Dave Berri's midseason analysis
by vjl110 on Jan 24, 2010 12:21 PM CST
Email
PrintTwo wolves are notable, and not in a good way.
"Turning our focus to the least productive players in the game we see that Ryan Hollins (-1.9 Wins Produced) is leading the way. He is followed by Darius Songaila (-1.5 Wins Produced). The least productive rookies include Earl Clark (-1.4 Wins Produced in only 249 minutes) and Jonny Flynn (-1.3 Wins Produced)."
Also notable is that Kevin Love (in limited minutes) is second only to Marcus Camby in per minute production.
Some people may not be fans of Berri's Wins Produced, and I think it does have some minor biases and oversites. However, given its predictive power year to year it is the closest thing to a "one stat to rule them all".
46 comments
|
1 recs |
Do you like this story?
Comments
Ya
I don’t know about Berri. Dude’s got some good arguments to make, but he comes across too strongly as another Winston/Hollinger type with his own “One stat to rule them all.” And while his stat is supposed to reveal how many wins or losses an individual player produces, I don’t think it’s an apples to apples comparison stat despite what Berri says. Ultimately each player is compared to team influenced offensive and defensive coefficient that modifies their score up or down. So on the one hand it makes sense that on bad teams you won’t have a lot of players producing wins, but on the other hand it seems rather circular in that players are penalized for playing on bad teams (or unfairly benefit from playing on good teams). Now I’m sure a lot of people will say that this is the benefit and insight of this statistic – but he rated Mike Miller as worth 10 wins last year. Was Mike Miller the second best – or best – player on the Wolves last year?
If anything the value of this stat lies in its relative measure of how much a player helps his team as compared to his teammates. So Jonny Flynn doesn’t help much because he’s not an efficient scorer, a good rebounder, and so far a terrible defender. Same with Ryan Hollins. For what it’s worth, though, he had Dennis Rodman as being more valuable than Jordon on some of those Bulls teams. Now you tell me what that means, especially considering we have our own Rodman-esque player in KLove.
"Come on Eddie, let's get serious."
Few Thoughts
I feel pretty good about a couple of my past statements looking this over.
1. – “Gerald Wallace was a Top 15 player”. This has been proven true this season. Perhaps even Top 10.
2. Lamarcus Aldridge is overrated- This has been proven true.
3. Rajon Rondo is a Superstar- true
4. Derrick Rose is incredibly overrated- True
5. Marvin Williams is better than Rudy Gay. This would seem silly on the surface. Although looking at WOW this doesn’t seem far off (Considering Shot Volume)
6. Kevin Love is the best player on the Wolves- True
7. Jonny Flynn adds nothing but scoring- True
8. Shelden Williams is a productive NBA Player- True
can i add one more thought?
9. the troy murphy problem still exists.
for all the progress made by win shares in the advanced stats department, WS still overvalues big guys who do nothing on defense except rebound and undervalue defense minded perimeter players who don’t rebound, but actively create the opportunities to get those rebounds.
that’s not to discount wins produced as a stat, just to point out its most obvious flaw. rebounding is very important to defense (and as a result, it’s very important to winning games, too), but there’s more to it than that. that’s why we need adjusted plus/minus and the like (noisy though they may be) to fill in the gaps. they do the job of catching the troy murphys of the world and as long as you only use APM as a story-checker, the noise isn’t such a huge problem.
heart of a champion, will of the warrior.
I should clarify
This has some good insights but is in no way definitive. Kevin Love isn’t the 2nd best player in the NBA. Troy Murphy is a above average player not an All NBA Type Talent.
by Jose Cordoba on Jan 26, 2010 10:05 AM CST up reply actions
Agreed.
I think WP has two significant flaws.
1.Underrates players with exceptional man defense, and overrates those that slink off of their men to collect boards (cough-Love-cough).
2.Underrates players listed as bigs that play the role of a small on the court, and overrates smalls with the skill-set of bigs. - for example: players like Mike Miller and Jason Kidd, get too much love for their rebounding b/c they get adjusted by their contribution relative to other PG/SGs, and bigs that don’t get rebounds but play perimeter D and float around looking for 3s on offense.
