The All-Star game starters are chosen by the fans. The All-Star reserves are chosen by the coaches. And the assistant coaches get to choose the T-Mobile Rookie Challenge rosters.
Previously I looked at the choices of the fans and coaches in terms of Wins Produced. Today I am going to offer my review of the choices made by the assistant coaches.
Evaluating the Rookie-Sophomore Rosters
When we look at the rosters in the Rookie Challenge game it’s important to ask: What are the criteria for selecting these players? Are these players supposed to be the “best”? How exactly is “best” defined?
Although I am not sure how the assistant coaches defined the term, if we define “best” in terms of “most productive”, we see that a few players fall short of this particular definition.
Table One: The Rosters in the Rookie Challenge
For the rookies, Al Horford, Sean Williams, Mike Conley, and Jamario Moon look like they belong in this game. Each of these rookies posted a WP48 [Wins Produced per 48 minutes] that was above the average mark of 0.100. But Yi Jianlian, Kevin Durant, and Jeff Green appear to be odd choices. The productivity of these players falls into the negative range.
For the sophomores, the quantity of quality players was higher. Ronnie Brewer, Rajon Rondo, Brandon Roy, Jordan Farmer, and Paul Millsap are all above average this year. Daniel Gibson, Rudy Gay, and LeMarcus Aldridge are not above average, but are clearly in the positive range. If the sophomore roster stopped at this point, the average WP48 would be 0.125. This would be consistent with a 51 win NBA team.
Unfortunately for the sophomores, the assistant coaches decided to add Andrea Bargnani to the team. Across the first half of the 2007-08 season, Bargnani was the least productive player in the league. His addition reduces the average WP48 to 0.088, a mark consistent with a 36 win NBA team.
With Bargnani on the team, the edge the sophomores should enjoy over the rookies is greatly reduced. Certainly the sophomores should be favored, but the margin is not quite what it was last year.
This analysis leads to three additional questions:
1. Who else could the assistant coaches chosen?
2. Do you need Wins Produced to make your selections?
3. Are the sophomores always favored in this game?
Who Else?
To answer the first question, let’s turn to Tables Two and Three.
Table Two: The Rookies at the Midpoint
Table Three: The Sophomores at the Midpoint
When we look at the rookies, the most glaring oversight is Joakim Noah. Noah leads all rookies and sophomores in WP48. Of course, Noah also yells at assistant coaches. So that probably didn’t help him get on this team.
When we look at the sophomores, Josh Boone and Kyle Lowry certainly could have expected a call. The omission of Boone is especially interesting. Boone is above average. He has also played nearly as many minutes as Bargnani. Yet Bargnani is going and Boone is staying home.
Making the Selection
Why was Bargnani chosen over Boone? One suspects two forces at work. First better scorers are probably chosen first. And by “better”, I mean players with a higher scoring average per game. Bargnani averaged 8.7 points per game across the first half of the season. Boone’s average was only 6.9.
In addition, Bargnani was the first overall choice in the 2006 draft. Boone lasted until the 23rd pick. It’s possible that the assistant coaches were still influenced by the initial assessment of the players. Of course, a year and a half has passed since the 2006 draft, so one might think that draft position should stop playing a role in evaluating players [although readers of a 1999 paper by Colin F. Camerer and Roberto A. Weber ("The Econometrics and Behavioral Economics of Escalation of Commitment: A Re-Examination of Staw and Hoang's NBA Data." Journal of Economic Behavior and Organization, 39, 59-82) would suspect otherwise].
When we turn to the individual statistics, we have plenty of reasons to think the evaluation of Bargnani would have changed since the summer of 2007. Specifically, consider Table Four.
Table Four: Durant, Bargnani, and Morrison
Table Four presents that mid-season performance of Durant and Bargnani. For comparison sakes, the performance of Adam Morrison – a player selected to the 2007 Rookie Challenge game – is also included.
The individual statistics for each of these players paint a clear picture. Durant is below average with respect to shooting efficiency, steals, turnovers, and assists. Bargnani comes up short with respect to shooting efficiency, points scored, rebounds, steals, blocked shots, assists, and personal fouls. And Morrison was below average with respect to everything except field goal attempts, turnovers, and personal fouls. In sum, the individual statistics suggest all three players should not be counted among the “best”.
When we turn to Win Score, we see the same story told by the individual statistics and Wins Produced. All three players did not play well prior to being selected by the assistant coaches. Again, all three could score (relative to other rookies and sophomores). And all three were very high draft choices. Still, each was far below average and should not be rewarded for what they have done thus far in the NBA.
