Last November I posted the “handcrafted” Wins Produced numbers for 2009-10. As I noted last fall…
….these numbers are slightly different from the numbers Andres Alvarez posts throughout the season. The numbers from Andres are referred to as “automated Wins Produced”. This is partially because the position adjustment Andres employs is derived from an algorithm that considers such factors as the position designations listed on-line, a player’s height and weight, and the height, weight, and position designations of his teammates. In general, this algorithm is good enough to tell us if a player was a center, power forward, small forward, shooting guard, or point guard. But sometimes it might place a player at a position “incorrectly.”
The approach I have taken could be called “hands-crafted” Wins Produced. Essentially, I go through each roster, and assign positions by considering height, weight, position designations (i.e. same factors as Andres) and also my understanding of what position the player is probably playing. This process is fairly tedious (hence the inability to provide updated numbers throughout a season). In general, Andres and I reach the same conclusion for most players (so the automated approach – since it is easier – is preferred). Sometimes, though, there are differences.
It is also possible that a person looking at the lists Andres and I offer would disagree. If that is the case, I have presented the ADJ P48 numbers (you can look here for what the means) and the position averages. This will allow one to calculate their own WP48 numbers for each player.
Now that the 2010-11 regular season is complete, I can now post the “handcrafted” numbers for this past season (yes, I can do this almost seven months earlier than last year).
The following two websites report these numbers. The first takes you to a ranking of all 453 players, starting with Kevin Love and ending with Andrea Bargnani (more on Bargnani below):
Table One: Ranking All Players by Wins Produced in 2010-11
And this table allows you to look at each team.
Table Two: Ranking All Players on Each Team by Wins Produced in 2010-11
As noted, you may not like how positions were assigned. So if you wish to see what a player would look like at a different position, here are the ADJ P48 position averages for 2010-11.
- Center: 0.425
- Power Forward: 0.377
- Small Forward: 0.272
- Shooting Guard: 0.229
- Point Guard: 0.272
To illustrate how these averages can be used, consider a question I was asked by Kevin Clark of the Wall Street Journal yesterday. Kevin noted — in It’s Carmelo Anthony, Plus the Not-So-Fantastic Four — that the Knicks closed out its last playoff game with the following five players on the court:
- Center: Jared Jeffries [ADJ P48: 0.241]
- Power Forward: Carmelo Anthony [ADJ P48: 0.364]
- Small Forward: Bill Walker [ADJ P48: 0.244]
- Shooting Guard: Roger Mason [ADJ P48: 0.170]
- Point Guard: Toney Douglas [ADJ P48: 0.271]
Kevin asked the following question: Imagine these five players played all the minutes for the Knicks in a regular season (and the added time didn’t impact per-minute performance). How many games would these five be expected to win?
Here was my answer (from Kevin’s artice):
Basketball analyst David Berri, an associate professor of economics at Southern Utah University, said the Knicks’ lineup in crunch time Tuesday night was far from impressive. According to his projections, that lineup would have produced just 17 wins over an 82-game regular season.
Yes, that isn’t a great collection of talent on the court for the Knicks.
One last note on these numbers. Brendan Nyhan pointed me to the following story (from Eric Freeman at Yahoo! Sports) on Andrea Bargnani yesterday.
Raptors GM thinks Bargnani can become a good rebounder
The article notes that Bryan Colangelo – the Raptors General Manager and person responsible for bringing Bargnani to Toronto – still thinks Bargnani can get better on the boards (yes, hope can spring eternal). But as Freeman notes, that seems unlikely at this point.
Freeman also notes that Bargnani plays like a 7-foot small forward. Let me close by noting that even if Bargnani were a small forward, with an ADJ P48 mark of 0.191, Bargnani would still be below average (but not the least productive player in the game).
– DJ
Mike
April 21, 2011
By wins produced, Love was worth over 25 wins for a team that won 17 games.
In the 9 games that he missed, Minnesota was outscored by 11 points per game. That mark is not consistent with a team that would win 0 games over an 82-game season.
Can you explain the discrepancy?
dberri
April 21, 2011
Mike,
You are looking at nine games? Small sample would be one obvious explanation.
There is diminishing returns in the NBA as well. I have written about this often in the past. The FAQ has quite a bit on this topic.
