Author Message
Crow
Joined: 20 Jan 2009
Posts: 829
PostPosted: Tue Mar 08, 2011 6:15 pm Post subject: Most important Heat star to take away Reply with quote
From recent Slate article:
Oliver stated with confidence that Dwayne Wade is the "most important guy to take away" on the Miami Heat—not LeBron James. "Not everybody knows that," he said—including many opposing coaches who appear to be keying on LeBron. So how did he know that? His computerized mathematical game-analysis tool, called Roboscout, told him. "It's not an obvious thing when you watch the game," he said. "But when you do the analysis, that comes out."
(Note: underline was added to highlight one of the issues to be discussed.)
Keeping with the focus on who is the "most important guy to take away", not “who is better”, here is some quick data that might be related to taking a guy away, at least it is stats down:
"Hold" Wade to 14 shots or less (about 25% less than average), Heat are 8-2. Hold James to the same, Heat are 10-1.
Hold to 40% raw FG% or less, Heat are 5-7. Hold James, Heat are 5-6.
Hold Wade to 4 or less assists, Heat are 21-9. Hold James, Heat are 4-6.
Hold Wade to less than 20 points, Heat at 8-8. Hold James, Heat at 8-1.
Hold Wade to under 15 on Game Score, Heat are 7-11. Hold James to the same, Heat are 8-4.
What else would you consider?
Free throw attempts? When Wade gets 6 or less, the Heat go 13-8. With James 18-5.
Force 5+ Wade turnovers, Heat go 4-11. James 19-11.
Holding Wade down goes along with the opponents winning more than with holding James down by the same standards in 6 of 7 cases. This brief look is consistent with what RoboScout is said to report.
But... how confident can you be in causality? Did teams especially try to take away Wade or James or both or neither or sort of or hard to say? How do you decide whether there was a major attempt to take a player away? Was the stats under-performance caused by such or did it just happen under normal defensive efforts?
With video you could count touches, ball denials, distance from the basket upon receiving the ball the desirability of locations reached with it, scoring possessions used / touches ratio, double-teams, tightness of man defense, quickness and frequency of help defense, etc. to make a case for or against efforts to take a guy away and this may becoming easier and more common soon with new tracking products out there but there was no previous indication that RoboScout was video-based and you would need this data for every team game and every opponent game and a solid way to rank it to know if the opponent's activity was normal or heightened.
And even with these data results, how confident can you be given the sample size for each is perhaps 15-20 games in size (with some overlap)? Is there enough to meet the traditional high standards of statistical significance in this case? If you can't, do you say you don't know with confidence and nobody "knows", or do you just do the best you can and make a recommendation (with either confidence or caution as you see fit) if the data "looks" to point one way?
You could take it down to lineup level and look at team performance when each was on the court just in games when either was "held down". But then whether the other was also held down or doing well could affect / complicate the data and it could affect the confidence level with conclusions. And the impact of Home, Away, Rest, SOS, the performance other teammates and the other team also are variables to consider along with other things like refereeing and coaching and luck and noise.
Can Roboscout make all these considerations and adjustments and give an answer with high confidence or "enough" confidence?
It is proprietary, so I don't know the depth of Roboscout's analysis or its degree of significance.
From the outside, you can either accept the statement of findings based on trust, or hold off from doing so since no explanation is given for how much it considers or how or the strength of findings and no public testing data is available (at least yet, though previous statements suggest it will stay private), or just think about it further on your own.
Was data from previous seasons used? Would you use data from other teams / other contexts for this question?
On their own Coaches will use observation and judgment to find their answer regarding is the "most important guy to take away" on the Miami Heat. The findings of quantitative analysis gives another perspective.
I don't know if it is easier to try to take away Wade and do it than with James, but when Wade is held down compared to his averages in the above data the opponents are usually more successful than when James is held down.
Maybe with the data, analysis and conclusion from the quantitative study Coaches will review the issue again and perhaps gain additional or better insights and then make whatever decision they are going to make.
This may be an example of limits to really "knowing" from either perspective. Often basketball decision-makers will have to make judgments from information that is not certain. Judgment from observation and lengthy experience is not likely to be perfect or even as strong as thought in some cases. Statistical significance for the quantitative analysis is nice... if you can get... but if not, then what? Managers will have to decide how to weight the available perspectives and proceed.
Still there may be a difference between what they do, what they really think they "know" and what they may say.
Last edited by Crow on Wed Mar 09, 2011 2:02 am; edited 1 time in total
Back to top
View user's profile Send private message
Jeff Fogle
Joined: 11 Jan 2011
Posts: 70
PostPosted: Wed Mar 09, 2011 12:44 am Post subject: Reply with quote
Was trying to think of anything to add crow, but you've covered so many possibilities. All I can say is that I try to re-read Taleb's "The Black Swan" every year (as opposed to Portman's, lol). That helps remind me of the limits that any model can have because the world is so complex...