I agree to a point but...
To his credit, the system does a really good job of maintaining consistency when players change teams or situations. For example both Gasol and Ariza scored almost exactly the same the year before joining the Lakers as after (although Ariza hasn’t maintained production after leaving the Lakers). Berri has checked consistency across decades of stats and found no significant error in prediction related to changing teams.
Additionally, the aggregated wins produced for the players on a team is very good at predicting team wins from year to year (and he has won the stats-off playoff picks competition at least the last two years.)
It does seem like the system overvalues rebounders to some extent, but I think much more of that comes from the fact that everyone else undervalues the hell out of them.
The wins produced has its problems but it is miles beyond PER (b/c it actually predicts wins) and miles beyond any +/- measures (b/c it actually makes consistent predictions without massive sample sizes). And it is miles and miles beyond subjective observation and biased perusal of box scores. It may not be the word of god but I trust wins produced over any other means of evaluating talent.
I will say this
while I don’t disagree with anything until the last paragraph, I’d say you are far underrating subjective observation. At least subjective observation to the trained eye. I’m not saying that subjective observation is better or worse at picking out guys in the middle of the pack, but I’d really have to guess that you could find quite a few guys underrated (and to a lesser extent, overrated) by WOW. I’m guessing good rotation-defense players give some of their wins to their teammates whose man they are rotating to. I’m also guessing that guys who do the “little things” don’t get as much respect as they deserve. Valuing Rodman over Jordan definitely backs up the rebounding over scoring thing. But at some point, if you score as efficiently as Jordan did, it’s better than rebounding. I mean, there are only so many rebounds to go around when you are that efficient after all.
I personally hate subjective observation,
Jordan by no means did poorly with WOW. I believe he had at least one season over .500 WP48 which is absurd, and beyond what anyone today is putting up. That is because Jordan was, as you say, an amazingly efficient volume scorer. Jordan was awesome by all measures. The problem with subjective observation in that case was an underestimation of just how good Rodman was. He was every bit the freak that Jordan was just in the less glamorous role of rebounding.
“I’m guessing good rotation-defense players give some of their wins to their teammates whose man they are rotating to”
WOW only uses box-score statistics so this would not be the case.
I think you have a point with the “little things” comment. The biggest weaknesses of WOW are “little things” and man defense. I think that is where +/- can be used to add more information (but you need to be really careful b/c of the consistency problem).
The biggest test for stats is the Rockets
Subjectively, they shouldn’t be good. But they have the 13th-best winning % in the league while building a roster based on much more advanced stats than the ones used in mainstream media. If they stay above .500 and/or make the playoffs, Daryl Morey should be Executive of the Year.
by pagingstanleyroberts on Jan 24, 2010 4:43 PM CST up reply actions
Wilkins
He’s also a great measure of subjective/objective. He doesn’t seem to do a lot, but the team has been more competitive when he’s played, and he’s 3rd in Wins Produced according to Berri’s calculations.
by pagingstanleyroberts on Jan 24, 2010 4:45 PM CST up reply actions
Dave Berri captured both
He called the Rockets before the season
(http://dberri.wordpress.com/2009/08/13/does-houston-really-have-that-big-of-a-problem/)
And if you look at the link to todays article he has Wilkins as the second best performer on the wolves to date. Although as Jefferson improves to his career average he will overtake Wilkins.
"[Rodman] was every bit the freak that Jordan was"
That is quite the statement. Jordan was a freak in multiple roles, and freak would be an understatement to describe much of his game. Rodman was a tremendous defender and rebounder, but I wouldn’t put him in the same echelon as Jordan.