JChan
April 21, 2011
Dave has said before that diminishing returns exist, they just don’t affect the overall picture very much. I’m guessing in this particular case, diminishing returns (or whatever the opposite is called) could explain it. If Love is not there to suck up every available rebound, certainly other Timberwolves players should get a few more than normal, which would mean that, on average, Love’s teammates probably are a bit more productive when he is out of the lineup. So the team would still be terrible, just not quite as terrible as you might expect. That’s my guess, anyway. Also, small sample size could have something to do with it as well.
Italian Stallion
April 21, 2011
Mike,
Some of the games Love missed late in the season gave Anthony Randolph a chance to play big minutes. AR didn’t produce a lot of wins for the season (he rode the Knicks bench) or in limited minutes for the T-Wolves, but he had a few huge games in that period while Love was out. On that roster, he was probably a big improvement over some of the other options.
Mike
April 21, 2011
A team that gets outscored by 11 points a game, as Minnesota was in the 9 games that Love missed, would be expected to win about 15 or 16 games over a full season (I’m using the pythag wins formula from basketball-reference).
How badly would a team have to be outscored to get 0 wins produced over a full season?
Mike
April 21, 2011
@Italian Stallion
Randolph has a .094 WP48 for the season, and Love’s is .474. Love earns his team nearly 5 times more wins per minute than Randolph. Yet when Love was out and Randolph was starting, Minnesota was outscored by 11 points a game instead of the 6.1 points per game that they were outscored by when Love starts.
That’s only a 4.9 point difference. That might be attributed to the sample size of 9 games, but 82games.com has the Wolves being 4.7 points better with Love on the floor than when he’s off it. That obviously is a big difference, but I can’t see how it translates to 25 wins, especially considering that Love’s replacements aren’t good.
IS, do you think that Love was worth over 25 wins to the Wolves this season?
arturogalletti
April 21, 2011
Mike,
The answer is about 15 points.
http://arturogalletti.wordpress.com/2010/07/25/the-worst-team-since-the-merger/
The 93 mavs got outscored by 15.2 points a game.
And they still won 11 games.
Curious Guy
April 21, 2011
“Now that the 2010-11 regular season is complete, I can now post the ‘handcrafted’ numbers for this past season (yes, I can do this almost seven months earlier than last year).”
And it’s very much appreciated! ;-)
“Essentially, I go through each roster, and assign positions by considering height, weight, position designations (i.e. same factors as Andres) and also my understanding of what position the player is probably playing.”
Have you tried using data on 82games.com? They have “Player Floor Time Stats by Position” for everyone with minutes assigned to positions and IMHO it would be easier and faster than considering all those factors above.
nerdnumbers
April 21, 2011
Mike,
I’m in the process of writing up a post on the Wolves. I do actually want to talk about science though! In ideal science you take a bunch of variables, change one at a time and then see what happens. That’s experimental science.
Now when we have the NBA we get to observe what happens and make judgement. That’s cool but the issue is many people treat this as experimental science. When Kevin Love got injured that was a variable. Awesome right? How will the Wolves play without K-Love vs. with him?
The issue is that Love leaving the game isn’t the only variable that changed: their opponents changed, they had other roster changes (Darko was actually out 5 of the 9 games Love was as well), young inexperienced players (who are typically inconsistent) got more minutes, etc. While it’s a fun game to take an event as a single variable it’s hardly ever the right move.
Power Overwhelming
April 21, 2011
Does Kevin Love have the highest wp48 of all time among players who’ve played at least 2000 minutes? Is his season the best, or among the best, seasons ever?
Power Overwhelming
April 21, 2011
I just checked Magic and MJ’s seasons. Magic had several seasons above .500 and Michael had one season above .500. So Love’s season isn’t the all-time best, but it is in very good company, according to Wins Produced.
Pau Gasol
April 21, 2011
Why am I listed as a strict center (position 5.00) when I’ve started over half my games at PF?
Arthur
April 21, 2011
These numbers are hard for a warriors fan (although there a few numbers that ever look good for them). It looks like they had the worst best player in terms of WP48 with Steph at 0.163. Granted a few other other teams best player did not get as many minutes so his wp is higher.
My question is what happened to David Lee? Is the kind of issue that Love would face if he switched teams? It is not a question of diminishing returns from his teammates, because they aren’t that good.
