degree how much in each stat category a certain player puts up. Play a hybrid F on the perimeter, they won’t rebound as well as if you play them closer to the basket, for example. The illusion is that like I say, the roles typically stay the same every year for each player. As a result you see this as evidence that most of player production is due to the player himself. That’s one explanation, but another is its evidence that a player’s role – btw a coaching decision – influences his numbers to stay the same, because that role never changes. So how much is the player and how much is the coach? Untangle that. And how consistent is WP year to year for the rare players who do change roles?
All of this skillfully sidesteps the “team-based regression coefficients are not necessarily applicable at the individual level” critique, of course.
ilikeflowers
May 14, 2009
R^2Collins,
Nothing much more to say, I guess we should just take your expert opinions as fact.
[I] My assumptions are way better than wow’s assumptions. Trust me.
Nick,
you are assuming that roles are important enough to override position. Maybe, maybe not. The model operates on the standard positions/strategies. It is not going to account for other positions/roles currently. I’m sure that there are strategies that wow is not going to address well. The real question is how much these alternate strategies/roles really differ from the norm and how much they are actually used. There are certainly plenty of questions that remain unanswered/unquantified.
Phil
May 14, 2009
dberri,
I do not see what’s so implausible about the idea that coaches can emphasize a player’s focus in certain areas, or put them in position to excel/struggle.
Rebounding is a big one – some coaches really emphasize offensive rebounding, favor better defensive balance. Some allow players to leak out on the break, others focus upon securing a defensive rebounding.
And does there really need to be a lot of elaboration to explain how (some) players can only focus on a couple of areas? A player with Curry’s conditioning and game sense, for example, is going to struggle to contribute in a lot of areas.
reservoirgod
May 19, 2009
Prof. Berri:
Are you aware of any research linking the consistency/inconsistency of production in basketball, baseball and football with the use of performance-enhancing drugs? I ask because with all of the PED stories in the news and the conventional wisdom that football players use PEDs more than baseball players (who use more than basketball players) I found myself wondering if that’s a significant part of the explanation.
bob
March 14, 2010
this method of WP is pretty darn stupid due to the fact that it even compares wad with kobe. 1. it i comparing kobe stats with a team full of allstars to wades team full of nobodies. Wade knows he ha to do everything for his team to win as kobe did when he was avergaing 35 pts 5 reb 5 ass… when kobe had no one he did everything for that team to win they went up 3-1 in a series vs the suns when the suns were prob the best team in the nba unstopable, when people say wade is the best it is a joke. lebron is better than wade easily so is kobe melo and wade r on the same level so compare them… kobes all around game is better than wade, kobe has big men on his team and barley any shooters and plays in the triangle offense which is why he only avg 4-5 ass everyone touches the ball in that type of off, where as wade has shooters all around him no tri offense and only manages about 2 more ass so now tell me whose better!