I believe I have brainstormed at least a little on the topic of GM and coach profiling before. I haven't looked back at what I might have said previously but I think it would be an extension to say these things:
Can you look at the stream of GM actions and biases and program a model to predict future actions in a way similar to what I imagine chess computer models and chess masters do? Is there any good literature on profiling GMs and coaches that folks can reference or summarize? Is there literature or practice that can taken from financial markets that could be applied or more fully applied to the markets GMs compete in?
In addition to trying to characterize / cluster GMs and perhaps improve ability to guess trends in behavior or specific actions (better than without this systematic effort), could you predict who is learning or not so much, getting better or worse based on "comps" of other past GMs or coaches who behaved and looked "similar" in demographics and in the data analysis? What about performance and psychological testing? Even physiological monitoring. If nutrition, sleep matter before a game, they matter before the draft, trades, signings, analytic work and major meetings too. Have the decision makers who probably generally consider themselves pretty important actors adequately studied each other and themselves based on that assessment of impact? I am sure management thinks about these topics but have they applied "analytic tools and methods" sufficiently? Could you get to point where you could say if a GM or Coach in general hasn't learn to be proficient or above average on skill x,y or z by season 3, 5 or 7 that the data shows that they are this likely to achieve or not achieve it later or "in time" or that this specific "type" of GM or coach will or won't achieve based on performance of similars?
GM and coach profiling, evaluation, prediction
Re: GM and coach profiling and prediction
Funny, this is very similar to what a certain team owner asked of me, whether I'd be able to reverse engineer what other teams are looking at for roster building (especially from the draft) . I told him it SEEMS like there wouldn't be enough data to come up with a solid grasp on team tendencies - taking into account coach & gm changes which, in theory, could definitely change the approach in terms of roster construction for that team.Crow wrote:I believe I have brainstormed at least a little on the topic of GM and coach profiling before. I haven't looked back at what I might have said previously but I think it would be an extension to say these things:
Can you look at the stream of GM actions and biases and program a model to predict future actions in a way similar to what I imagine chess computer models and chess masters do? Is there any good literature on profiling GMs and coaches that folks can reference or summarize? Is there literature or practice that can taken from financial markets that could be applied or more fully applied to the markets GMs compete in?
In addition to trying to characterize / cluster GMs and perhaps improve ability to guess trends in behavior or specific actions (better than without this systematic effort), could you predict who is learning or not so much, getting better or worse based on "comps" of other past GMs or coaches who behaved and looked "similar" in demographics and in the data analysis? What about performance and psychological testing? Even physiological monitoring. If nutrition, sleep matter before a game, they matter before the draft, trades, signings, analytic work and major meetings too. Have the decision makers who probably generally consider themselves pretty important actors adequately studied each other and themselves based on that assessment of impact? I am sure management thinks about these topics but have they applied "analytic tools and methods" sufficiently? Could you get to point where you could say if a GM or Coach in general hasn't learn to be proficient or above average on skill x,y or z by season 3, 5 or 7 that the data shows that they are this likely to achieve or not achieve it later or "in time" or that this specific "type" of GM or coach will or won't achieve based on performance of similars?
I think something like this could be done more for baseball - where you have multiple levels of minor leagues for each organization, involving 100s of players.
Speaking of baseball, & kinda unrelated to the original question - it still appears most teams do a piss poor job of D league utilization, where D league player acquisitions seem so random - & so many players are beyond the age of expected upside. There seems to be little consideration that MAYBE if an organization took their player analysis deeper well past their current NBA roster & try to actually get real "potential" into their D League roster - maybe they'll find some future gems more often than almost never (extreme case obviously is Whiteside - but I am about certain there are guys that could be the next Tony Allen, or Patrick Beverley, or Korver out there). Heck, Kyle Singler & Lance Thomas still get NBA minutes (let alone D league minutes) with no perceivable upside whatsoever.
Re: GM and coach profiling, evaluation, prediction
Interesting to hear a major certain someone has contemplated the subject of my remarks. Not surprised really. The most probable could have been. At first contact, might not be a lot to say with confidence but lots of eventual insight came from someone tinkering on something privately, not knowing if anything useful would come from it.