I am not too familiar with how many stats are calculated, but if they only use boxscore statistics, this is already introducing some large flaws into the “one stat to rule them all.” It is missing an extremely large part of the game that the eyes can easily pick up on. Again, I’m not going to ask my friend who casually watches basketball who to draft/trade for, just as I wouldn’t look at one stat or only stats to decide. To a trained eye, observation can be extremely effective. Also, to a trained mind, statistics (in comination) can be very effective. But I would much rather take my chances on somebody only having seen every single game of theirs rather than by only looking at their stats. Defense, IMO, is a very underrated statistic. Probably because it’s very difficult to quantify. But I would say that it’s pretty easy to see (again, trained eye, large sample size) who is and isn’t a good defender.
come to think of it
baseball has fielding percentage…why can’t basketball have rotation percentage?
by TheEvilProfessor on Jan 24, 2010 11:07 PM CST up reply actions
Nobody would put Rodman in the same echelon...
except for blind stats. That is the whole point.
There is only one thing that is important for a statistic and that is its predictive power. WOW is extremely predictive. It does a very good job of calling which teams will struggle and which will succeed based on the talent they have on the team.
So yes, by only using the box score stats it would appear that WOW is missing out on a huge part of the game BUT, because it has so much predictive power, either that stuff doesn’t matter as much as we think or, more likely, something about those non-boxscore abilities is subtly captured in box-score stats. For example, a guy that gets a lot of steals and rebounds is also likely to be a good defender.
(If you peruse the list in the linked article WOW seems to consistently rate guys like Gerald Wallace that are “underrated” by the popular press b/c of their defensive ability really high)
It is easy to pick-out things that we think are important that aren’t captured in WOW (the same cannot be said for adjusted +/-), but the difficult part to wrap our heads around is the fact that it really doesn’t matter. The independent variables that go into the model are irrelevant when all you are interested in is the predictive power of the score that it shoots out. And so far, no other statistical model comes close.
plus when you consider
that lineup success is all about having synergistic lineups, which are almost impossible to contstruct using any stats, it really hammers home that there never will be one stat to rule them all. And to be honest, I never want one. If they come up with it, I can guarantee you that your employer will come up with one too.
Employee Question:
“Why did I just get fired boss?”
Boss Response:
“Because your employee composite score just fell below my replacement level cost and I’m looking for an upgrade.”
Conclusion:
Ouch and/or burn.
by TheEvilProfessor on Jan 25, 2010 9:44 AM CST up reply actions
Whenever this issue is brought up, you
should counter with the “synergistic lineup” argument. Sums it up simply.
Take two teams, each has a PG with a 9.5 PER, and a SG with a 10.2 PER, but they don’t complement each other on the court. Now trade the PGs….Two months later the PGs each have a 15.2 PER, and the SGs have a 14.1 PER. Did the players suddenly get better, or do stats fail to tell the whole story?
Decision making chart
Using Stats to make your decisions = EPIC FAIL
Using Stats to help your logical mental though process in order to make more informed decisions = ROCKSTAR
by TheEvilProfessor on Jan 26, 2010 7:49 AM CST up reply actions
I think we're just talking semantics right now
I agree with you that WOW does a good job and I’m not challenging that another stat is better (whether that’s true or not). I just think it ignores a large part of the game that can’t be (or has yet to be) quantified. It’s very scientific in its approach and like science, in a perfect world or a vacuum, the stats are perfect and make sense. Crappy analogy alert: It’s like gravity. If you drop something in space it will accelerate at exactly this speed. Put it into our environment and gravity can only give you so much of a prediction of how fast the object will fall and/or accelerate. You need to take into account different variables and a couple things that can not be measured to more accurately predict the true measurement.
As for your subjective hate of subjective analysis, we’ll just have to agree to disagree. I prefer to use both qualitative and quantitative analysis (not saying you don’t, but I obviously value it more) to determine/project wins. Like I said earlier, I don’t know exactly how each stat is calculated, but I’m pretty sure that there hasn’t been a stat yet that will project how injury-prone someone is, their mental problems and the effect on the game, potential (and bust-potential), locker room presence, etc. There are just so many other variables that you can know but you can’t measure. For example, my opinion of Evan Turner is higher because I have heard about his work ethic and the kind of humble person he is. While the latter might not affect how he plays, it affects how I relate to him as a player on my favorite team and tells me that he will probably be a better teammate than Born Ready.
you are correct that our disagreement is one of degree...
