The NBA season ended last week and now we have (thanks to Basketball-Reference.com) all the data needed to measure each player’s Wins Produced (and WP48, Win Score, etc…) for the 2008-09 season. With data set in hand, there are numerous stories to be told. While most people start with the best (i.e. MVP, All-NBA, Rookie of the Year, etc…), I thought I would start with a story that is often overlooked. Specifically, who are the least productive players – in terms of Wins Produced — in the NBA?
Before we get to the list, let’s define what it means to be “unproductive”. A productive player will tend to shoot efficiently, grab rebounds, gets steals, and avoids turnovers. So an unproductive player is one that tends to shoot inefficiently, fails to rebound and get steals, and is prone to commit turnovers.
When we pair productive players with unproductive performers we see that the latter can minimize the success of the former. In other words, a productive player can get a rebound or steal, only to see the unproductive miss the shot or commit a turnover. Or the productive player can miss a shot, only to see the unproductive fail to get the rebound. In sum, just as a team cannot succeed without productive players, it can also find failure when it employs too many unproductive performers.
The Least Productive Lead the Way
So who are the unproductive performers? During this past season 445 players were employed by NBA teams. So the bottom 10% would consist of 45 names. And topping this list – as Table One reveals – is Al Thornton of the LA Clippers. Consequently Thornton has been named the Least Productive Player – or LP2 – for 2008-09.
Table One: The Bottom 10% in 2008-09
Obviously Thornton and the Clippers struggled this year. But as we scan down the list we can that the Clippers’ problems are not all about Thornton. In addition to Thornton, the LA Clippers employed Ricky Davis (ranked 7th), Tim Thomas (ranked 11th and a Clipper for part of the year), and Steve Novak (ranked 31st). So although the Clippers also employed Marcus Camby (the 13th most productive player in the game), the abundance of unproductive players did much to limit this team’s success.
A similar story could be told about the Toronto Raptors. Toronto employs both Jose Calderon (20th most productive player) and Chris Bosh (26th most productive player). But the Raptors also have Andrea Bargnani (5th least productive), Jason Kapono (6th least productive), Roko Ukic (27th least productive), and Jake Voskuhl (28th least productive). When Toronto fans seek to understand why this season was so unsuccessful, this quartet – that played more than 25% of the team’s minutes – should certainly come to mind.
Three other teams also employed at least three players listed in Table One. The Wizards received -4.47 Wins Produced from Nick Young (10th least productive), Darius Songaila (23rd least productive), and DeShawn Stevenson (33rd least productive). Meanwhile the Minnesota Timberwolves were led to 58 losses by Bobby Brown (3rd least productive and a Minnesota employee for part of the year), Jason Collins (12th least productive), Brian Cardinal (24th least productive), and Sebastian Telfair (25th least productive).
Bringing Down Boston
And then there are the Celtics. The Boston Celtics won 62 games this past season. When we look at Wins Produced we see the team received 48.3 Wins Produced from Rajon Rondo (6th most productive), Paul Pierce (29th most productive), Kevin Garnett (30th most productive), and Ray Allen (35th most productive). But the team also received -4.44 Wins Produced from Glen Davis (9th least productive), Brian Scalabrine (20th least productive), and Stephon Marbury (41st least productive).
Unfortunately for the Celtics, KG is probably lost for the entire playoffs. And perhaps more dismaying to Boston fans is that the team has turned to Glen Davis to take many of Garnett’s minutes. Had Davis played all the minutes played by Garnett this year, Davis would have seen his Wins Produced fall from -1.98 to -4.12. Plus the Celtics would have lost the 10.16 Wins Produced from Garnett. The outcome of these changes is that the Celtics would have seen their Wins Produced fall from 61.06 to 48.76 (assuming the per-minute performance from Davis did not change). In other words, moving from Garnett to Davis takes the Celtics from a title contender to a team that will struggle to defeat the Bulls.
We only heard that KG was lost for the playoffs the day the TrueHoop picks were due. And my schedule last Thursday prevented me from doing much thinking about this before I had to submit my choices. As a consequence, I thought the Celtics would still be good enough to defeat the Bulls in five games. Although this is still possible, it seems unlikely. Yes, the Celtics did win on Monday night. And yes, Davis actually played well. But given what Davis did in 2008-09, it doesn’t seem possible for Davis and the Celtics to take the next three from Chicago.
And this means that once Chicago wins another game in this series, my rooting interest in this series moves from the Celtics to the Bulls. Everyone in the Smackdown picked Boston, but only Neal Paine and I took Boston in five games. So to stay even with the other competitors, once “Boston in Five” can’t happen anymore I need Chicago to take the series (and then everyone loses the points).
If Boston keeps playing Glen Davis, it seems possible that the defending champions will indeed lose in the first round. And the legend of Derrick Rose – which I will address in a future post – will certainly grow.
– 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.