Are the Sophomores Always Better?
I am going to end this column with a brief comment on the sophomore’s record in this game. Currently the second-year players have a five game winning streak and have won six of the eight challenges held. This should not surprise, since an average rookie posts a WP48 of 0.047 while an average sophomore has a mark of 0.076. In sum, sophomores tend to be better than rookies.
Last year the sophomores were quite a bit better and the game was a blow-out. This year, as noted, the teams are closer. At least, they are closer as long as Bargnani plays and fails to produce. If somehow Bargnani can stay on the bench, or better yet, if he starts playing like a number draft pick, the sophomores should again cruise to victory. At least, they would cruise if our sample was greater than one. With such a sample, anything could indeed happen (although I still suspect the sophomores will win).
- DJ
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.
It’s a shame Boone isn’t going. Since being inserted into the starting lineup he’s averaging 10.4 points, 8.8 boards and just under a block and steal per game on 53.5% FGs. That FT% is hideous, but he’s been a reasonably effective player as a starter this season.
Andrew
EmptyTheBench
I’d be interested to see what effect draft position has. As best I can see Bargnani is getting notice because of where he was drafted, but . Is there some sorting that happens after year one where draft position stops being so ridiculously responsible for choosing the team?
(How did Mickael Pietrus wind up on the Soph list? While he plays like he’s new to the league from time to time, he’s been around several years now.)
Andrew Thell is right. Boone, who hasn’t really played that many minutes in his career, is coming on fast and looks to be a very good big man, esp. once he gets his FT up to the 60% range. He did a nice job defending Dirk and Jefferson this week, although he had lots of help on both.
I’ve been a reader of the Journal for a while, and I always come back to the same conclusion: not everything is as easy as a wins per 48 minutes breakdown, or Win Score number. Durant, Yi and Green aren’ t ‘odd choices,’ they’re just good rookies who aren’t particularly efficient. Same as many other ‘good’ players who don’t get the benefit of the Win Score.
What do you mean by “good”, Chris. If they aren’t particularly efficient (and they aren’t) what is “good” about them? How are they helping their team when they aren’t efficient? Or can you be “good” without helping your team?
Chris–how is a player ‘good’ if they don’t help their team win?
While WP isn’t perfect, those players aren’t helping their teams win. I don’t think they’re hurting their teams, (while they have negative scores, there is value in having young guys “see the elephant,”) but to me, good implies ‘above average.’ Durant and Yi show potential–I’m not as sure about Green–but they have plenty of room for improvement. Flashes of brilliance should not be mistaken for their average contribution.
Noah, Horford, Moon, Conley, Williams, and Scola have impressed me in what I’ve watched–not in a “they’re the best player on their team” way, but in a “look, a rookie that helps his team” way. And when it comes to rookies, that’s really all that should be hoped for.
Off topic, if my back-of-the-envelope calculations are reasonable then the Jason Kidd trade would push the Mav’s into Celtics/Lakers territory.
Dr. Berri,
Like your famous prescient column on the implications of the Iverson trade last year are you going to do a prediction column for Mavs/Nets after the Kidd trade goes through? Seems another good test case. :-)
I’d love to see a prediction on how Mavs/Nets do with the roster changes!
1. I love your site.
2. I actually belive the rocket. But it’s starting to get lonely out here on the ledge.
3. I love your analysis about 3 great players in the NBA. I wonder how that applies to the 2008 SA Spurs. Right now (without injured Tony Parker) the Spurs have only two top players (Manu G. and Tim Duncan). They seem to be able to win better than 64% without a third great player. IMO, Oberto + Finley have been solid (WOW w/ in 110?) per min played, but still….
Does this cast doubts on the “big three ” hypothesis, or does 65% = a mediocre team?
Do the v. good Kidd + Lakers trades spell doom for the 2007 champs?
KIdd won’t help the Mavs as much as Berri thinks because the thing Kidd is best at for a guard (rebounding) is the thing that is most overrated in Berri’s system.
Iverson helped the Nuggets more than Berri predicted because the thing Iverson is the worst at for a guard (rebounding) is the thing that… again… is most overrated in this system.
Randini,
That’s why a prediction post would be good. I don’t think Dr. Berri overweights rebounds and the prediction will be support (or refutation) of my belief!