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 829
PostPosted: Wed Mar 09, 2011 2:10 am Post subject: Reply with quote
Yes the NBA is pretty complex and pretty hard to model with high accuracy and stability and that is good to remember, often, while still trying.
For closer Heat observers, how often would you say teams have made trying to take away Wade or James the top priority, and how often have they made Bosh the priority, or prioritized the big 2 evenly or the big 3 evenly, or mostly focused on executing their own offense and defensive systems? How have they done with each strategy in your estimation?
For the past 5 game losses, Wade has fallen short by the above criteria about 5 times as much as James. There is some reason to lean toward the conclusion that trying to take Wade away over James would be the better choice. Even though it is small sample. I am not against interpreting data and making calls on data short of statistical significance. But I wanted to recognize the small sample and the limitations of knowing and the choice to make and present a strong conclusion.
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 829
PostPosted: Wed Mar 09, 2011 4:21 pm Post subject: Reply with quote
With regard to Black Swans, would Adjusted +/- be considered a scalable or non-scalable variable?
Back to top
View user's profile Send private message
Jeff Fogle
Joined: 11 Jan 2011
Posts: 70
PostPosted: Wed Mar 09, 2011 4:59 pm Post subject: Reply with quote
I'm too deeply mired in Mediocristan to be certain...
Back to top
View user's profile Send private message
Jeff Fogle
Joined: 11 Jan 2011
Posts: 70
PostPosted: Wed Mar 09, 2011 8:00 pm Post subject: Reply with quote
I always get those reversed in my head because I think of "scaling" a building or a mountain or something. Then, I imagine a graph showing book sales that have mostly zeroes but then Harry Potter books. I couldn't scale that tall bar graph for the JK Rowling books. But, that's the "scalable" variable while stuff like human height or weight is non-scalable. I could climb over bar graphs of human height! Should have majored in math rather than journalism back in the day. Word definitions always getting in the way...
Looks like Adjusted plus/minus is non-scalable in terms of its range...unless you're counting a crash to zero from a serious injury as a black swan or something. Can't imagine a future superstar who's so much better than the rest that they would correspond to Rowling book sales over a typical book's sales. Maybe Tiger Woods number of tournament victories would seem scalable considering his dominance at his peak. But, it's not like he was shooting 55 every round or something.
Am I in the ballpark or did you have something else in mind?
Sincerely,
Mired in Mediocristan
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 829
PostPosted: Wed Mar 09, 2011 9:41 pm Post subject: Reply with quote
Quotes from recent True Hoop post on Miami Heat and dribbling:
Against these [top] teams, the Heat have scored under 100 points per 100 possessions, about 10 points worse than their average...
why is this happening?
When trying to answer this question, the first thing that jumps out to me about the Heat as a team is that these guys dribble a lot.
...Dribble charts show where and how much teams dribble in order to score, and the Heat have a big red area indicating that they dribble to score more than any other team."
Questions:
Do the Heat in fact dribble a lot against the top teams? The data provided appears to be against all teams.
Do the Heat dribble more than their average against each team or just some of them? Is the average against them more or less than the Heat's average? More than the league average against these top teams or less? More than other contenders against these teams or less?
How correlated are the dribble counts against the top teams with the Heat winning? Are we again talking about a sample size of a dozen or so games?
How correlated are the Heat dribble counts against all teams with the Heat W-L? How correlated are the dribble counts of all opponents and other contenders and their W-Ls against these top teams?
What is the significance of these findings?
What other Factor trends or sub-Factor trends are occurring when the Heat play these top teams? Surely there are some, though unnamed. What % of the offensive fall-off is reasonably associated with these other elements? What % of the problem is over-dribbling being asserted to be? Why not provide an estimate?
Dribbling more than average is not necessary a problem. The Heat after all are 5th best on offensive efficiency. If they dribble more than average overall, how does it work in general but not against the top teams? Are turnovers up? Are at the rim shots a lower % and cut-off mid-rangers higher?
What is the relationship between dribbling and getting foul calls in general and against top teams? If it generally helps get foul calls, are they getting the calls as much as normal or lighter? Are the calls on average considered close to "fair" or not?
How many possessions on average do the Heat "over-dribble"? When? Is it worse on the road vs home? Is the timing different? Is it particularly worse at the end of the game? Does zone vs man defense impact this?
What defines over-dribbling beyond being above league average? Is there only way to play? What is the average and how much are the Heat above it? How many dribbles is acceptable? What do Coaches say on the topic? What do players say? Are there different schools of thought? Are there any advantages to recognize on the other side of the balance?
Are all dribbles being treated equal in the count toward over-dribbling or wouldn't some dribbles be more or much much more important and more negative than others? What are they and when?