Wages of Wins Journal 
mrparker
April 21, 2009
I like this approach. I often think of teams as being led to 60 wins but rarely think of teams as being led to 60 losses.
I still have questions about derivatives with this formula. It makes sense that Glen Davis cost Boston some wins. On the other hand a team consisting of all negative players wouldn’t produce negative wins. In that respect I don’t think its possible to derive expecting wins by simply adding and subtracting when it comes to a negative player and a positive player. I wonder if there is a tweak that would give the most productive players less credit while giving the least productive players no less than 0 wins.
Tball
April 21, 2009
A problem with Glen Davis is the team decided they needed a low post player to take 10-15 foot shots when Garnett was not on the floor, to space the floor for their offense. In a contest with Powe and Perk, Davis ‘won’. He is taking only about 30% of his shots from near the basket, where last season that number was closer to 60% (these are from memory, not sure where I read them). This offensive style leads to a lower FG% and fewer rebounds. I’m not defending Davis, his play does not hurt any less by understanding it, but I think the coaching staff should have scrapped the plan long ago.
Also of note, he has improved from -0.99 after 33 games and -0.85 at the midway point to -0.58. If he’d played all season for KG, perhaps he’d have improved to 0.00.
Meanwhile, perhaps Scal’s injury helps counter KG’s. And Powe’s injury last night means more time for Davis.
brgulker
April 21, 2009
while giving the least productive players no less than 0 wins.
But that doesn’t make sense. Such an approach would fail to adequately explain the negative consequences of unproductive players, wouldn’t it?
Oren
April 21, 2009
“On the other hand a team consisting of all negative players wouldn’t produce negative wins.”
True, but then again on such a team players would perform differently and would likely be slightly more effective.
In that regard, it’s on the same vein as saying there’s something wrong with WP because a team primarily playing five good bigmen wouldn’t be as effective as their numbers would indicate.
stephanie
April 21, 2009
In the TNT halftime segment of the Philly-Orlando game they showed a compilation where Dwight would establish extremely favorable paint position, including sealing his man under the basket and no one on his team could get him the ball, either because they didn’t see the obvious play or because they weren’t good enough. Stuff like that just vanishes into the ether.
I’m guessing Orlando will continue to struggle without Jameer and a less than normal Hedo. Great for Cleveland/LA fans I guess.
mrparker
April 21, 2009
Oren,
I’m not saying anything is wrong with wins produced. I’m saying if you want to use it to predict the impact of gaining or losing a (-) player its not nearly as accurate as when you replace positives with positives. In that respect I think something can be thought up so that wp will help us understand the impact of gaining and losing (-) players. Remember when an average Jason Richardson was supposed to help Charlotte gain 10 games or something like that.
Again, I’m not saying that wins produced doesn’t do what it was orginally designed to do. I’m saying it can help us better in another fashion and maybe some derivative can be found.
I’ve found a way to predict future wp48 of college players when they make the transition and this is sort of what I’m alluding to. I use a different method but it correlates with wp48 pretty well. I’m trying to start a brain storm here but noone seems to care that negative players aren’t really negative players but just extremely unproductive.
brgulker
April 21, 2009
On the other hand a team consisting of all negative players wouldn’t produce negative wins
What I could have said better:
A team full of unproductive players wouldn’t win negative games, obviously. But, what the negative WP48 captures is the net negative impact that unproductive players have on teams.
The way I understand the negative WP48 is quite simply that X player cost his team wins over the duration of the season. Simply making that number 0 would not quantify the net negative impact.
brgulker
April 21, 2009
I’m trying to start a brain storm here but noone seems to care that negative players aren’t really negative players but just extremely unproductive.
Isn’t a ‘loss’ really a negative win, though?
Ryan
April 21, 2009
I like how you say, “if” they keep playing Glenn Davis, as if they have a choice. They don’t!!!! They have Kendrick Perkins, Big Baby, and that’s pretty much it. Leon Powe tore his ACL on Monday night, so he’s done. Mikki Moore can’t understand the defense (and is pretty terrible at other aspects of the game as well) and Scalabrine (Mr. 20th least productive player this year) is just now being cleared to play after 3 concussions and being inactive for 2 months. So, don’t talk about it being a choice, it’s a necessity. If they don’t play Glenn Davis, they have ZERO chance of advancing. Put that in your wages o’wins and smoke it.
Peter
April 21, 2009
There’s more troubling news for Boston:
If Rajon Rondo’s ankle sprain lingers, the team might have to turn to Stephon Marbury more, which damages more productivity.
That could make Celtics-Bulls from a five or six game Boston win to seven, or even to a Chicago upset.
In the long run, this scenario benefits Cleveland and Orlando the most because the Celtics’ losses increase the chances that both teams will meet in the Eastern Conference finals.
brgulker
April 21, 2009
@ Ryan, this post was written before the extent of Power’s injury was known. Take a chill pill.
@ Peter: I think this series is over, unless they can ride Pierce, Allen, and Rondo 43+ minutes per game, and assuming they can all be very productive during those minutes.