Nice website, look at mine when you have some spare time
http://toronto-lawyer.blogspot.com
Where’s the evidence that iverson made the nuggets better? They were really good at the beginning of last season, and started to struggle when half their team got suspended and they traded andre miller. I’m a huge iverson fan, but it’s hardly obvious that iverson has significantly helped the nuggets, and didn’t berri correctly predict that the sixers would improve? The kidd trade if it happens most likely would improve dallas in the short-term. That said i wouldn’t trade 10 productive years of harris for 2 good years of kidd.
Also, is there any data for the WP48 for lebron james his rookie year? Cuz he shot a pretty low percentage too if i recall correctly, and i was wondering how productive he was compared to durant, and other big name rookies who are supposed to be great.
Lebron had a 0.099 his rookie year.
For more info on #1 picks, check out when-will-we-know-about-greg-oden.
Animal,
If/when Kidd doesn’t help Dallas as much as WoW would predict (or at least, his ability to rebound above average won’t translate into a net rebounding advantage for his team), do you think that WoW will be conveniently overlooked and the emphasis will be on the risks associated with acquiring an aging guard who is on the downside of his career?
Mark, even if you think WoW overweights rebounds I would be careful about projecting that Jason Kidd doesn’t help the Mavs quite a bit. If I recall, Kidd always ranks very, very highly in plus/minus because he is very good defensively and presumably makes everyone around him much better. Ignoring WoW, I suspect plus/minus would project Mavs are helped a lot. (Disclosure: I’m a big fan of Jason Kidd and the Nets.)
You don’t have to wait to see what happens when Kidd joins a new team. Look at his career. In 7 seasons with NJ, Kidd has averaged 281 rebounds above his position average per season. But his teams have averaged just 3 rebounds above average per season (i.e. almost exactly average). In the previous 4 full seasons with Phoenix, Kidd averaged +191 rebounds per season, but his team was -1 (again, almost exactly average). WP says Kidd’s rebounds were worth about 8 wins above average per season for his team. But it’s not clear the rebounds actually materialized for his teams.
Bad teammates? Maybe. But we see the same pattern with Ben Wallace. And Garnett. And Camby. Or look at this year’s WP stud, Dwight Howard: he’s supposedly contributed 248 rebounds above an average center (close to half his WP value), yet his team is only +19 for the entire season, barely above average. (Meanwhile, Curry’s teams have been above average over his career, despite the damage allegedly done by his poor rebounding.)
Are you saying you’d rather have eddy curry at center than ben wallace or marcus camby?
Guy Molyneux,
What if Kidd’s rebounding allowed other players to focus on other stuff such that the team wins inreased as dberri projects but the team rebound total doesn’t increase one for one?
Guy Molyneux,
Did the Nets’ team improve similar to what dberri would have projected (based on trailing stats to that point) before Kidd joined the team? That’s what you should be looking at and not just total team rebounds.
“You don’t have to wait to see what happens when Kidd joins a new team.”
In the past, when he’s joined teams, they’ve improved, usually by a substantial margin. His rookie season, the Mavs improved 23 wins.
When he went to the Suns, in his first half season there, they went 23 and 10. They were 17 and 32 in games without him before the trade.
When he went to NJ, the team turned from a 26 game winner to a 52 game winner.
Those are dramatic improvements each and every time and dramatic of the magnitude in line with the rather high WP Dave’s metric indicates. Did he add rebounds to the team total? Maybe not. Did he add wins? Yeah. He did.
Jason makes a good point. Mr. Molyneux, if you’re trying to disparage WoW then you really shouldn’t be picking Jason Kidd as a test case to do so! Kidd’s new team improved exactly as one would expect after each trade.
This is not for this specifically, but just in general. Shouldn’t pythag wins be used? When speaking of what Kidd did, or really what anybody did, shouldnt it be talked about in point differential since we are talking about statistics. For example, when Jason Kidd improved the Nets from 26 to 52 wins, maybe they should have won more or less according to pythagorean wins, which is only what Kidd can account for. H e can not account for the luck.
Also, in all of these situations that Jason brings to our attention, while Kidd certainly improved the teams he was on, it is not as if he was the only addition or improvement to the team. Maybe he was in the Phoenix case because it was a mid-season trade, and I don’t know if there were any other trades or injuries effecting the Suns off the top of my head, but I know with NJ and Dallas Kidd can not be credited with the sole reason for his teams improvement. Certainly a part of it, but definitely not all of it.
And definitely so in Dallas, considering I would bet in his rookie year he did not accumulate the 20+ wins WoW routinely says he gets now.