Using a combination of cluster analysis and hunches from observation and evaluation you might be able to identify tendencies for at least some. The two together should have more potential than either alone. Looking at similars could be waste of time but there might be some tendency for GM and Coaches who played high level ball to do things a bit different than those how did not. And perhaps tendencies by position played or organization where they understudied, degree, etc. You don't know what might fall out and be revealed until you do the analysis. Who is more comfortable with risk with players, with their own career? How is more influenced by star flash, metrics, mock drafts / high school recruiting, practice style, personal appearance, personality, others lifestyles, etc. GMs tend to have longer tenure than Coaches. there is more data on each than is being used hard, probably. What if you paired a programming / stat whiz with a former FBI profiler (and / or a psychiatrist, linguist, etc.) ? Adam might not want to hear about that though.
Analysts who move from one team to another are a potential security risk, but of course so are asst. coaches, head coaches, executives. But it is just one or a few prior teams. ( Unless a covert information sharing alliance were to have a moment or a plan. Risky. Probably unlikely but ... I imagine some execs share some info with some friends / lesser rivals about greater rivals. Doing it with analytic material is a different level but things aren't always completely locked down. How far would people go if they thought somebody had a secret powerful tool / "sauce"? Folks sometimes go far in other multi billion dollar industries. Mouths talk, ears hear, fingers type.) Hire a former political or military intelligence officer or "corporate spy" / "market analyst" (don't have to break law while paying unusually close attention) Just musing at the moment... (Though coincidentally there is a NBA exec who was an intelligence officer.)
Even more taboo would be profiling / analyzing owners, since they may give final approval or block some decisions that can have major impact. Team surely do to some extent. But again, the extent could be extended to more analytic means or beyond.
As some try to go granular into the optic play and sub-play data, others could try to decode / understand better at top team strategic levels. Others and their own. There are probably worthwhile discoveries / surprises out there for teams about themselves that they don't current understand (or at least not adequately acknowledged, contemplated, accepted consciously or changed).
Using a combination of cluster analysis and hunches from observation and evaluation you might be able to identify tendencies for at least some. The two together should have more potential than either alone. Looking at similars could be waste of time but there might be some tendency for GM and Coaches who played high level ball to do things a bit different than those how did not. And perhaps tendencies by position played or organization where they understudied, degree, etc. You don't know what might fall out and be revealed until you do the analysis. Who is more comfortable with risk with players, with their own career? How is more influenced by star flash, metrics, mock drafts / high school recruiting, practice style, personal appearance, personality, others lifestyles, etc. GMs tend to have longer tenure than Coaches. there is more data on each than is being used hard, probably. What if you paired a programming / stat whiz with a former FBI profiler (and / or a psychiatrist, linguist, etc.) ? Adam might not want to hear about that though.
Analysts who move from one team to another are a potential security risk, but of course so are asst. coaches, head coaches, executives. But it is just one or a few prior teams. ( Unless a covert information sharing alliance were to have a moment or a plan. Risky. Probably unlikely but ... I imagine some execs share some info with some friends / lesser rivals about greater rivals. Doing it with analytic material is a different level but things aren't always completely locked down. How far would people go if they thought somebody had a secret powerful tool / "sauce"? Folks sometimes go far in other multi billion dollar industries. Mouths talk, ears hear, fingers type.) Hire a former political or military intelligence officer or "corporate spy" / "market analyst" (don't have to break law while paying unusually close attention) Just musing at the moment... (Though coincidentally there is a NBA exec who was an intelligence officer.)
Even more taboo would be profiling / analyzing owners, since they may give final approval or block some decisions that can have major impact. Team surely do to some extent. But again, the extent could be extended to more analytic means or beyond.
As some try to go granular into the optic play and sub-play data, others could try to decode / understand better at top team strategic levels. Others and their own. There are probably worthwhile discoveries / surprises out there for teams about themselves that they don't current understand (or at least not adequately acknowledged, contemplated, accepted consciously or changed).
Re: GM and coach profiling, evaluation, prediction
Opponent (and self) prediction is a thing for Rock-Paper-Scissors programs (http://ofb.net/~egnor/iocaine.html). Chess programs generally just look for the best move.Crow wrote:...
Can you look at the stream of GM actions and biases and program a model to predict future actions in a way similar to what I imagine chess computer models and chess masters do?
I think that first-party prediction would be a good place to start: For example working out a good way to compare two players that were scouted by different scouts, or making a model for predicting the GM's own ratings. Once those models are established, it's possible to start looking at the more challenging problem of predicting outside GMs.