But frankly, I think Chicago’s big men will simply overpower the Celts inside. Thomas, Noah, and Miller should be able to outplay the Celts big men
Also, you might see Chicago put a concerted effort on pushing the tempo, to wear down a very thin rotation.
Peter
April 21, 2009
Addendum: My last statement came out garbled; the Cavaliers would not be able to face the C’s until the conference finals.
However, it’s not a good sign for the Celtics that Noah has exceeded his season averages in both games thus far.
mrparker
April 21, 2009
I give up.
brgulker
April 21, 2009
You haven’t even made an argument, mrparker. Someone questions your point, and you resort to playing the victim?
I’d honestly like to hear what you have to say, but if you’re going to bail on every conversation in which you get challenged, you won’t get very far.
ChicagoismynewBlog
April 21, 2009
Congrats on the success of your blog! When you get a chance, take a look at my very own WordPress blog. Good luck in the future also.
http://chicagoismynewblog.wordpress.com/
Clarence
April 21, 2009
Dave,
as a raptor fan it makes me a little sad to see Bargnani ranked so low yet again. This year he’s played great (particularly since J O’neal went out) and raptor fans have high hopes of him now for the future. But apparently he’s still ranked ridiculously low by your formula? Why is that? Does the WoW formula really put that much emphasis on rebounds? Bargnani’s percentages are great all over the floor, he gets a block a game – not great numbers, but I’m still surprised it’s deserving of 5th least productive).
Jason E
April 22, 2009
Bargnani is a poor rebounder. He’s very, very poor for a big man and really nothing better than average for a SF. Since he’s rarely played with two guys who could legitimately be considered ‘bigs’ to allow him to really be a SF and not have that hurt the team because of a dearth of rebounds from the front court. Rebounds really are that important and Bargnani really did hurt Toronto’s rebounding.
I’m not the biggest fan of plus-minus, but plus-minus does support Dave’s conclusion vis a vis Bargnani. For the minutes Bargnani was on the bench, Toronto actually very mildly outscored their opponents. While in games, the Raptors were heavily outscored. Is it all entirely his fault? Perhaps not, but that no other Raptor, when removed, saw the team head into positive plus-minus space should be a warning flag and perhaps good indication that Dave’s ranking is not ‘ridiculously low’ but a rather accurate portrayal of Bargnani’s contributions (or lack thereof) to the Raptors. I don’t see how anyone can realistically say he played great unless playing great and winning are not in any way related.
Michael
April 22, 2009
Clarence, Bargnani is a forward who averages 5 rebounds and 1 assist per game, whilst shooting 45% from the field. Those numbers for a forward are very bad. Yes he gets a block and half a steal a game, but he turns the ball over about 1 and a half times a game which cancels that out. He has improved but he is still very bad, and the fact that he plays so many minutes (nearly 2500 this year) means that he is almost destined to be near top of the profs least productive list!
Mr Parker, if you think of ‘negative wins’ as ‘losses’ then you avoid that problem. I.E player x produced negative two wins for team 1 = player x lost the team 1 two games. Then if you take the losses produced by the unproductive player and add them to the total team 1’s replacement player at his position would have produced in those minutes, you get the number of wins that player x has actually cost the team overall.
Golden Graham
April 22, 2009
I’d be interested in reading what Bargnani’s WP48 is as a starter vs. reserve.
17.7 ppg, 5.9 rbg, 1.4 apg in 34 minutes.
45.7 % from the field, 42% from behind the line.
“Clarence, Bargnani is a forward who averages 5 rebounds and 1 assist per game, whilst shooting 45% from the field. Those numbers for a forward are very bad.”
Bargnani’s TS% and EF% are .559 and .512
Bosh’s are .569 and .492 .
That’s for the entire year, including his dismal play as a reserve. Yes, his FG% is low for a big man (45%) , but he takes a lot of 3s, and consequently he’s an efficient offensive option. Unless you’re saying Bosh isn’t. Or Nowitzki.
Nowitzki: .564 TS% .498 EFG%
I’ve been a huge AB detractor, but he’s made huge steps forward as a starter, at least offensively. He’s still a rebounding liability, but he’s made strides forward as a defender, and his offensive game, as aforementioned, is much more efficient than it used to be.
Golden Graham
April 22, 2009
Oh, and here’s a link to his splits for you guys to go over: http://www.basketball-reference.com/fc/psplit.cgi?player=bargnan01&year=2009
It’s very interesting how his increase in minutes and the absence of JO from the lineup have correlated so closely. But then D Berri has always maintained that more minutes generally equals increased production.
Michael
April 22, 2009
Yes his EFG and TS% are inflated by his 3 point shooting, but his Adjusted FG% and points per shot sits right around the average. You know that comparing Bargnani’s shooting to Bosh or Dirk Nowizki is not really the issue. Bargnani is a bad player largely because he is a poor rebounder and worse than average at getting steals, which means his net possessions score is way below average.