There is only one major problem that I have with the role of integrating subjective and objective measures and that is the “this stat is wrong b/c it says player X is good and I know that player X is not good”.
Otherwise I am fully aware that any one stat requires a human interpreter that needs to use some less formal sense. i.e. so and so is injury prone, so and so has a drug problem, or so and so is so physically gifted his upside has to be higher. I get that.
The issue becomes one of extrapolating to individual contribution
It seems to me, if I understand correctly, that the model is validated only for team performance. The weighting of variables is derived from regression of how they influence team wins, and the model is then tested by team composite win scores (derived from the calculation of these variables across individuals and then summed) scored against actual team wins. However, there is no validation for the use of individual player scores within a team, and especially across teams.
For instance, across the NBA, generating an assist has a certain predictable contributory effect to a team win, calculated by averaging across all teams. However, the importance of an assist on one team, playing one system, will not be as great as on another team. This would have some impact on the predictability of win score for that team, but a greater impact on the win contribution of a particular player on that team.
But I’m no statistician, so correct me if I’m wrong.
I'd actually be interested to see
how many of these statistics would hold up for an average guy, say you or me, playing on a pro team. Would my [adjusted] +/- be terrible even if I was surrounded by the best 4 players in the league? Would I have a -30 win share and completely detract Lebron’s effect and then some? Wonder what my PER would be…
Sorry, completely irrelevant and pointless, just an interesting thought.
Subjective observation is the key
and will separate the great talent evaluators from the pack. Think of it this way, if stats were all that mattered, why would scouts and GMs be necessary, let alone things like pre-draft workouts.
That said, stats are a great counter balance to prevent clouded judgement.
So what does that tell you about Love's limited minutes in the last few weeks? --- TANKING!!
Rambis and Kahn are limiting Love’s minutes to prevent too many wins that would hurt draft position.
by Kevin Love Jefferson on Jan 24, 2010 4:55 PM CST reply actions
I wish this was true
Because it would then mean that Rambis and/or Kahn had SOME notion of advanced statistics.
Berri's analysis also tells us again that Sessions has been more productive than Flynn.
Flynn is being played for development…. but Love isn’t!
by Kevin Love Jefferson on Jan 24, 2010 4:57 PM CST reply actions
Loving the Brew?
On the warts thread, I was pretty brutal on our boy Corey. However, I think there is substantial data in this analysis to prove that his “warts” come with a substantial enough cost that they don’t need to be accepted as a mildly annoying consequence of his frenetic efforts on defense.
Now, if his offensive improvement continues, these data will not look as horrific by the end of the season, but right now, according to these stats (and I already noted my caveat about correctly portioning team performance across individuals) Corey’s production rate has cost the Wolves 1.4 wins through the first half of the season. That’s not a lot, but there are only a couple of players who have had the same negative impact across the league (one notably being our own Mr. Hollins). There are other players with a bigger negative impact per minute, but they do not see the court enough to have Corey’s overall effect.
The cost is more obvious when you consider replacing Corey’s minutes with another NBA player. I started with Kelenna Azubuike, because a) I like his game, b) he’s not a star, and c) he is similar in that he alternates between the 2 and 3. Giving Corey’s minutes to Azubuike produces, by my calculation, 6.4 wins. All together this is an 8 win swing by giving starters minutes to a player who has not (over the first half of the season) provided the kind of output that has been shown to win games.
Of course, Azubuike was great before his injury this season, and showed a very high WP48 (above Kobe’s) , so redoing the analysis with a less impressive wing like James Harden (say the Wolves got him in the draft) shows a swing of over 4 wins. Still well in the lottery, but less painful. Swap Brewer for Azubuike, undo Love’s injury and give him all of Hollins minutes (giving hiim the same amount of court time as Big Al so far) and we have as many wins as the Thunder. OK, that seems a bit farfetched, but the game gets a little addictive after this season’s struggles. Bottom line, where are the data suggesting that playing Brewer helps this club?
