A few days ago I was asked by Benjamin Polk – from A Wolf Among Wolves – to look into how the Timberwolves were allocating minutes. Ben’s story was posted at TrueHoop, and then a follow-up story was posted at A Wolf Among Wolves. The follow-up story noted that
- Kevin Love is immensely productive.
- and his teammates are not.
Love is primarily known as an amazing rebounder. And whenever a rebounder is ranked high with respect to Wins Produced, someone tends to argue “rebounds are over-valued by Wins Produced”.
In response to such a comment, I thought it would be a good idea to finally post some responses to Frequently Asked Questions and Frequently Offered Comments. This page – which is a work in progress – addresses such issues as why Wins Produced differs from popular sentiment, alternative measures of player performance in the NBA, and the issue of rebounds. As noted, this page is a work in progress. But hopefully what is posted helps.
One should note… as indicated at A Wolf Among Wolves, the model employed to study how minutes are allocated in the NBA needs to be updated. This update will be part of a paper I hope to present this summer at the Western Economic Association meetings. So look for more on this in the future.
– DJ
EntityAbyss
January 1, 2011
Well, I like the faq page. It’s not complete, but does a very good job explaining certain things. People are gonna start asking for stuff on usage and shot creation, so that maybe should be added.
ilikeflowers
January 1, 2011
professor,
Since Guy has been shown as willing to outpost all comers, there is some chance that what he says will be assumed to be true simply because it has such visibility and is spoken with such marked…um…
assuredness. In his last comment here he has some very concrete and disparaging opinions about the elasticity argument the gist of which is,
These elasticities tell us very little about how big a role rebounds play in determining players’ WP48. For that purpose, one uses standardized coefficients not elasticities.
Some of his post is irrelevant and some of it actually argues against the point that he’s attempting to make, but I can’t specifically address his elasticity claims. From looking up elasticity and standardized coefficients in wikipedia it appears that he is at a minimum oversimplifying and is likely just wrong. Can you clarify elasticity vs standardized coefficients as used in the FAQ?
dberri
January 1, 2011
ilikeflowers
When a person has been shown to be wrong over and over and over again (seriously, how long is the list of statements this person has gotten wrong?), why do people keep taking their arguments seriously?
Elasticity is a common approach in economics — taught in microeconomics around the world — to determine how responsive a given Y is to changes in a given X. In other words, it is a perfectly appropriate approach to this particular question.
By the way… I have been traveling this past week, but I think Andres sent me a response from you regarding apples and oranges. Not sure where that was posted, but it was very funny.
Adam C. Morrison
January 1, 2011
“When a person has been shown to be wrong over and over and over again (seriously, how long is the list of statements this person has gotten wrong?), why do people keep taking their arguments seriously?”
Because that’s what open-minded people do.
EvanZ
January 2, 2011
Every introductory book on regression analysis explains that measuring changes with respect to standardized coefficients (i.e. coefficients normalized with respect to 1 standard deviation change in that variable) is as useful or even more useful than simply looking at elasticities. The interpretation of the model often changes drastically.
To anyone actually interested in learning more about this and not simply deferring to authority, I would recommend one of the “little green books” published by Sage:
“Applied Regression: An Introduction (Quantitative Applications in the Social Sciences)” (1980) by Michael Lewis-Beck
It’s literally “little” (80 pages) and is even available as a Kindle e-book.
Flowers, why don’t you read that book and see if you feel the same way about Guy’s comments, especially since Guy has been banned from this blog and can’t respond on his own. Also, see my recent post on the “Four Factors” where I perform this kind of analysis, and explain why it’s necessary:
http://thecity2.com/2010/12/21/regressing-point-differential-on-the-four-factors-part-2/
Italian Stallion
January 2, 2011
I enjoyed the frequently asked questions section, but think you are going to continue get questions about both rebounding and scoring.
Some people believe that the value of rebounding is not being distributed properly by any model (count me among them). Based on decades of observation I would argue that you can’t simply assign a standard value to all defensive or offensive rebounds.
It seems to me that the value of a rebounds changes depending on who gets it.
When a guard gets a long rebound, there is typically no one else in the area. So he deserves full (or close to full ) value for that rebound. When the primary rebounder (usually the C) gets a defensive rebound, there are often 1-3 other defenders in the area that could easily have gotten it, but they defer to the primary rebounder. So perhaps the value should be distributed between them in some way.
When a PF or SF gets a defensive rebound, it’s sometimes in an area where the primary rebounder would not have gotten it but someone else might have.
It’s almost as if the general formula should be that the more significant rebounding is to your role on the team, the less value should be given to each rebound you get. The fewer you get, the more value each has.
Also, related to this, if you reduce the value of a rebound to the player who got it to below 1, you cannot simply redistribute the difference to the remaining players equally because not all of them were equally likely to get it.
On the scoring side, this model puts an extraordinary amount of the value on the efficiency and not on the scoring itself.
I don’t think you are ever going to convince most people that if one player (A) scores 30 points per 36 minutes at a slightly below average efficiency that his scoring is less valuable than a player that scores 10 points at a high efficiency (B).
It’s entirely dependent on the make up of the team.
Every teams need scorers, not all players have equal scoring talents, and not all teams have an equal or adequate number of scorers to work with.