His effective 3 point shooting in no way makes up for this. Again he has improved, but he is still bad.
ok
April 22, 2009
Since this is all about the numbers, have a quick look-see at Bargnani’s numbers since Jan 1, vs. Aldridge’s. Is Aldridge a ‘bad player’ as well?
Part of the problem (for Bargnani) is that the stats used to derive WoW’s are taken from the entire season; it’s no secret he played like utter, total, complete garbage for the first couple months, but it’s also true that since Jan., he’s been much more efficient. His low rebounding numbers will always hurt him in these statistical analyses, that’s just the way it is. He’ll improve, but won’t ever be anything more than average at best. He’s an offensive-oriented player with improving defensive skills who will always be deemed ‘unproductive’ by these measures because of his rebounding deficiencies.
I’m more curious to see what gems are at the top of the board for ‘most productive’ players.
Phil
April 22, 2009
ok,
Aldridge is a more productive player than Bargnani according to WP, though still below average. Aldridge better on the boards; otherwise they’re quite similar.
ok
April 22, 2009
Phil – no argument, but that’s where teammates come into play, no? I mean, it can’t hurt LA that he has Vanilla Gorilla or Oden playing in the frontcourt with him (both defensive-oriented, excellent rebounders), and with wings like Roy, Outlaw, Batum & Fernandez creating space/shots for themselves & others. Other than Bosh, who do opposing teams have to worry about stopping on the Raptors? Part of the reason Bargnani (& Bosh, to a lesser degree) have been exposed as less than stellar defenders (& in Bargnani’s case, rebounder) is the pathetic ‘D’efense on display by the Raps’ guards & wings for most of the season. If no one on the team (prior to Marion’s arrival) could prevent dribble penetration in the slightest (& no one could), it puts the front-court defenders in a continual state of help-&-rotate…and out of rebounding position.
mrparker
April 22, 2009
brgulker,
Victim? Come on man take a joke or else I’m getting Al sharpton. I’ll argue this till I’m blue in the face because its something that has f’d a couple of season win bets for me that I miscalculated. I guess noone else has tried to do that.
Oren
April 22, 2009
“Isn’t a ‘loss’ really a negative win, though?”
“Mr Parker, if you think of ‘negative wins’ as ‘losses’ then you avoid that problem. I.E player x produced negative two wins for team 1 = player x lost the team 1 two games. ”
I don’t think this works. After all, a team that has an average of a 0 WS would be projected to go 0-82. A team with a WS of a negative win score would be projected to win negative games.
For example, a team of Al Thorntons won’t win -8 games and lose 90.
“I’m saying if you want to use it to predict the impact of gaining or losing a (-) player its not nearly as accurate as when you replace positives with positives. ”
I suppose that it does make sense that if a guy plays only one game, he can’t cause his team to lose a game and a half.
Kevin
April 22, 2009
To change the topic some, coaches have little say in creating a roster (with a few exceptions), they do have a say in allocating minutes. Perhaps a better way to judge coaching ability is to see how well they do this, rather than winning %, since as a few here have noted, sometimes the coaches have no good options (the Glen Davis point). For instance, Sam Mitchell did well to recognize Jamario Moon’s overall worth, despite his being deferential on offense, last year, though this year he was traded as though he were worthless, and now the Heat have decided to employ James Jones at SF for 30+ minutes(!).
On that thought, did Don Nelson throw this season by limiting the play of his productive youngsters (he admitted early the Warriors were not contenders, and he does seem to be an expediency-minded dude) and playing Captain Jack for 40+, or did he simply have poor options/fail to recognize options? Did he throw the season, or was he just a bad coach (or did he have no options)?
Rick Adelman is regarded as a good coach, and yet he has failed to control Ron Artest’s shooting (which would dramatically improve his productivity if he NEVER SHOT A 2-pt-j again), and has decided to start Aaron Brooks over Kyle Lowry. Is there a justifiable reason for this?
Perhaps the value of coaching is buried under the vastly more important question of player talent. But coaches do make impactful decisions, like allocating minutes and player positions (imagine a slower moving Warriors team which employs a proper frontcourt).
Looking forward to your next book, prof Berri.
Nick
April 22, 2009
Oren, a team would not go -8 & 90 (as projected), but that team would probably lose by a larger margin than the 0-82 (projected) team would.
Remember these are not absolute numbers. If this team were to play in another non-NBA (lesser) league, this difference would be more noticeable. For example a team of 12th men from small NCAA Division I schools, would probably project to 0-82 (just guessing). But a team of amateur 12th men from poor high school teams would project worse. Then 12th mean from NCAA would probably beat the High School team. But either way, both teams would go 0-82. This is what this data would be capturing. That this team has even of a less chance to win the odd game.
