Data Crunchers Look To Quantify "Chemistry" In NBA

http://offthedribble.blogs.nytimes.com/2012/02/28/n-b-a-gets-a-nice-look-from-the-data-crunchers/?action=click&contentCollection=Pro%20Basketball&module=RelatedCoverage&region=Marginalia&pgtype=article


Data Crunchers Look to Quantify Chemistry in N.B.A.
By JOSHUA BRUSTEIN  FEBRUARY 28, 2012 1:05 PM


Deron Williams and Chris Paul seemed set to leave their teams, and were caught up in the annual frenzy of speculation about where they would fit best.

According to a statistical analysis of 4,718 N.B.A. games that looked to determine how well teams played together, both the New Orleans Hornets, who employed Paul, and the Utah Jazz, who employed Williams, would have been better if the teams had simply traded their star players for each other.

Or at least that was the conclusion reached in a paper called “N.B.A. Chemistry” that will be presented at the M.I.T. Sloan Sports Analytics Conference later this week. About 2 percent of potential trades in which teams exchanged starting players would improve both teams, the paper found.

The Sloan conference has become the premier event for those who like to combine their rebounds with regression analyses to find new insights into sports. “N.B.A. Chemistry” is a finalist in the conference’s research contest. And while the prototypical stat geek is a baseball fan, more than half of the papers to be presented at the conference deal with the N.B.A.

“I think basketball is sort of the next wave, in the sense that it’s not as picked over as baseball, so there’s still a lot of work to be done,” said Eugene Shen, a co-author of “N.B.A. Chemistry.” “On the other hand, it’s not as difficult as football, because football is so team-oriented.”

A paper titled “Effort vs. Concentration” looked at how players respond to pressure, looking at free throws (a skill that requires quiet concentration) and offensive rebounds (where effort is the main factor). Players for the home team shoot better than those on the road, the paper found. But that changes when the game is on the line. Suddenly, players at home fare significantly worse, losing as many as 2 percentage points from their free-throw percentage in situations of “moderate pressure.”

The authors of the paper cite psychological research showing that people have trouble performing tasks requiring concentration if they are forced to think about what they are doing, and the tense silence at the end of the game does just that. Visiting players are not much affected by pressure situations, though, suggesting that perhaps fans would do better to put down the wands and give opposing players some time to think late in tight games.



A graph showing how home teams tend to grab more offensive rebounds in pressure situations.


By contrast, home teams grab a greater percentage of offensive rebounds as each potential rebound becomes more important. This makes sense, the paper argues, because people tend to try harder when they receive a little friendly encouragement.

Many statisticians are looking to quantify team chemistry, the seemingly indefinable quality shared by teams whose play on the court seems simply to flow.

A paper called “Experience and Winning in the National Basketball Association” assumed that teams would build chemistry as they played together longer, and looked to see whether stable teams were more successful in the playoffs. It found that each year of shared experience between two players counted for 0.06 playoff wins. This does not seem like much, but it adds up quickly for teams who maintain a similar lineup from year to year. The paper found that teams composed of long-term teammates win an average of 1.62 playoff games more than less familiar teams.

Another approach to chemistry is to look at which types of players mesh well together. “Big 2’s and Big 3’s” puts N.B.A. players in 14 different clusters, from the “limited, role-playing centers” to “wing 3-point shooters” to “role-playing big men without an exceptional skill.” It then looked at teams by which clusters their best players fell into. Teams whose best players were similar were generally less successful, with the exception of teams that used two high-scoring shooting guards, who consistently played better than expected. Unsurprisingly, small forwards that can shoot 3-pointers were the most compatible with different types of players.

Shen and his colleagues went even further, and looked at which skills go well together. It found that certain types of skills are enhanced when there are many people on the team who excel at one aspect of the game. Players who force turnovers tend to compound their success. So do offensive rebounders, while skilled defensive rebounders tend to crowd one another out. Productive scorers, unsurprisingly, did not mesh well together.

The research calculated that good chemistry can be responsible for up to six wins over the course of a regular season.

It also looked at some of the major player moves in the last several seasons, finding, for instance, that both the Denver Nuggets and the Knicks improved with their trade centered on Carmelo Anthony. But Denver benefited more.

As for LeBron James’s Decision, Shen said that Miami was one of the worst teams James could have picked in terms of chemistry. The Cleveland Cavaliers presented James with the best synergies, but that was largely because he made his teammates better, rather than the other way around. Also, according to their model, Dwyane Wade is not a good fit with James because their main skills, offensive scoring and ball handling, are wasted in combination with one another.