Jason/Kent:
I was replying to Mark’s comments about whether Kidd’s arrival would produce the expected change in rebounds. I think we already know the answer to that. And I’m not holding up Kidd as a “test case” for WP — my point is precisely that virtually ALL “great rebounders” do not deliver the rebounds to their team that WP seems to predict.
Could Kidd nonetheless be just as valuable as WP says? Sure, he could be (you know what they say about stopped clocks, blind pigs, etc.). However, one player’s teams’ performance after his arrival doesn’t prove anything by itself, as I’m sure Jason would agree. You’d have to control for other changes in the team, and look at a lot more players, to know if changes in team performance match what WP predicts. (And as we all know by now, the only attempt thus far to measure that has shown WP to be a rather poor predictor.)
The other problem I have with Jason’s Black Box version of WP — which says that we don’t really know how these players help their teams, they just do — is that it would seem to imply that WP is double-counting player performance. If Kidd causes other players to shoot more efficiently, defend more effectively, or whatever, then that production is being captured in those players’ individual stats as well. So aren’t Kidd’s teammates then being overvalued by WP?
And if the Black Box Theory is going to be the new basis for belief in WP (it certainly isn’t the authors’ original argument), shouldn’t someone try to find som actual evidence to support it? Just asking….
Guy, very good point about double-counting. I hadn’t thought of that.
Guy, you have a very interesting and peculiar concept of “proof.” No, one player does not validate or invalidate a model.
But are you really saying that Kidd’s high WP and the observation that he’s improved every team he’s ever gone to is a broken clock issue? That it just happened to be so? It appears to me that this is what you’re suggesting.
You appeared to focus purely on the parenthetical portion of Mark’s sentence, Guy, in a particularly obsessive manner focusing in only on one portion of the question. Forgive me for not being able to read your mind when you did not explicitly mention that you were disregarding his larger question. Forgive me for concluding that your statement about how “don’t have to wait to see what happens when Kidd joins a new team” references only rebounds. Since the bulk of your posts seem to boil down to “rebounds are not additive”, something you’ve voiced ad nauseum, I guess I should have assumed that you were responding only to a portion of one sentence since the bulk of your posts seem to be so incredibly rebound-centric, that you would disregard a larger issue of what was going on with the team.
I take issue with your dismissal of my position as a “black box.” We’ve been through this before and you still seem to come back to the totally irrelevant “what Dave was originally thinking” issue. Again, that’s something for him to comment on. I don’t read his mind any better than I read yours.
But I do take issue with the notion that there’s ‘double-counting’. It does appear you have zero idea of what I was actually suggesting. I am suggesting *nothing* that involves double counting. This is in fact *not* the case if, as you’re so obsessed with saying, rebounds are not additive. If Kidd’s rebounds *do not* add to the team’s total (as you have said they do not appear to) but his presence seems to have a monumental improvement on the team of similar magnitude (sensu lato) to that predicted by his WP, we are not double counting anything at all. We’re seeing a shift in tasks and responsibilities, his rebounding detracting from rebounds others would/were getting but providing them with opportunities to replace their rebounds with some other statistical measure. This hypothetical (one that can be tested, albeit with a requirement of considerable input of data) is not double-counting. Now as I’m so fond of saying, this is an empirical question, one that can be investigated. You are correct that it should be investigated. Care to do it? It seems to bother you deeply. Perhaps you’ll sleep better aftewards.
While I don’t know Guy’s problem, my problem with Jason Kidd in WoW is that according to WoW he has not gotten any worse as he has aged. To me, since his knee surgery, he just has not been the same player, defensively or offensively. And if you are to assign credit to Kidd for getting the Nets from 26 to 52 wins, than he should also be assigned credit for going down from a 50 win team to a now 40 win team at best. And this year much, much worse. Now I don’t know what Kidd’s WP48 is this year, but I am sure it is about the same as last year and the year before that, even though the team is largely the same. So if you want to give him credit for improvement in the past, I don’t see why he doesn’t get credit for them being worse off now.
Also, I am a huge JKidd fan. I thought he was robbed of the MVP by Tim Duncan when came in second place, and for a while, until the knee surgery, he was the premier point guard in the NBA. So for awhile I did agree with WoW, but not anymore.
and Jason, you said “Now as I’m so fond of saying, this is an empirical question, one that can be investigated. You are correct that it should be investigated. Care to do it? It seems to bother you deeply. Perhaps you’ll sleep better aftewards.”
So does this mean you are not 100% positive of the validity of the WoW metric or at least the rebounding portion of it? While I know you believe the test will prove the validity of WoW, you don’t actually know the answer?