It’s all relative.
pattymc
April 22, 2009
I would contend that the bulls might be a perfect opponent for Big Baby. He is playing against a former teammate in college who Baby has clear coordination and strength advantages (but def not explosiveness). The other main big he contends with is Noah, who is also less strong/coordinated than Big Baby. Davis has looked pretty good in 1 on 1 situations at the elbow against Thomas.
I’d say it’s better for him not to go into the post against those guys because he wouldn’t be able to get his shot off. But, an open jumper, or a drive after those guys fly at him out of control is right in his wheel house (as he’s shown so far this series).
Brad Miller, however, has made Davis and the rest of the Celtics post players look bad. He is the only bulls player that has exposed Davis for what he is on D – too short to stop jumpers/passes from a skilled big, and not saavy enough to stop the back cuts from bigs when there is more than one decent passer for the bulls on the floor (sorry everyone but Miller, Rose and Heinrich).
Michael
April 22, 2009
“I don’t think this works. After all, a team that has an average of a 0 WS would be projected to go 0-82. A team with a WS of a negative win score would be projected to win negative games.”
Oren, I understand your point from a theoretical standpoint, but in practise how exactly does a team “win negative games?” If a team consisted entirely of negative WP players, the absolute worst they could do is lose every game. Negative wins is a nonsensical concept in absolute terms. However if applied to a minority of poor individual players on a given team it can show who is costing that team games. Like I said if player X produces -2 wins he has cost that team two wins if his replacement is as bad as he is. If however his replacement would have produced 10 wins in those minutes, then playing player x instead has actually cost the team a total of 12 wins.
Jason E
April 22, 2009
It happens fairly regularly that those attempting to ‘improve’ the model to fit some preconceptions of how it should behave forget how models are made and what they are for.
WP was not designed from observations of teams employing and using large numbers of players with negative wp48 generating ‘negative wins’. It was created based on observations of NBA basketball, where the very, very worst teams do not go without some victories in the season.
Models may not perform well when asked to evaluate systems outside of the realm of what the initially observed. Saying that a team of Al Thortons would not win negative games (or even would lose by larger margins) is not a flaw in the model.
The problem with recentering the scores such that negative scores were not possible is that this may not accurately reflect what the effect of removing a player from a lineup will be. If adding an Al Thorton really does result in 8 fewer wins over the course of a season *given otherwise a team comprised of NBA players* and not some attempt to field an all-time worst team, then the score is accurate. A zero would not accurately reflect the real impact.
mrparker
April 22, 2009
Jason E,
I’m not suggesting that wp be recentered. I’m suggesting some sort of derivative that helps predict future impact of gaining or losing negative wp players. WP is obviously fine the way it was constructed.
Shayan
April 22, 2009
Statistics saying Andrea Bargnani is the fifth least productive player is a classic example of stats misleading in a huge way. If you’ve actually seen him play in the games this year its absolutely ridiculous to come to that conclusion.
Jason E
April 22, 2009
Ridiculous? Is it ridiculous to expect that a guy who is playing well actually helps his team improve? Bargnani did not help his team when in games compared to when he was on the bench.
“Strangely”, the statistics correctly identified the results that occurred when Bargnani played. Were he valuable, one would expect he wouldn’t have made the Raptors worse when he played, but he did in the only measure that matters: how efficiently you score points relative to how efficiently your opponent plays. Your incredulity towards the statistics based on what you saw doesn’t provide a more accurate picture given how poorly the team performed with him playing.
I suspect that “wathing him play” is misleading. His play did not result in good things happening for the Raptors. Whatever subjective opinion of how he ‘looked’ really doesn’t matter.
Michael
April 22, 2009
“Statistics saying Andrea Bargnani is the fifth least productive player is a classic example of stats misleading in a huge way. If you’ve actually seen him play in the games this year its absolutely ridiculous to come to that conclusion.”
Shayan, if you look more closely at the data that is presented, Bargnani is actually 39th out of the 45 players listed per 48 minutes. The reason he is listed 5th overall is because with the exception of Al Thornton and Al Harrington he has played many more minutes than most of the other players on the list. So no, he isn’t the 5th worst really, he has just had such a negative impact due to the large number of minutes he was given this year. I stated above the reasons for why he really isn’t an effective forward, however as a Raptors fan you will be aware that he is still very young and can therefore obviously get better!
Oren
April 22, 2009
“For example a team of 12th men from small NCAA Division I schools, would probably project to 0-82 (just guessing). But a team of amateur 12th men from poor high school teams would project worse. Then 12th mean from NCAA would probably beat the High School team. But either way, both teams would go 0-82. This is what this data would be capturing.”
I agree that this is what the data would be capturing. And surely, both of these teams WP would be negative. However, I don’t understand why this would be the case.
Ultimately, if you played an infinite amount of games eventually both teams of 12th men would beat the NBA players. The only caveat to this claim is that it may take more then 82 games.
In that event, this team of 12th men would go 0-82 but should have a positive WP(something between 0 and 1 game) because while they wouldn’t win one game a year, they win one every ten(hundred) years or so. Instead of having a negative WP(because all of their players would likely be the worst at their respective positions).