Are there games where they over-dribble but over-dribble in a better, more effective way and win more than they tend to do against top teams and could / should they try to use that pattern more and the 2 star player strengths / comfort zone that way?
Has the Heat over-dribbling stayed the same all season, gone down or up in general and against top teams over time? Is the Heat a unique case or does over-dribbling tend to accompany teams with high turnover? Or weak PGs? Or low inside scoring? Or modest 3 point shooting frequency?
Why are they over-dribbling? What specifically and physically are defense doing to encourage or cause this?
What if anything has the Heat noticeably tried to address this? Is this something they "fix" and then it slips or is it chronic? Was this a problem with previous Wade teams? Spoelstra-Wade teams more less than Riley-Wade?
Which players are over-dribbling? Does it vary by whether 1, 2 or 3 stars are present on the court? Are they doing it more or for a worse affect on W-L than last season, in some cases on other teams? Were their teammates dribble rates higher, lower or about the same then as with the support cast on this Heat team?
How does overall player scoring and efficiency change at game level with their dribbling? Do some players have a positive performance relationship with dribbling or "over-dribbling", while other are negative or is the tide strong one way? Are the effects more with the player who is over-dribbling or is it on the other players and the teal-level data?
Is there any link between efforts to "take a player away" and over-dribbling? Which is having more impact? Are these issues somewhat opposite of each other? If not, why not?
Is over-dribbling more or less a problem in the playoffs? Has an over-dribbling team won the title?
If they dribbled less, what would they do (besides pick 'n roll) and is there data to suggest they can do that well against these top teams? And is pick n roll actually lower dribbling than average? By how much? What is the marginal efficiency of pick 'n roll for the Heat? On average how frequent and successful is pick 'n roll against these teams and where does the Heat's frequency and success rate against them sit against other teams against these top teams? How does it compare to the Heat's overall average pick n' roll frequency and success rate against the rest of the league? How does it compare to other team's overall average pick 'n roll frequency and success rate against the rest of the league? Is more pick 'n roll really a likely solution? Have they tried to do more pick 'n roll than average against these top teams? What happened?
Same considerations for catch 'n shoot and anything else beyond over-dribble possessions.
Much more analysis along these lines or other angles would be helpful to address this issue adequately. One issue in a bigger picture.
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 829
PostPosted: Sat Mar 12, 2011 1:50 am Post subject: Reply with quote
Some revised comments on the scalability question (and basketball Black Swans):
As you suggest Jeff, it is probably right to say Player Adjusted +/- treats and / or finds player value as non-scalable, a bit more so in RAPM than traditional Adjusted +/-.
Adjusted +/- for lineups, sub-units (like pairs) and clutch time might be or are somewhat more scalable, if you allow shades of grey between non-scalable and scalable.
I'd think a Championship Added version of Player Adjusted +/- though is probably demonstrating scalable player value. (similar in intent as past work by Scott Sereday and Neil Paine).
I guess it might help to look closer at and compare the scatter plot distributions of all these metric values.
Basketball Black Swans often have pretty rare physical sets or I guess you could broaden out to pretty rare physical / mental sets.
Players like KG, Shaq, Pierce, Gasol, Duncan or in earlier days Jordan, Hakeem, Jabbar, Rodman, B. Wallace, R Wallace, Magic, etc.
Are (or were) LeBron, Dirk, Paul, Yao, Kidd, R Lewis (previously really high on Adjusted+/-) and Nash Black Swans who didn't have enough help or had bad luck or were they fundamentally lesser in championship power?
Certainly they really look hard for and try to get super stars, true super stars. test for them, try to build around them and give freedom to them. But I don't know how much the NBA can "develop" a potential Black Swan into one or try to copy one and get near the impact.
Maybe some version(s) of Adjusted +/- can play an important role in identifying and understanding them. Has there been a recent title winning superstar / Black Swan who wasn't at least very good on Adjusted +/- that season? I didn't find one. Kobe Bryant is having a bad year on 1 year Adjusted +/- and that might be pivotal or not.
Anything that can help with earlier identification, reduction of impairments or greater synergies with support elements would be very valuable.
Back to top
View user's profile Send private message
Jeff Fogle
Joined: 11 Jan 2011
Posts: 70
PostPosted: Sat Mar 12, 2011 1:41 pm Post subject: Reply with quote
Not sure if "black swan" is the ideal term for players. I think Taleb was referring more to "events" that were rare but had a high impact on history. I think the overdose of Len Bias was a Black Swan...or even San Antonio catching a big lottery break so they could draft Tim Duncan.
http://en.wikipedia.org/wiki/Black_swan_theory
Might be better to come up with a different term for unique players. Or use "stars" or "superstars" or something like that when referring to the impact a star player has on a game/team/season...then use "black swan" for an event that wasn't naturally predicted by past history but then strongly influenced events going forward.