But Shen acknowledged that mathematical formulas do not always have the answers.

“If you’re LeBron James, synergies aren’t the only things you consider,” he said.




bob
MY NOTE: If this topic is of particular interest to you, go to the link and read the article. There are several hot links to MIT Sloan papers that are referenced in the article (e.g. "Effort vs. Concentration: The Asymmetric Impact of Pressure on NBA Performance") and you can read the academic papers directly. I went through this one, it's pretty darn geeky, but you can't claim they talk down to you. Of course, there's the "Safe Harbor" disclaimer at the very end of the article. Stats aren't everything. Yep. In fact, the only thing that is everything is everything itself. Just another piece, to be applied judiciously and proportionally.


.

"Lies, damned lies, and statistics" is a phrase describing the persuasive power of numbers, particularly the use of statistics to bolster weak arguments. It is also sometimes colloquially used to doubt statistics used to prove an opponent's point.

The term was popularised in the United States by Mark Twain (among others), who attributed it to the 19th-century British Prime Minister Benjamin Disraeli (1804–1881): "There are three kinds of lies: lies, damned lies, and statistics." However, the phrase is not found in any of Disraeli's works and the earliest known appearances were years after his death. Other coiners have therefore been proposed, and the phrase is often attributed to Twain himself.

Next thing you know, someone will try and quantify the laws of attraction.  I like blondes, my height or taller.  My wife is blonde, my height, and taller in heels.  I wonder how that works out statistically?

NYCelt wrote:Next thing you know, someone will try and quantify the laws of attraction.  I like blondes, my height or taller.  My wife is blonde, my height, and taller in heels.  I wonder how that works out statistically?

In my expert statistical analysis, I'd say that works out pretty well.

Is it possible that the phrase "lies, damned lies and statistics" was created by a person whose opinions were constantly refuted by statistics ?

swish

The "new age" basketball stats such as this one and the plethora of stuff issued "New" basketball statistics such as this one and those released by Synergy are interesting and, at a cursory glance, seem to offer greater validity than "pop" stats like PER. For one thing, they are not limited by dependence on traditionally record statistics (which minimize defense); and, for another thing, they select the components of the statistical formulas not judgmentally but through regression analysis.

Regression analysis shows the relationship between multiple variables and a "dependent variable" (which is the factor that's being acted upon by the multiple variables). For example, does the effect of having two types of players playing together under a certain amount of pressure increase or decrease the number of "playoff" wins and by how much?

But there are two caveats that involve subjectivity and can't be measured statistically. First, is the "dependent variable" (the phenomenon that is being acted upon) really measuring what it's purported to measure. For example, are "playoff wins" an appropriate dependent variable by which to represent "chemistry?" Does the fact that a combination of two or more skills or other factors increases the number of "playoff wins" really signify that chemistry is at work?

I've always thought of chemistry as an extra, indefinable dimension that helps more tangible ingredients (such as skills) to blend together productively. What was it that made the Russell Celtics, who obviously had many finite and individual skills, blend those skills in such a way that the players almost seemed to be reading one another's minds? Of all the many options on Red's plays, what made the guys on the floor (without exchanging any signals) segue into a particular option en masse? Was it some sort of collective sixth sense? Was it something as obvious as situational repetition? Whatever it was, I don't believe it could be quantified by statistics.

The second caveat is that, for the results of regression analysis to be interpreted as influencing a dependent variable, there must be a logical explanation (often a subjective one) for a causal relationship between the two. In Darrell Huff's tongue-in-cheek "How to Lie with Statistics," Huff mentions bizarre circumstances in which regression analysis suggests a strong relationship between variables when there is no logical explanation as to the cause of such a relationship. Something like (and I'm making this up, but it's no more weird than Huff's examples) a strong relationship between the African children's test scores and the number of cows and horses in Brazil (I threw that one in for Bob Heckler).

For several years, my company was charged with predicting (which shows that I am able to make predictions, but only for money) the quarterly sales volume for each of the Dunkin' Donuts shops in the U.S; and the predictions had to be for half a year in advance. We used regression analysis; and the number one chore was to identify variables that were truly related to Dunkin' Donuts future sales.

We tried hundreds of variables, and some initially seemed to be successful predictors but were not dependable over time. One main reason why they were undependable predictors was that there was no logic as to why they should predictably influence Dunkin' Donuts sales for extended periods of time, despite the fact that regression analysis initially indicated a strong statistical relationship between these factors and Dunkin' Donuts sales.

