v-zero wrote:Weights for box score stats are estimates. Not all assists, shot attempts, blocks, steals, fouls, etc etc are made equal, and in the case of assisting there are certain players whose assists are of high value, and then there are those whose assists are really just passes that happen to be the last pass before a shot.
Estimate but it's empirical and potentially the best estimate. I believe theoritical estimates generally forget relations and effects of other factors. Look at field goal miss and blocked shot example.
Still if I wasn't lazy and had analyse-ready PBP data of 1997-2012 I would use advanced stats. In fact I would use anything I could extract from PBP data. Unforunately, I don't know how to pull data besides using Excel's data import functions. In the same time, extracting stats from PBP data is an another issue for me. It would take too much time...
What I'm suggesting is also an empirical estimate. Regression will show (see the formulation of ASPM for example) that a usage-assist interaction is a better empirical estimate of value than mere assists. I can tell you for free that simple regression on assists isn't the best way of estimating the value of assists.
v-zero wrote:What I'm suggesting is also an empirical estimate. Regression will show (see the formulation of ASPM for example) that a usage-assist interaction is a better empirical estimate of value than mere assists. I can tell you for free that simple regression on assists isn't the best way of estimating the value of assists.
Well, usage alone is a big estimate. I was looking into player possession formulas, I wasn't satisfied with them in the end.
Even "Assists" are actually a poor estimate of assists.
Playing at home, the Clippers get 27% more assisted FG than the Magic do, relative to their road rates.
Assuming the home scorekeepers of the league are unbiased about awarding assists to all visiting teams, a better estimate of player assists can be had pretty easily: just use their away rates.
When assist (or any other production) rates are randomly distorted, there will of course be poorer correlations to wins or point differential. When you make a correction -- even a reasonable estimate -- the correlation is stronger.
When you don't correct for differences in pace, rebounding and scoring differential, home vs away, starter vs sub, you're assuming a factor of 1.
This is a very crude estimate to apply to all players on all teams.
What do you suggest me then? Shouldn't there be a raw box-score metric? PER's empirical version... I believe I can't produce a better advanced model than your model or anyone else's here because I don't have enough time to learn data scraping. Because I will never use box-score data to build a model that involves advanced stats. All I can do is generate ideas without executing them. I can't get PBP data neither so this is the best I could do.
I believe the future is simulation. Not only in this field but everywhere. I would like people here to help eachother and develeop a very complex model to run simulations for each game than making their own version of advanced models. I would help for it too.
Simulation is an interesting field but it is quite questionable as to its value in a situation as well-defined as a basketball game. Because of the strict rules in basketball there is a general, simple pattern to games and the score/statistics of the game to which classical statistics can be applied with great success. Simulation is more useful in more complex systems, such as weather systems, where any attempt at an analytic solution to a problem is met eventually with the problem of complexity.
I can, however, see a use for the simulation of a basketball game wherein the properties of the players can be varied to attempt to theoretically derive the importance of interactions in basketball, but a large fine-grained dataset on player movements relative to each other and the ball would be required for that. One day, maybe.
Simulation is an interesting field but it is quite questionable as to its value in a situation as well-defined as a basketball game.
naysayers - no vision...
I can, however, see a use for the simulation of a basketball game wherein the properties of the players can be varied to attempt to theoretically derive the importance of interactions in basketball
yes, yes, exactly.... your vision is improving...
but a large fine-grained dataset on player movements relative to each other and the ball would be required for that.
no, what is currently available today works just fine...
bchaikin wrote:I believe the future is simulation.
here, here...
Simulation is an interesting field but it is quite questionable as to its value in a situation as well-defined as a basketball game.
naysayers - no vision...
I can, however, see a use for the simulation of a basketball game wherein the properties of the players can be varied to attempt to theoretically derive the importance of interactions in basketball
yes, yes, exactly.... your vision is improving...
but a large fine-grained dataset on player movements relative to each other and the ball would be required for that.
no, what is currently available today works just fine...
One day, maybe.
* sniff * ...
'No vision' is not a real response, try to be a bit more informative and a little less patronising and then I will have a discussion with you. As it happens I have done some simulation work on the many body problem in Quantum Mechanics, so I have a decent idea on the level of data required.
Do some searches. Ask some questions. Bob isn't shy about sharing information about his simulator. It used to be available for free on the web. Not sure if that's the case anymore.
I have done some simulation work on the many body problem in Quantum Mechanics, so I have a decent idea on the level of data required.
perfect then...
think of the predictive ability of those who worked on the strong force and how they were able to deduce the existence of quarks strictly from mathematics. people like murray gell-mann and sheldon lee glashow were able to predict and instruct others on how to find yet to have been found baryons simply because they understood the underlying mathematics of the system (i.e. nature) before others did...
or even earlier, think of the periodic table, using mathematics and experimentation dmitri mendeleev was able to make predictions of yet to be discovered elements because he understood the underlying mathematics of the system (again nature), again, before others did...
how about quantum chemistry? neils bohr's simplified model of the hydrogen atom works well even today for many macroscopic applications, but the underlying model has been improved with the mathematics of quantum mechanics (the s, p, d, f orbitals, electron clouds, etc) such that even better predictions can be made...
granted these are far more grandiose examples, but then why should the field of sports (or any field for that matter) be any different? once you find an underlying mathematical pattern in a real life scenario the rest is fun to develop, tweak, improve, test, and implement, and it's predictive ability is easy to test...
I would need to see the positive results above classical-style basketball statistics to have that mean more than simply being an interesting take.
just because you may not know about them does not mean there aren't nba basketball simulations out there. try a few and see if the results are positive enough for you. if not, you yourself can always improve upon them, just as others have...
'No vision' is not a real response, try to be a bit more informative and a little less patronising and then I will have a discussion with you.
Although I don't agree with Wins Produced method of position adjustment and do really believe in average weight of assists in PTR, because of the nature of box score; it doesn't reflect guards' some abilities which improves forwards and centers' box-score productivity thus I decided to include top 10 lists categorized by player position (guard, forward, center).
I need a suggestion here. Should I categorize players by their rough positions such as guard, forward, center or should I go with PG, SG, SF, PF, C? First option has an advantage of covering multiple positions thus somewhat minimizing the error comes from ongoing player position changes in games . Second option has the advantage of more specific and valid comparisons between players but it's error margin is bigger.
If you think I should go with the second option, where do you think I should take off the players' primary positions? ESPN, Fox, Yahoo etc.