BTW, I doubt that this has a major impact on the predictive accuracy of the model. As Jason mentioned, there are relatively few players with a negative WP. I suspect that if one does consider a negative wins produced to count as a “negative win”, that it will have little impact.
Italian Stallion
April 22, 2009
Bargnani is always going to look horrible on this model because he doesn’t rebound well for a big man and he’s mostly an outside shooter, so his efficiency is lower than the typical big man.
The first problem with this analysis is that IMHO this model overrates low usage efficiency and underrates high usage scorers. It’s almost the complete opposite problem of PER which overrates inefficient high scoring.
The second problem is that this model considers Bargnani as a Center/PF when he really has the skill set of a SF and plays that way.
If your role is to score, shoot from the outside, open the floor, take some of the tougher shots etc…. you are automatically going to have a lower efficiency and NOT be in a position to get as many rebounds. That’s Bargnani!
The problem Toronto has with Bargnani right now is that they have this really huge SF type player and as a result need to fill out the squad with smaller players that do some of the things that most big guys do. That’s not so easy to do.
They need a very efficient lower usage role player than can rebound exceptionally well (perhaps a David Lee type).
They would give them the proper mix on the court.
However, if they ever get the right mix, a guy like Bargnani will be extremely valuable (even though he will continue to rate poorly here). His size will allow him to be way more effective than the typical smaller SF and the team will lose little in terms of rebounding and efficient scoring. He can get his shot almost any time he wants because he can shoot over people and draw other big men away from the basket.
If you think of him in terms of a SF instead of a C/PF, what I am suggesting will make some sense to everyone that is not a total slave to their own numbers and models.
Italian Stallion
April 22, 2009
I’m surprised more emphasis wasn’t put on Marbury’s role in Boston’s decline.
Did anyone really think that Boston was immune to the curse of Marbury?
I made it way better than even money there would either be a meltdown in the locker room, a tragedy, a series of injuries, or some other unforeseen disaster soon after Marbury joined the team. LOL
Seriously, he’s playing terribly and it is amazing how this kind of thing follows him everywhere he goes. ;-)
Jason E
April 22, 2009
When you say a model “over-rates” or “under-rates” what do you mean, IS? The model only ‘over-rates’ rebounders if the rebounders do not actually provide the benefit that the model says they do. It does not under-rate high usage low efficiency scorers if they do not actually provide a boost to their teams’ record. Is this the case or does it simply not conform with your gut feeling of how things should be.
It does not appear that the model has problems accurately evaluating Bargnani’s contributions (or lack thereof) to the Raptors.
If it were possible to play Bargnani as a SF, he would come out better in the rankings. Is it though? Does he really allow a team to use a pair of rebounders down low? It doesn’t seem like Toronto has used many lineups with a couple of guys who really can be thought of as ‘bigs’ in addition to Bargnani. Why?
It still sounds as if you have some notion of “fairness” that has to be taken into consideration. The model is not about being fair, but about accurately predicting results. Criticizing a model without indicating where the *results* are faulty is not terribly meaningful.
Golden Graham
April 22, 2009
The model accurately predicts results, but does it accurately reflect a player’s contributions to those results? I have faith in its predictions based on the sum, but occasionally I question its ability to accurately measure its parts.
Shayan
April 22, 2009
Jason E,
I see you are sticking to heart with the statistics, and sure, the numbers may be what they are, and I can understand Ukic, Kapono, and Voskuhl being put there, but to put Bargnani in there is one thing calling him the fifth least productive player really doesn’t do it any justice. Yes, he’s a bad rebounder for his size, and has screwed up plenty but he plays like a forward and is more perimeter oriented, and has progressed a lot this season. So maybe stats wise, he was the 5th least productive player.
But saying “Is it ridiculous to expect that a guy who is playing well actually helps his team improve? Bargnani did not help his team when in games compared to when he was on the bench.”
YES, that is ABSOLUTELY ridiculous because he helps this team greatly when he’s on the court. You must not have actually watched many Raptors games this season because anyone who has would know that.
Sam Cohen
April 22, 2009
Jason E- could you give a fuller explanation of this statement?
“If adding an Al Thorton really does result in 8 fewer wins over the course of a season *given otherwise a team comprised of NBA players* and not some attempt to field an all-time worst team, then the score is accurate.”
I understand your general point about not asking a model to evaluate things that are beyond its scope, but I’m curious as to why asking it to evaluate a team of bad NBA players would be considered beyond the scope of the model. After all, bad NBA players are still NBA players.
Just looking at the “top” 25 people on the list, Sebastion Telfair (PG- # 25), Antoine Wright (SG/SF- # 2), Al Thornton (SG/SF- # 1), Al Harrington (PF- # 8), and Andrea Bargnani (PF/C- # 5) all started more than 41 games this season, so it doesn’t seem completely outlandish that the five could all play on a (very bad) team together.