Surely different playing styles (ie keep Lee close to the basket for rebounds) have to help numbers somehow. Or are the warriors as usually very unlucky?
arturogalletti
April 21, 2011
Arthur,
Player Month Minutes Played Month Wins Per 48 Minutes Month
David Lee 1 276.73 0.18234709
David Lee 2 454.34 0.046341494
David Lee 3 517.63 0.141674572
David Lee 4 447.49 0.054824394
David Lee 5 546.77 0.151272953
David Lee 6 391.14 0.319845274
That elbow injury killed his (and the warriors by extension) season.
nerdnumbers
April 21, 2011
We’ve got to nominate the best repeated arguments without context for this season.
David Lee isn’t good (leaving out his injury)
Kevin Love isn’t good (leaving out his team)
Derrick Rose is MVP (leaving out his FG% and Turnovers)
Others?
Italian Stallion
April 21, 2011
Mike,
You have to read my comments carefully. Randolph pretty much didn’t play at all for the Knicks and when he did he was horrible. He was more productive for the T-Wolves and especially productive in a handful of the games that Love didn’t play in.
I’m not sure what point differential would satisfy you that Love has a lot of value, but I think between diminishing returns (in reverse) and the play of AR a lot has been accounted for.
I tend to not think Love was worth 25 wins, but that’s because I try to build diminishing returns into my thinking on rebounds.
Italian Stallion
April 21, 2011
Arthur,
IMO, Lee’s injury impacted at least part of his season, but I think he also suffered from some diminishing returns on the boards and from playing PF instead of C.
He was the primary rebounder on the Knicks. With GS, he was often paired with another very good rebounder in Bierdrins. So they probably took a little from each other.
In addition, one of the reasons Lee was more effective on offense for D’Antoni is that he often played him at C. At C he gets a lot more open looks from mid range and is often quicker to the basket than his defender and can fish with either hand. Of course he gets killed on the defensive end also, but he puts up box score stats.
When Bierdrins got hurt late in the season, Lee played more C and surprise surprise, his boxscore numbers all went back up. Biedrins was no longer clogging the middle on offense and Lee was getting better looks (and I assume at least a few extra rebounds, but I didn’t check)
Alien Human Hybrid
April 21, 2011
@IS-
Didn’t Lee play major minutes with Zach Randolph?
Italian Stallion
April 22, 2011
Alien,
Lee played with Zach. I can’t recall the exact numbers, but I do recall debating the issue of his rebounding on another stats blog back then. I was arguing for some level of diminishing returns. I seem to recall they went down a little when he played with Zach, but I can’t swear to it.
He was a highly efficient scorer back then, but he was way more limited on offense. He didn’t run the pick and roll really well until D’Antoni became coach and added the jumper as a major weapon last year.
I watched a lot of GS basketball this year because I am a huge Lee/Curry fan. I don’t think they ran the P&R nearly as often or as effectively as the Knicks did.
Mike
April 22, 2011
@nerdnumbers
You’re right. 9 games is a small sample size. It seems off to me that Love is credited with 25 of Minnesota’s 17 wins. It seems off that Minnesota w/o Love outperformed WP48’s predictions. But this one case does not prove WP48 is wrong.
Except Love isn’t the only player with weird numbers. Noah has a .254 WP48, much higher than those of his backups Kurt Thomas (.076) and Omer Asik (.113), but Chicago has a better record (26-8, .765 win%) without Noah than with him (36-12, .750 win%). Why does Mike Miller have a .202WP48 when his team is 11.8 points worse with him on the court? http://www.82games.com/1011/1011MIA.HTM
Those are only 3 obvious examples from scanning the rankings. Something seems off.
Italian Stallion
April 22, 2011
Mike,
I think you should forget about adjusted +/-. I’ve seen so many absolutely preposterous examples over short periods of time I consider it almost irrelevant.
I also think that if you are going to look at a teams record/point differential with various key players out you have to adjust for home/away, strength of schedule, and whether it was an easy/difficult part if the schedule in terms of traveling, B2Bs etc…
Mike
April 22, 2011
I never mentioned adjusted +/-.
How do find the things you mentioned in the 2nd paragraph. I have no reason to believe the Bulls faced a particularly difficult portion of their schedule in the 48 games that Noah played, but I could be wrong.
A.S.
April 22, 2011
Much appreciated that Prof. Berri takes the time to do these numbers – and so quickly after the end of the season this year!
I think Pau Gasol has a legitimate beef, above, about being listed as a straight “5”, given that he played at LEAST 761 minutes with Andrew Bynum (also a straight “5”) on the floor. (According to 82games.com, the Lakers second most used 5-man unit was Fisher-Bryant-Artest-Gasol-Bynum, which unit played 761.4 minutes.)