Brewer has improved but if you look at his efficiency. If he sustains his offense and becomes a lock down defender he's a keeper.
Brewer’s efficiency per 48 minutes
Nov 16
Dec 15.8
Jan 19.2 (thru 1/25)
Other players for comparison.
per 48 minutes for the season:
Anthony Parker 13.6
Nick Young 14.3
Courtney Lee 15.8
Sefalosha 15.9
hinrich 16.6
Salmons 17.2
Ariza 17.3
J.R. Smith17.9
Ronnie Brewer 18.5
Ray Allen 18.8
Gordon 19.0 (clippers)
Jason Terry 19.0
Ben Gordon 19.1
Iverson 19.2
Willie Green 19.3
Foye 19.6
Bayless 19.8
Mayo 19.9
stephan Jackson 20.0 (Bobcats and Warriors)
Wilson Chandler 20.3
Harden 20.8
Vince Carter 21.8
Crawford 21.9 (Hawks and Warriors)
Kevin Martin 22.6
Richardson 22.8
mike Miller 22.8
Monta Ellis 23.7
Joe Johnson 24.7
Iguodala 24.7
Tyreke Evans 24.8
Ginoboli 25.6
Roy 26.9
Azubuike 27.8
Bryant 29.7
Wade 32.4
by Kevin Love Jefferson on Jan 26, 2010 2:29 PM CST up reply actions
Jonny Flynn adds nothing but scoring.
I have seen this written many times now and am very curious what others feel each other member of the roster adds. I am not disagreeing about Jonny. I just think a discussion about it would reveal that way too many of the players on this roster contribute only one thing and sometimes nothing.
It's very true
and also why our team is so… bad.
Love and Jefferson both score and rebound. Love also passes.
Brewer scores on some nights and hustles on every night.
Sessions adds patience.
Wilkins does a little of everything, but nothing well. Same with Gomes.
Flynn adds scoring and swagger (sometimes).
Hollins adds comic relief.
That's not entirely true
he adds foul trouble to the other team. At least when they call it. :)
by TheEvilProfessor on Jan 26, 2010 5:07 PM CST up reply actions
Haha
no I meant it’s very true that many of our players only add one or two things to the team. Not that Flynn only adds scoring.
Of course you bring up another good point here though, Hollins also adds foul trouble to our team. Not in that he fousl out, but he puts us in the penalty too early.
I love the look on his face when
he gets called for a foul after he completely misses the ball on a block attempt and nails the guy’s arm.
“Who me? I got all ball!”
by TheEvilProfessor on Jan 27, 2010 8:31 AM CST up reply actions
When Hollins was younger
he was told he could be a combination of Duncan and Rasheed Wallace. Unfortunately this is the only part of either of their games he picked up.