Why should this line-up be considered outside the scope of the model?
(The last time I took an econometrics class was five years ago, so feel free to point out something blindingly obvious if I’m missing it.)
dustin
April 22, 2009
Bargnani is bad according to PER, Wins Produced, and adjusted +- . I’m not sure how you can say your eyes are better than 3 systems that come to the same conclusion.
dustin
April 22, 2009
Also, there is no excuse for bad rebounding on the defensive end. He has a 16.4 defensive rebounding rate to go with his horrible offensive rebounding rate (3.4).
Shayan
April 22, 2009
Ok guys, I see your points. Bargnani is statistically horrible in every way, there’s no point in me arguing it because the numbers are what they are – all I’m saying is I don’t think the Raptors are better with him on the bench – guy is part of our nucleus and for good reason; he’s got a hell of a game and unreal potential – by no means anywhere near a complete player of course – he’s had a lot of ups and downs and is still developing. He had a brutal first half of the season and then really came along in the second half, and is only going to get better. But if you guys believe the Raps are better off with him not playing, then that’s your opinion and I respect it, I just think its the complete opposite but that’s what makes sports so fun.
dustin
April 22, 2009
Shayan,
That is provably false. From
http://www.82games.com/0809/08TOR14.HTM
The raptors are only .5 points/possessions better on the offensive end when bargnani is on the court. However, they also give up 6.9 more points/possession defensively while he is on the court. (Compared to when he is on the bench, in both cases)
Whether or not this is all his fault is another matter, but it is provably false that the raptors play better when he is on the court.
mrparker
April 22, 2009
Though you do have to take those +/- numbers with a grain of salt. By those numbers Chris Paul was a mortal last season.
Jason E
April 22, 2009
It was the *adjusted* plus minus that showed Paul to be not particularly great. I know the adjustment is supposed to control for quality of teammates, but it seems at times to detract from just looking at the on-off net. Paul was +10 in that regard. His very limited off court minutes and the large number of minutes he shared with other starters allowed a small, small sample to skew the “adjustment” from that figure.
82games is not presenting the “adjusted figures” in the table linked above.
On the team full of bad players:
The model was never designed with observations of a team with that many poor players, so how it would behave under those conditions isn’t tested. My guess is that the ‘diminishing returns’ witnessed at the top end at times when teams are loaded with great players would apply to those truly terrible teams as well, where whole wouldn’t be quite as bad as the sum of the parts. In the same way that adding a bunch of scorers presents problems because there’s only one ball and not everyone can shoot, the guys who are suffering because they miss too many shots will not miss as many when there’s other guys to share the futility with. Someday I fully expect the Clippers or Warriors to test this though.
Phil
April 23, 2009
Imagine a player that shoots 90% from the field and is otherwise average. He is a boon to his team. But if another player does the same and averages an additional five steals, he is an even greater boon. Agreed?
Now imagine the opposite; a player that shoots 10% from the field but is otherwise average. He is detrimental to his team. But a player that shoots 10% and also turns the ball over an additonal five times is even worse. Agreed?
Then why is it so difficult to grasp the concept of negative wins? We differentiate between the effects of the first two players that I listed, and for obvious reasons. Negative wins allows us to differentiate between the second pair as well, and for the exact same reasons.
I think it’s pretty implicit that point differential, WP, etc are capped at 82 losses (or wins), and typically have a small margin of error; about three games if I recall correctly. Furthermore, the sample size for teams with fewer than 10 wins is one. Ditto fewer than 10 losses. Perhaps at the extreme end of the bell curve in terms of wins/losses, such systems lose a lot of accuracy. This would not be too surprising; such is often the case with extreme outliers in a statistical model.
Also, what Jason E said.
dustin
April 23, 2009
Yet another metric that says Andrea Bargnani is not good!
http://www.basketball-reference.com/blog/?p=2191
Mark
April 23, 2009
If I understand WP correctly, WP actually predicts team Efficiency Differential and then Efficiency Differential is used to predict number of wins. If so, the problem under discussion is with using Efficiency Differential to predict wins, not with using WP to predict efficiency differential.
A team with a -50 Efficiency Differential is going to lose every game. So is a team with a -100 Efficiency Differential. But the second team is clearly much worse and will have a lower total Wins Produced because Wins Produced is based on Efficiency Differential.
mrparker
April 23, 2009
Mark,
Great way of wording your argument. My question then becomes at what point is a player bad enough for his contribution to become 0? And then the follow up question is should there be differing degrees of being a 0?
brgulker
April 23, 2009
Honest question:
Isn’t this debate largely theoretical and not at all practical?
In other words, there are no teams in the NBA who are bad enough to have a negative net WP (correct acronym?), and isn’t it a good question to ask if there ever will be?
So yes, in theory, a team with a negative net WP should have a negative record (based on this model) — theoretically.