If player A is on a team with a lot of great scorers, then he is clearly hurting the team by shooting so much at below average efficiency. It’s the role of the coach to make sure the shots are distributed better.
However, if player A is on a team with a lot of Bs, then he may be doing the best thing possible by taking some lower efficiency shots because they are better than the alternatives.
There is value in the ability to create and make shots at even reasonable levels of efficiency. Almost obviously, PER and NBA efficiency overstate that value, but IMHO the people that often say that this model overrates rebounding are not correct. It underrates scoring and that’s why the top of the chart is so dominated by rebounders.
entityabyss
January 2, 2011
Itallion stallion, the main argument I’ve been seeing is rebounding being overvalued by WP. Using many examples, it’s been shown that scoring efficiency is based on shot selection and not shot amount. There’s an article on this on both WoW and arturo’s blog. I don’t know how someone can see that and still argue.
As for rebounding, I’ve posted the effects that marcus camby and ben wallace’s rebounding has had on their teammates. The effects were very small AND even if you give them all the blame for diminishing and subract wins from them from the rebounds left that were expected, they’d still be very good.
ilikeflowers
January 2, 2011
Evanz,
from my own ‘research’ on elasticity vs standardized coefficients (I just googled it and read some of the results) it looks like elasticity is used because it can reveal effects that the standardized coefficients underestimate or just miss. I didn’t see any cases in which the reverse position was true (although this of course doesn’t mean that there aren’t any I may have just missed them). I did see where there were drawbacks to using the standardized coefficients (i.e. incorrect conclusions) that elasticity didn’t suffer from. Is there a specific reason to favor standardized coefficients over elasticity in the case of basketball stats other than it’s statistics 101, i.e. the simplest tool? What’s wrong with using a percentage (unitless) based measure? I actually did this research prior to asking the prof for more info.
I read your post and while it was interesting since it touches a little upon an area that I find interesting (being are there cases in which teams significantly and systematically deviate from the wp48 predictions?), it didn’t address the elasticity vs standardized coefficients argument (at least as far as I could tell).
ilikeflowers
January 2, 2011
Evanz,
from my own ‘research’ (prior to asking the prof for clarification) on elasticity vs standardized coefficients (I just googled it and read some of the results) it looks like elasticity is used because it can reveal effects that the standardized coefficients underestimate or just miss. I didn’t see any cases in which the reverse position was true (although this of course doesn’t mean that there aren’t any I may have just missed them). I did see where there were drawbacks to using the standardized coefficients (i.e. incorrect conclusions) that elasticity didn’t suffer from. Is there a specific reason to favor standardized coefficients over elasticity in the case of basketball stats other than it’s statistics 101, i.e. the simplest tool? What’s wrong with using a percentage (unitless) based measure? Why wouldn’t that be the preferred measure?
I read your post and while it was interesting since it touches a little upon an area that I find interesting (being are there cases in which teams significantly and systematically deviate from the wp48 predictions?), it didn’t address the elasticity vs standardized coefficients argument (at least as far as I could tell).
ilikeflowers
January 2, 2011
I’m having difficulty posting so if I’ve posted multiple times I apologize.
A.S.
January 3, 2011
The FAQ is pretty good, I thought. One thing I found interesting, though, was the discussion of the consistency of rebounding and how that explains in part why Wins Produced is a good measure. Yet, when you look at the chart that shows that you can cut the value of defensive rebounds in half and it will produce similar #s, Prof. Berri discusses 3 players in particular: Camby, David Lee and Troy Murphy. And 2 of those 3 players have had massive drop offs in production this year – Lee now has a WP below .100, while Murphy actually has a WP in the negative range. I know that the circumstances have changed for both players, but I found it interesting that these are among the players that Prof Berri is using to show the superiority of his model.
entityabyss
January 3, 2011
Well A.S., both of these players have suffered major injuries. David lee’s career almost ended this year and he has a hole in his elbow. It’s been causing trouble with his shooting and his rebounding has gone down from the beginning of the year.
nerdnumbers
January 3, 2011
A.S.
I’d say 2 players in a small period of time may not be the best proof. Additionally you have chosen two players that have been impacted by injury (also Murphy is older, Dr. Berri’s research has shown players start to decline after 30)
I am very happy to have put up day to day numbers but it certainly has helped some with overreacting. Just a point at the start of the season (around 10 games in), Carmelo Anthony was playing as a superstar, Kevin Durant was playing terribly and Lebron James was playing average. I could have stepped in at that point and said “How can we claim Melo isn’t as good Durant and LeBron? He’s playing better than both.” Just waiting a mere 20 games would make that claim silly.
However you do make a good point. Using one player as an example has the weakness that a bad patch or an injury can alter that player. Still career wise (don’t forget Camby, Murphy and Lee have all played multiple years well before this one) are good examples.
A.S.
January 3, 2011
Andres,
Oh, I totally agree that we’re talking about a relatively small sample (although getting toward the halfway point of the season) and two players that have had some injuries (although neither player has lost a lot of games – most of Murphy’s DNPs have been CD). I just thought it was striking that, in reading through the FAQ, the only players used as examples were the three I mentioned, two of whom have had major, major decreases in production this year.