I’m not 100% positive about much of anything, certainly not of any statistically derived argument. I know that there’s error in the WP metric. It’s my opinion, based on looking at data and based on looking at how the metric was constructed, that the error is low. But of course all of us are filling in some gaps with intuition and I’m more than willing to accept that data may show the intuition wrong. I do not “believe” that the tests will show anything. I suspect that they’ll confirm more than they’ll refute, and I can and have given reason for my suspicions. That there’s “an answer” is also not really an applicable statement to a statistical model and “prove” is not a concept with real meaning in science. Model approximate reality to a greater or lesser degree. They are not reality.
I have not yet seen these data or analyses though and find the obsessive arguments that it must overvalue rebounding to be headed down a dead end. The notion that a diminishing returns on rebounds invalidates the metric is not one I share and do not find that line of argument particularly convincing.
What I find fascinating is that Guy Molyneux so dogmatically and obsessively believes that rebounds are overvalued in WoW that on a quotidian basis he posts comments on multiple blogs about it. The nature of belief and interests is intriguing from a psychological perspective. Where does Guy’s passion for the issue emanate from and why does he so intransigently choose a side without empirical support? It’s all very interesting.
Antonio, Richard Jefferson was healthy when the Nets won 49 wins two seasons ago. His w48 went from around 250 pre-injury down to 50.
Guy, does you’re rebound argument hold in general and how what does it say when the same method is applied to scoring?
Mike. There are two way to rate an action:
Sucess – Failure
Sucess – Oportunities
Both are near the reality, but they are extremes and the reality is a mixture of both. WOW chose one of them for scoring, other metrics chose anotherone. But, WOW didn’t choose anyone to rate the rebounding action, at least I can’t see a stat or a statistical method to account for rebounding failures or opotunities.
Mike H:
I haven’t looked at the full range of other statistics. I have looked a little bit at shooting efficiency, and as best I can tell there is not a diminishing returns effect there. (Which makes sense, since we can count opportunties consumed for scorers — FGAs — but not for rebounders.) When I have a chance, I’ll try to post some of those numbers.
Kent: Have you looked at the work done by Eli here: http://www.countthebasket.com/blog/2008/02/05/diminishing-returns-and-the-value-of-offensive-and-defensive-rebounds/ The comments cover similar work done by myself and — more importantly — Ed Kupfer. We all find a very weak link between individual and team rebounds. My position may be wrong, but it’s not accurate to say it is “without empirical support.”
Jason: Once again I find myself mystified by your tone. You seem to take these discussions terribly personally. That’s too bad.
“You appeared to focus purely on the parenthetical portion of Mark’s sentence, Guy, in a particularly obsessive manner focusing in only on one portion of the question. Forgive me for not being able to read your mind”
I agree my original response wasn’t clear that I was addressing myself only to Mark’s rebound point. So I clarified that. I didn’t criticize you or anyone else for your quite-reasonable interpretations of what I said. Sheesh…..
“This hypothetical (one that can be tested, albeit with a requirement of considerable input of data) is not double-counting. Now as I’m so fond of saying, this is an empirical question, one that can be investigated. You are correct that it should be investigated. Care to do it?”
Let me see if I understand: You come up with a highly speculative theory that players with high rebound totals may improve their teammates in a variety of unspecified ways, despite not actually increasing team rebounds much. But it’s not your job to come up with even one piece of evidence to support this idea, rather it’s the job of other people to disprove your speculation? Good luck with that approach (I’ll pass).
“since the bulk of your posts seem to be so incredibly rebound-centric, that you would disregard a larger issue of what was going on with the team. ”
Yes, I plead guilty to being particularly interested in the issue of diminishing returns on rebounds at the moment. But I do have some thoughts on the usage-efficiency issue as well (the second biggest problem with WP), and I’ll try to find the time to share those at some point too.
“The notion that a diminishing returns on rebounds invalidates the metric is not one I share and do not find that line of argument particularly convincing”
I don’t believe anyone has said it “invalidates the metric.” What I’ve said is that it eliminates a critically important assumption underlying the authors’ own case for WP. Diminishing returns doesn’t disprove the validity of WP, but it does “unprove” WP — it leaves WP without any empirical foundation (at least until Jason can find evidence that his Black Box really works). The assumption of minimal diminishing returns is essential if you want to use coefficients derived at the team level to assess the value of individual players’ value. I honestly don’t see how you can disagree with that…..
Guy, elsewhere in another thread, you said that there was not diminishing returns in turnovers. I asked where these data were. You did not reply.