The lesson from the book that always grabs me about building models is the much-cited turkey analogy. A turkey builds a model of expectations for what is day is going to be like based on all of his previous days. He wakes, up, gets fed a bunch of grain, enjoys the beautiful weather, plays with his turkey buddies, goes to sleep, wakes up... He develops extreme confidence of what his next day is going to be like, until a certain Thursday in November.
Also like all the anecdotes about "experts" on CNBC who's predictions fail to match what subsequently happens. Natural flow from that into the Michael Lewis book "The Big Short" and other books about failed hedge funds who trusted their models too much.
Complicated world. Fun to try and see what jigsaw pieces can go together. Modeling can be humbling given the bedlam that may be lying in wait...
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 829
PostPosted: Sat Mar 12, 2011 2:08 pm Post subject: Reply with quote
Alright. Yes, Black Swans was referring to events. Maybe it is not the right term to use for basketball superstars but I was mainly interested in the non-scalable / scalable question that arose in connection with Black Swans and transferring it to the basketball context.
When I briefly touched on the rare physical make up of some of the superstars, I guess I was thinking that their arrival and success was something of an announcement of a new type (the big PG; the agile, really long guy with all the offensive and defensive skills, the center with an unstoppable jumper, any big with a strong 3 pt game, the super-sized wing, etc.). A new type that cut against the history of position expectations or expectations based on size or at least represented heightened success of a player type (lightning guards, agile / versatile big man defenders, play-making SGs) and that could be considered something of an event in that it might cause league change in player search and player use.
Back to top
View user's profile Send private message
Jeff Fogle
Joined: 11 Jan 2011
Posts: 70
PostPosted: Sat Mar 12, 2011 3:16 pm Post subject: Reply with quote
I see what you're saying Crow. Not sure if I can think of the right word for a new type of breakthrough player in that vein. Kind of like a mutation that triggers evolution in the sport...but they're not really mutants, or "random" mutations like in natural selection. Played around with some words but nothing seemed right. New types or evolutionary ticks. I googled evolutionary ticks and learned about the evolution of ticks (lol).
To me, I think most of the stat stuff we're looking at in the NBA is going to be non-scalable. We're focusing very much on the right end of the bell curve of people who play basketball. It's just that tip. In a way...book sales are scalable...but once you're only looking at the different Harry Potter books as compared to each other, they're so close together that it's non-scalable. If the NBA had 10,000 teams and me and you got to play....Kevin Garnett is going to seem scalable. Within the context of a pre-selected elite, not so much.
So, if we're trying to come up with mixes that are greater than the sum of their parts, I think we're still dealing with non-scalable rather than scalable. Or, if we're looking for a certain type of impact player, same thing.
Would be interested to hear any other perspectives on that. Wouldn't pretend to be an expert on scalability or power curves.
In terms of anticipating black swans. That seems very hard to do by definition. I think teams can try to prepare themselves for the future occurence of black swans (building depth and versatility for when a high impact injury comes along). They can have a sense of humility about possibilities rather than just assuming everything is going to be great...and invest their resources accordingly (overconfidence leads to bankruptcy, so don't be overconfident...extreme risk-taking is more dangerous than it seems because only the survivors are around to tell you how great it worked out, etc...)
In terms of building models...stuff that always worked out in the past isn't guaranteed to work in the future...
Back to top
View user's profile Send private message
huevonkiller
Joined: 25 May 2010
Posts: 15
Location: Miami, Fl
PostPosted: Sat Mar 19, 2011 3:51 pm Post subject: Reply with quote
My impression is that this is being over thought.
In the madness of the Heat's losing streak, there were a lot of reactionary articles and conclusions. All it means is the LeBron James takes over in hostile situations, and defers when his team is doing well. On the Road he averages 27.7 points, and his other statistics are better. All of his 50 point games have been on the road, and he would have had 50 again against the Hawks if it wasn't a blowout.
Back to top
View user's profile Send private message
Mogilny
Joined: 05 Aug 2010
Posts: 25
PostPosted: Sun Mar 20, 2011 4:20 am Post subject: Reply with quote
On both the Taleb and OP topic - isn't the noise/info ratio too high to draw any real conclusions from the data regarding who to take away?
Back to top
View user's profile Send private message
bbstats
Joined: 25 Apr 2010
Posts: 46
PostPosted: Sun Mar 20, 2011 12:51 pm Post subject: Reply with quote
Someone should do a correlation study on D-Wade's ASPM versus the Heat's Actual minus Expected point differential.
_________________
http://thebasketballdistribution.blogspot.com
http://twitter.com/bbstats
Most important Heat star to take away (Crow, 2011)
Re: Most important Heat star to take away (Crow, 2011)
Anybody want to summarize who teams tried to take away from the Heat in the playoffs and how and how well or not well they did with that task and what were the spin-off effects? Who will Dallas try to take away and will it work?