Win Shares as a predictor/function of team success
I went to look at the players the Wolves employed during their years in the playoffs. Of the 61 guys who played for the Wolves during that span, there are 26 players who ever finished with negative win shares, whether with the Wolves or another team:
Dean Garrett: -0.1, 02/03 with Wolves and Warriors
Stephon Marbury: -0.3, 08/09 with Celtics
T-Hud: -0.2, 99/00 with Clippers; -0.1, 07/08 with Warriors
Trenton Hassell: -0.1, 02/03 with Bulls
Tom Hammonds: -0.1, 00/01 with Wolves
Bobby Jackson: -0.6, 96/97 with Nuggets
Chris Carr: -0.3, 99/00 with Warriors and Bulls
Mark Madsen: -0.3, 07/08; -0.1, 08/09
Reggie Slater: -0.1, 95/96 with Mavericks, Nuggets, Raptors
Kendall Gill: -0.4, 00/01 with Nets
Stanley Roberts: -0.1, 93/94 with Clippers; -0.1, 98/99 with Rockets
Hollywood Robinson: -0.1, 98/99 with Clippers and Wolves
Rod Strickland: -0.2, 04/05 with Rockets
Doug West: -0.1, 89/90 with Wolves; -0.1, 00/01 with Grizzlies
Reggie Jordan: -0.2, 99/00 with Wizards
Michael Olowokandi: -0.3, 98/99; -0.8, 99/00; -0.4, 00/01; -0.2, 02/03, all with Clippers; -0.1, 06/07 with Celtics
Stojko Vrankovic: -0.1, 90/91 with Celtics
Mike Wilks: -0.3, 05/06 with Spurs and Cavs
Micheal Williams: -0.2, 88/89 with Pistons; -0.1, 95/96 with Wolves; -0.1, 98/99 with Raptors
Anthony Goldwire: -0.1, 02/03 with Spurs and Wizards; -0.1, 05/06 with Clippers
Robert Pack: -0.6, 95/96 with Mavs
Darrick Martin: -0.2, 01/02 with Mavs; -0.1, 07/08 with Raptors
Paul Grant: -0.1, 98/99 with Wolves and Bucks
Cliff Rozier: -0.1, 97/98 with Wolves
Shane Heal: -0.4, 96/97 with Wolves
Will Avery: -0.5, 99/00 with Wolves; -0.4, 01/02 with Wolves
What does this indicate? I think a couple of things: 1) Even decent players can be unproductive on really bad teams (Bobby Jackson and early-career Robert Pack); 2) Win Shares don’t always stay consistent over a career (Jackson was positive for every other year; and 3) a team’s worst players can’t be more than -0.3 in win shares. Note: right now, Pavlovic is -1.0. Maybe observation and stats can tell the same things. The Wolves had some seriously deficient players during that run, but they didn’t usually have a statistically-negative impact, with the Wolves or with their other teams.
by pagingstanleyroberts on Jan 27, 2010 12:14 AM CST reply actions
In Fairness...
“Win’s Produced” and “Win Share’s” are two different statistics. Berri uses Win’s Produced.
I know
But they’re a similar stat. Basketball-reference’s site allows for searches; otherwise, I can’t compile the data.
by pagingstanleyroberts on Jan 27, 2010 3:18 PM CST up reply actions
Here’s a quick rundown on the difference between the two from GSoM’s resident stat guru, jae:
"Win Share" and "Wins Produced" are entirely different calculations. They factor different things into their equation.
"Win Share" for a team takes the team’s actual victories into consideration, then (more or less) partitions them out to players on the team according to their playing time and statistical contribution. It will add up to being rather close to the team’s actual wins, which isn’t surprising since it’s using that as part of the equation. Like the name suggests, it credits a player for his "share" of the team’s victories. Evaluating it vs. team wins isn’t meaningful. Win Share also tends to reward higher volume scorers, even at lower efficiencies, though not nearly as much as some other methods.
"Wins Produced" estimates a player’s change in the probability of his team winning based on the player’s individual statistical contributions, with the equation derived from regression of the statistical contribution against probability of victory. The team’s record is not part of the equation. It winds up being remarkably close to actual team victories. More than any other "box score" method, it only rewards high volume scorers if they do so at above average efficiency.
heart of a champion, will of the warrior.
by cap'n hack on Jan 27, 2010 5:17 PM CST up reply actions 1 recs
Comments For This Post Are Closed
Secondary Sidebar
User Tools
Related FanPosts on Canis Hoopus
FanPosts
Community blog posts and discussion.
Recommended FanPosts
Recent FanPosts
Next Game
In Association With
Ball Don't Lie - NBA - Yahoo! Sport
- Days of NBA Lives: Wherein Roy Hibbert tap-dances to Maine
- Grizzlies Valentine’s Day cards are the key to love
- Kevin Love would like the NBA to take the MLB’s approach on All-Star games
- Kobe Bryant fires a few hilarious salvos at Jeremy Lin
- Brandon Jennings states the obvious regarding his 2014 free agency, everyone goes nuts
Certain photos copyright © 2012 by Associated Press or Getty Images. Any commercial use or distribution without the express written consent of Associated Press and Getty Images is strictly prohibited.