Guy,
I hadn’t seen that work. Thanks very much for posting it. Apologies for what I wrote. That certainly offers backing for some of your assertions.
I do like dberri’s emphasis on *efficient* scoring, but I’m sympathetic to what this empirical work suggests about the weighting of rebounding.
Harold A, welcome back!!!
Pingback: Marty Burns Ranks the Best « The Wages of Wins Journal
Jason:
Using 06-07 data from 82 Games, I looked at TOs by position compared to TOs at a team’s other 4 positions. There is a negative relationship, but weak. For each extra TO at a position (vs. the position average), the other four positions combined record .17 fewer TOs. My guess would be that this at least somewhat reflects how frequently a player has possession of the ball — a high usage player has more opportunities to commit a TO, and gives his teammates fewer opportunities. If you controlled for that, perhaps there would be no relationship at all. But in any case, diminishing returns on TOs appears to exist but be pretty small (with caveat that this is based on just one year of data).
P.S. Let me know if you’d like to see the data.
Guy,
the other question about the Kidd info is, are you measuring productivity? Total rebounds isn’t really important from this standpoint. What is the change in rebounds per missed shot offensively and defensively before and after his arrival?
Also, if a typical offensive rebound leads to a higher percentage shot than a typical defensive rebound then that may mitigate the effect of diminishing returns when comparing offensive rebounds valuation versus defensive rebound valuation.
An interesting side note is, how does the value of a rebound change by position and player? I wonder how often Kidd grabs a rebound and dishes out an assist for a layup or dunk.
Mike H:
On shooting efficiency, I don’t see any evidence of diminishing returns. In fact, the more efficient production a team gets from a position, the MORE efficient the other four positions are. I first looked at a “Win Score” model of shooting efficiency, which is Points – FGA – .5*FTA (essentially, points above/below average due to efficiency). For each additional efficiency point generated by a position, the team gains 1.5 points. Looking at this as an efficiency rate — Pts/(FGA + .5*FTA) — I find the same relationship: an improvement in efficiency at one position is associated with an improvement about 13% as large at the other positions. 13% may not sound like a lot, but since there are four other positions, it means the total gain at the team level is about 50% higher than the gain at the first position.
Now, this is only one season (06-07), and there are factors that could explain this other than an actual player-interaction effect, including:
* Some teams value shooting efficiency and select multiple players with this skill; or
* Rich/smart teams get more efficient shooters.
However, if efficient shooters do actually have a positive impact on the efficiency of their teammates — such as by attracting more attention from defenders — that would certainly be a very important thing to know. I’m not saying that’s the case — I don’t know — but seems like a question worth researching.
On Kidd: my comment really wasn’t intended to be about Kidd specifically, but about the general pattern that players considered to be “great rebounders” do not appear to actually increase team rebounds by the amount they should according to WP. Here is a comparison of the rebounds per game ostensibly added (above position average) by a few players often cited here, and their teams’ total rebounds above average. I’ve looked at seasons when the players played a lot of minutes, posted good rebound totals (except Curry), and played only for one team.
Player/Years/Self/Team
Wallace (1999/00-date) +3.1 +0.3
Garnett (1999/00-date) +3.2 +0.6
Camby (2003/04-date) +2.8 +0.0
Marion (2000/01-date) +1.1 -1.0
Kidd (1997/98-date) +3.0 +0.0
Rodman (1991/92-1997/98) +6.6 +2.3
Curry (career) -1.1 +0.2
[Example: Wallace in 2002-03 was +414 rebounds over an average forward given his MP, which should have added 414/82 = +5.0 Reb/game for Detroit (which was –0.4 Reb/Gm as a team.]
Again and again, we see that the teams are merely average despite the contributions of the “great rebounders.” Rodman is an exception — his teams have generally been well above average on rebounding — yet still the team advantage is only about 1/3 of what Rodman supposedly was generating himself.
I think these numbers are pretty revealing. As I dug through the data, I was surprised to learn how little teams vary in their rebounding ability. The range is generally very narrow, which means the actual range in player rebounding ability is much narrower than what rebound totals seem to suggest (and also much narrower than the range of scoring talent).
“Also, if a typical offensive rebound leads to a higher percentage shot than a typical defensive rebound then that may mitigate the effect of diminishing returns when comparing offensive rebounds valuation versus defensive rebound valuation.”
Interesting thought. You should look at Eli’s research (link above), and his links to Ed Kupfer’s work, which seems to show that the diminishing returns effect is huge for DRBs (maybe as much as 90%), but less for ORBs (perhaps 50%?).