It wasn't until after more than a year of experimentation that we found three predictors that stood the tests of time and logic. The regression formula revealed not only the fact that the three factors collectively were good predictors over time, but it also indicated the degree of predictive influence exerted by each of the three factors.

I can't recall all three factors, but I know one of them involved the wholesale coffee price index. The logic behind the influence of that variable on Dunkin' Donuts future sales was that fluctuations in wholesale coffee prices at one point in time tended to influence Dunkin' Donuts retail prices about six months down the road.

As these "new" basketball statistics become more popular (which I expect they will), I'll be interested in monitoring (as best I can) to what extent they satisfy both of these caveats over time.

Sam

sam wrote:The "new age" basketball stats such as this one and the plethora of stuff issued "New" basketball statistics such as this one and those released by Synergy are interesting and, at a cursory glance, seem to offer greater validity than "pop" stats like PER.  For one thing, they are not limited by dependence on traditionally record statistics (which minimize defense); and, for another thing, they select the components of the statistical formulas not judgmentally but through regression analysis.

Regression analysis shows the relationship between multiple variables and a "dependent variable" (which is the factor that's being acted upon by the multiple variables).  For example, does the effect of having two types of players playing together under a certain amount of pressure increase or decrease the number of "playoff" wins and by how much?

But there are two caveats that involve subjectivity and can't be measured statistically.  First, is the "dependent variable" (the phenomenon that is being acted upon) really measuring what it's purported to measure.  For example, are "playoff wins" an appropriate dependent variable by which to represent "chemistry?"  Does the fact that a combination of two or more skills or other factors increases the number of "playoff wins" really signify that chemistry is at work?

I've always thought of chemistry as an extra, indefinable dimension that helps more tangible ingredients (such as skills) to blend together productively.  What was it that made the Russell Celtics, who obviously had many finite and individual skills, blend those skills in such a way that the players almost seemed to be reading one another's minds?  Of all the many options on Red's plays, what made the guys on the floor (without exchanging any signals) segue into a particular option en masse?  Was it some sort of collective sixth sense?  Was it something as obvious as situational repetition?  Whatever it was, I don't believe it could be quantified by statistics.

The second caveat is that, for the results of regression analysis to be interpreted as influencing a dependent variable, there must be a logical explanation (often a subjective one) for a causal relationship between the two.  In Darrell Huff's tongue-in-cheek "How to Lie with Statistics," Huff mentions bizarre circumstances in which regression analysis suggests a strong relationship between variables when there is no logical explanation as to the cause of such a relationship.  Something like (and I'm making this up, but it's no more weird than Huff's examples) a strong relationship between the African children's test scores and the number of cows and horses in Brazil (I threw that one in for Bob Heckler).

For several years, my company was charged with predicting (which shows that I am able to make predictions, but only for money) the quarterly sales volume for each of the Dunkin' Donuts shops in the U.S; and the predictions had to be for half a year in advance.  We used regression analysis; and the number one chore was to identify variables that were truly related to Dunkin' Donuts future sales.

We tried hundreds of variables, and some initially seemed to be successful predictors but were not dependable over time.  One main reason why they were undependable predictors was that there was no logic as to why they should predictably influence Dunkin' Donuts sales for extended periods of time, despite the fact that regression analysis initially indicated a strong statistical relationship between these factors and Dunkin' Donuts sales.

It wasn't until after more than a year of experimentation that we found three predictors that stood the tests of time and logic.  The regression formula revealed not only the fact that the three factors collectively were good predictors over time, but it also indicated the degree of predictive influence exerted by each of the three factors.

I can't recall all three factors, but I know one of them involved the wholesale coffee price index.  The logic behind the influence of that variable on Dunkin' Donuts future sales was that fluctuations in wholesale coffee prices at one point in time tended to influence Dunkin' Donuts retail prices about six months down the road.

As these "new" basketball statistics become more popular (which I expect they will), I'll be interested in monitoring (as best I can) to what extent they satisfy both of these caveats over time.

Sam

sam,


To be honest, my brains started to fall out about halfway through that academic treatise.  It had a lot of math and formulas (enough that I wasn't able to translate them into anything but "pink hearts, yellow moons, orange stars and green clovers").  Maybe you'll be able to decipher them and determine where their foibles lie.


bob
P.S.  If/when I go to Brazil it won't have anything to do with cows and horses.




It might not have anything to do with her either. I just love rainforests.






.

Bob,

Whoa, now I understand why you've been taking more pleasure excursions than usual. Actually, I can envision the two of you, dressed in leopard skins and swinging on vines in the rainforest.

Sam