I am also wondering about Kevin Durant’s position as “4.32”. That seems rather dubious to me. To be true, one would have to think some like that Durant spent a third of his time as a “5” and the other two-thirds of his time as a “4”? I think it is much more likely to say that Durant is a “3.32”, not a “4.32”. Last year, he was a straight “3”.
A.S.
April 22, 2011
Also, I wish I understood the relationship between Adj. P48 and WP48. Going to the link explaining the calculations, I see that WP48 = Relative Adj. P48 + 0.099. And I think Relative Adj. P48 = Adj. P48 – league average for the position.
But applying that formula to the numbers given for 2010-11 in the chart, I get league averages of:
C:.433
PF:.371
SF:.280
SG:.227
PG:.262
But these are different than the numbers set forth in the post. What am I doing wrong?
Jeff
April 22, 2011
I have never seen a stat more retarded that that.
Greyberger
April 22, 2011
“I think you should forget about adjusted +/-. I’ve seen so many absolutely preposterous examples over short periods of time I consider it almost irrelevant. ”
He’s not talking about adjusted +/-, he’s talking about plus minus. It’s a matter of record that Chicago outperforms opponents more with Gibson/Asik on court than Noah, or that Miami’s net point differential is 11.8 pts worse with Miller on than off.
What that means or if it means anything at all is a matter for discussion. But it’s not the results of a metric, it’s a measurement of what happened.
Nathan Walker
April 22, 2011
I hope “Wages of Wins II” involves a 10-page apology for overvaluing rebounds. This list makes me very sad :(
Itallion Stallion – Regularized adjusted+/- is way more predictive than any other public system. Of course +/- has poor results over small sample sizes, that’s why you look at the regularized, wait til the season is over, or just look at 2-year or 3-year data.
Seriously, other than the problem of diminishing returns (i.e. The Heatles) & styles of play (i.e. Melo to the Knicks), I honestly can’t think of a single reason to look at any other system than Regularized +/-.
Italian Stallion
April 23, 2011
Mike,
You mentioned Mike Miller’s value on and off the court. That’s why I brought up Adjusted +/-.
This year in mid season I started charting the point differential of all the teams adjusted for schedule strength/home and away and can quickly glance at each team’s schedule to see if it had a lot of tough games clustered, had a tough road trip, a lot of B2Bs etc.. I also make notes of injuries.
Italian Stallion
April 23, 2011
greyberger,
I thought it was understood that raw +/- numbers are so preposterous as a measure of anything that I elevated it to adjusted +/- which I consider to be only mildly less preposterous.
Italian Stallion
April 23, 2011
Nathan,
I wasn’t referring to straight plus/minus even though the original poster may have been. I was referring to adjusted +/-.
I have seen examples that were so beyond preposterous for 40 games or more that I can’t take any of it seriously (as least as a gambler).
It’s possible that over longer periods of time all the small sample and other technical issues get worked out, but by the time I wait 2-3 years, many of the very young players in the NBA are much better than they were in year 1 and many of the old players are worse. A lot of the rest of have suffered injuries and are not the same over the length of the sample.
Maybe some day all this stuff will be worked out, but right now I think the best approach is box score data because it’s fairly consistent over shorter periods of time. There’s room for more things to be tracked in the box score, a tweaking of the values, and astute game watching, but I think that’s the way to go for now.
Mike
April 23, 2011
I would be more likely to use regularized +/- if I understood how it worked.
@IS What strength of schedule did the Bulls face in the 48 games that Noah played compared to the 32 games that he did not play?
Italian Stallion
April 23, 2011
Mike,
I’d love to be able to answer that question, but the usefulness of that information didn’t occur to me until the middle of the season. So I didn’t start tracking it manually until March. It would be a time consuming project for me to create a program to do the calculations and retroactively figure that out for the full season.
I’ve been using an outside source. In March I asked them if they had historical data, but they didn’t and don’t plan on providing it.
I am contemplating creating something in the off season so I can do all this stuff on my own, but for now I’m in the same boat as you. I know the questions, but I don’t know the answers. Sorry.
Demarcus Cousins
April 23, 2011
I thought I was the best player in the rookie class? And John Wall wasn’t #### without me?
Philip
April 24, 2011
Nathan, IS,
I’ve never seen any iteration of +/- which has much predictive value compared to WP (or any box-score based metric, for that matter). Which one does?