Guy
First, I’m enjoying our discussion.
Second, back to the discussion:
I wasn’t really getting at shooting efficiency. I had read the diminishing returns post some time ago and noticed that diminishing returns seems to affect scoring almost exactly the same as offensive rebounding. [Am I going crazy? I just checked the post again and I can't find the scoring data now. Has it been changed?]
Those rebounds are totals though aren’t they? What about the rebound rate? If a team improves its offensive efficiency after adding player X, yet still maintains essentially the same total team offensive rebounds as before, then their team offensive rebounding has improved.
the general pattern that players considered to be great rebounders do not appear to actually increase team rebounds by the amount they should according to WP.
Is this really what WP is saying though? To me it’s saying that the value of a rebound contributes to a win approximately the same value as a point scored. Which is just another way of saying that rebounds and points improve offensive efficiency by approximately the same amount.
It would be nice to see some data on the value of offensive and defensive rebounds from the standpoint of how many points they typically generate and compare this to how many points a typical shot attempt generates. This should result in an apples-to-apples comparison (unless there’s something subtle that I’m overlooking). Or perhaps how well rebounds correlate with offensive and defensive efficiency can be examined. These are all factors that may mitigate the diminishing returns. I’m sure that dberri or someone else has probably already done this, I’ll have to check.
Also, referencing Rosenbaum’s paper it looks like NBAEFF, PER, WS, and AWS (WS with reduced DRB value) all perform similarly as far as how much of the variation in future wins that they predict (in the limited year to year examples presented). In particular AWS and WS are closely related which seems to backup dberri’s position that reducing the rebounding value doesn’t produce significant changes (AWS, PER, and NBAEFF are better when predicting year 1:1, only NBAEFF is better when predicting year 1:3 – which just begs for the inclusion of the missing years 1:4+ since NBAEFF and WS are both trending such that they seem to be the best over the long).
“Those rebounds are totals though aren’t they? What about the rebound rate?”
It makes a difference to look at reb%, but not a big difference. Each extra player rebound yields about .2 team rebounds; each extra point of player reb% yields a .25% (quarter point) increase at the team level.
“Is this really what WP is saying though?”
Yes, it is. Dave B was explicit about this in his recent interview: “One last point… it is assumed that a player statistics are his statistics. In other words, if you score 10 points (or get 10 rebounds), you get credit for these points (or rebounds). If you take 5 shots, you are charged the cost of these shots. It’s important to note that there is some interaction between the player statistics, but I have found this effect to be rather small. For the most part, what you see in the NBA is what you get. So it makes sense to credit a player for his statistics (by the way, I would not make the same argument for NFL players).”
This isn’t that complicated. The question at hand is “how do NBA teams win (and lose) games?” If rebounding is a big part of the answer, then we should generally see two things: 1) winning teams having a large rebound advantage over losing teams, and 2) large differences among teams in rebounding rates (compared to other statistics). However, we don’t see either pattern. Rebounding explains relatively little of the variation in team wins, yet it explains a lot of the variation in players’ WP.
“Also, referencing Rosenbaum’s paper it looks like NBAEFF, PER, WS, and AWS (WS with reduced DRB value) all perform similarly as far as how much of the variation in future wins that they predict”
I think you’re misreading the data a bit. If you look at the summary table (table 7), you’ll see that WP performs notably worse than both AWS and MP in predicting future wins, a slightly worse than PER. The AWS contrast is most relevant re: rebounds, since AWS uses radically different values for rebounds. (All of these correlations look pretty good because there is relatively little change in team composition. So for example, if WP gives a rebounder credit for what is really the ability of the team to reduce opponents’ FG%, the yr-to-yr correlations will still look good. This means that small differences in correlation matter. And certainly if you can make better predictions just using MP, that’s bad news for any “advanced” metric. )
“It would be nice to see some data on the value of offensive and defensive rebounds from the standpoint of how many points they typically generate and compare this to how many points a typical shot attempt generates. This should result in an apples-to-apples comparison”
We could see that extra rebounds above average at the player level don’t rise the average team level, since probably some players are stealing opportunities to others, or just they are spot where opportunities are in more quantity. It’s possible that extra points by some players do the same effect. How may points a typical shot attempt generates? the Lgave. PPS, that lead us to Lg.aveFG% or TS%. How may rebounds a typical deff. rebounding attempt generates? The Lg.DRperAtt., that lead us to Lg.aveDR%. Yes, the R% is a higher number than the scoring%, but scoring produce extra points, you know, less efficient action, more points generated.