But doesn’t the reality that 1) No team will ever lose 0 games and 2) No GM will ever construct a team that will ‘produce negative wins’ sort of make the argument nothing more than a theoretical debate?
Again, I’m not as well-versed in the metrics as many here are, and I’d welcome correction on this one.
mrparker
April 23, 2009
btw,
I’ve thought about it and my hypothesis is that (-) players are actually zeroes and there lack of ability can inflate the wp48 of their capable teammates. Its possible that their negative is spread out as a positive to their positive teammates. I guess I’d have to be “unlazy” enough to go back and look at some of the worst teams to find out if this holds true
AdamB
April 23, 2009
Negative Wins = Losses is wrong interpretation of the WP model.
The basic concept is that an average team will score exactly 0.5 win in a game and 41 wins in season as well. In the very same time, 0.5 losses/game will be produced. Here you can see, LP48 can be defined as WP48. For a team that wins every game, sum(WP48) = 1, sum(LP48) = 0. Per definition, games played per 48 minutes is 1, and wins + losses = games played.
That brings sum(WP48) + sum(LP48) = 1, for a team.
In that equation 5 players are summed up, so we divide by 5 to get it for one:
WP48 + LP48 = 0.2
It means that player X with WP48 will produce 0.2 – WP48 losses per 48 minute. (Not minus WP48.)
I think the problem of the model is maybe that WP48 is linear function of difference of the acutal player’s statistical parameters from the league averages. In an improved wins produced model, extremly unproductive players’ WP48 should converge to 0, and on the other hand, extremly productive players’ LP48 to 0. That statement would limit both quantities between 0 and 0.2. The player rankings would not change, though.
The reason the WP model is so powerful in giving back teams’ wins total, that it works very well around 0.1 WP48, and if you average players’ WP48 on a team (weighted with playing time, obviously), that will give around 0.1 WP48, but never negative or bigger than 0.2.
The model I suggested above would lost the simple addtitivity because of the nonlinearity, therefore maybe wouldn’t give such good win totals; but it would be interesting in individual perspective.
Michael
April 23, 2009
Professor, would I be right to assume your next post will focus more towards the top 10% of players in the league this year? (now you have all your season data.) I’m especially interested in whether CP3 or Lebron finished on top!
Oren
April 23, 2009
Suppose I have a team filled with players with less then a .00244(1/410) WP score. This team would be projected to win fewer then 1 game.
If these players had lower then a .00122 score, they’d be projected to win less then half a game.
Lower then a .00061 score, a quarter of a game.
I would think that if you wanted to make sense of a negative WP score, you’d have to convert it from a negative number into a decimal on this scale. However, I suspect that while doing this may clarify something that is confusing, it wouldn’t significantly improve the predictive power of the model.
Mark
April 23, 2009
mrparker:
If we are talking about a player’s contribution to Efficiency Differential, then you need to consider the range of possible player contributions. This range can vary from a large negative number to a large positive number. A player who contributes zero to efficiency differential is average. A player who contributes a negative efficiency differential is below average. With efficiency differential, there is no problem of an artificial cutoff at zero.
Berri takes the analysis a step beyond efficiency differential because it is more intuitive to think of a team as a 65 win team than a +10 efficiency differential team.
mrparker
April 23, 2009
I completely understand the model. The model is not in question. I’m only worried about predicting the change in wins when a negative player is replaced by a positive player(Jason Richardson replacing Adam Morrison for example). The change in wins in that type of situation is almost always overestimated(trust me I gamble). As opposed to say the Iverson vs. Billups or Miller swaps where all player involved were positive and the change in wins was easy to predict.
P-Dawg
April 23, 2009
I dunno. I’ve been a big fan of your analysis, but I don’t think it adds up with Baby. I compated his numbers to Pierce’s, who is in the top 10% according to you, multiplying by 1.74 (to equalize in terms of minutes on the court). If Baby played the same number as minutes as Pierce, he’d come out with 7rpg (40% more than Pierce), 1.6 apg (about 40% of Pierce), 1.6 TO (about 43% less than Pierce) and 12 ppg (about 40% less than Pierce). Additionally, Baby’s shooting percentage is just a bit lower than Pierce’s. I’m not trying to argue that Davis is as good as Pierce, but just going on the criteria you’ve identified and stacking up guys in the top and bottom ten percent of productivity, Davis doesn’t seem all that bad. (Oh, and he’s shooting 50%, pulling down 6 boards, scoring 22 ppg with only .5 TO/g so far in the series. How can he be a main reason the Celtics are struggling?
brgulker
April 23, 2009
P-Dawg:
You’re forgetting that Pierce is a SF and Davis is a PF.
Hence, you’re not adjusting for position, which is important for Dberri.
jbrett
April 23, 2009
If I’m Jason Collins, I’m very unhappy with the way this list is organized. “I should be number one,” I can imagine him saying. “All I need is enough opportunity, and I could be less productive than anyone.” (Melvin Ely could perhaps use a small sample size to argue his way into the conversation.)