Then the stat “rebounds per rebouning attempt” is that would lead us to an apples to apples comparison.
And Mike, I don’t try to despice or underrate rebounds. By 4Factors-101, we know that rebounding as a whole (at both ends of the floor) has the same influence on wins than scoring: about 20% both actions, and we also know that the defensive part against scoring, that is defending the shot, which has another 20%, is already included in the stat named “deff. rebounds”, a shortcut created by regression that is neither good, nor bad. But, please, let’s rate the rebounding action in WPII, and let’s use AWS.
dberri’s quote was the assumption driving the development of the model wasn’t it? If you look at what the WP equation is saying based upon its relative valuation of rebounds and points then it is saying that they both contribute equally to wins produced. Of course this assumes that the equation largely means what it says in a naive sense, and is not in fact measuring some other non-obvious factor in a black box manner.
NBA teams win and lose based on offensive and defensive efficiency to a high degree. I’m really interested in how rebounding affects efficiency (wins) relative to other factors in the equation that are generally not pointed out (at WOW) as being incorrectly valued.
Table 7 is looking at the entire picture and certainly with their data over the time periods analyzed WS performs the worst (although similarly to the best, AWS). It looks like AWS explains 60% of the variation and WS explains 56%. Whether or not that’s significant enough from a non-statistical point of view to require a change in the model is in the eye of the beholder. It still has to be noted that we need more mid to long term data especially since the correlation rankings in the year-to-year tables show that WS has a smaller (absolute) slope than two of the other measures including AWS. Also it would be very helpful to see the splits when dealing only with teams that have undergone major personnel changes and those whose lineups have remained consistent. I think that the models that best predict efficiency after major roster changes and over a longer time period (say the average player’s prime years) will be the ones that are best at identifying a player’s individual abilities, whichever these turn out to be.
I tend to ignore the non-per-minute based approaches since they don’t tell you who should be playing more minutes – and injuries, investment, and such have a clearer impact.
“I’m really interested in how rebounding affects efficiency (wins) relative to other factors in the equation that are generally not pointed out (at WOW) as being incorrectly valued.”
According to 4 factors:
-To force missed shots (40%)
-To obtain the possession after the missed shots (20%)
-To force turnovers (25%)
-To avoid shooting fouls (15%)
In a regression DRebs take the missed shots weight too, then this action tend to be biased to bigs (not too far from reality). He probably won’t need to include oppFGMissed in the WPII’s team defense adjust, but oppOR and oppDR are obligued to be included.
Sometimes neither I understand my english. I should say: “In a win regression from boxscore stats, DRebs take the weight of forcing missed shots too, sice it doesn’t exist a boxscore stat to account for missed shots forced, then the action of forcing misses tend to be biased to rebounder bigs (which is not an accurate defensive approach, but not too far from reality, since about a 60% of FGA are attempted inside the painted zone, although the efficiency is higher here too)”
And I am not too convinced to give the benefit of distance to perimeter defenders.
Harold A, why do you hate rebounds so much?
I don’t hate rebounds, by the way I would prefer an average unidimensional rebounder than an average unidimensional scorer, probably the first one could give me more defense game, and a little more “wins” or less “loses”, since I will limit the FGA of both. But when you introduce in a rating a R%=100% for everybody (which is not the reality), then rebounding appears out of scale with respect to others action of the game, and total rebounds take the command of the rating, and will produce such strange things like that Camby is playing two times better than Yao. Camby could make up his bad offensive (and some teammates’s bad defense) with his defense and a pair of more rebounds, but two times better than Yao?
nothing against you harold, but you always seem to bring an end to great conversations by butting in the middle of an argument with a post that is usually off topic and hard to understand. than the people before discussing just end.
That’s because you are focused on diminishing return (players competing against teammates for rebounds), and the weight of a rebound from a team win regression (How to apportion the shot defense). And all of that is good for a dominical debate, but that doesn’t carry to anything because the rebounding problem is not there. The first thing is maked up just rating against the position LgaveR%, did I say “%”, which is the lack of the rating; the second one, you just will need to reinvent the boxscore, to probably find in the end that the best rebounders and blockers defend more shots.
all i am saying is, instead of just coming and ending this debate, let them discuss their issues. there was a productive conversation here, until you joined. and of course you bring up the same points as always that people have already addressed or are just done talkin about
I am not sure the conversation was that productive :)
Pingback: Meet Carl Landry « The Wages of Wins Journal
Pingback: Greg Oden or Kevin Durant? « The Wages of Wins Journal