Comments on: Predicting a Player’s Impact on Teammates’ Three-Point Shooting http://www.hardwoodparoxysm.com/2009/11/05/predicting-a-player%e2%80%99s-impact-on-teammates%e2%80%99-three-point-shooting/ Unbiased opinions from extremely biased people Wed, 23 May 2012 04:55:00 +0000 hourly 1 http://wordpress.org/?v=3.3.2 By: JordanPushedOff http://www.hardwoodparoxysm.com/2009/11/05/predicting-a-player%e2%80%99s-impact-on-teammates%e2%80%99-three-point-shooting/comment-page-1/#comment-27313 JordanPushedOff Sat, 07 Nov 2009 01:31:13 +0000 http://www.hardwoodparoxysm.com/?p=3637#comment-27313 Jon, Really interesting stuff. I'm pretty sure that what you did here was compute 24 different simple bivariate correlations (or in other words, a regression with only one right-hand-side variable) -- i.e. you correlated MinutesPerGame against 3PA, then you correlated MinutesPerGame against 3P%, then you correlated PointsPer40 against 3PA, then you correlated MinutesPerGame against 3P%, etc. The problem with this is that the "true" model of player performance should involve all 12 explanatory RHS variables simultaneously. That's why your R-squared values are so low for each of those bivariate correlations. ...And in turn, the problem with the scenario I just described is (i) all the RHS variables [particularly PER, which is an artificial construct from other stats] would be mutually correlated, which the econometricians call "multicollinearity", and (ii) you would have a reverse causality problem, which the econometricians call "endogeneity". So I don't have any good solutions (that's why I'm not a professional statistician, or a professional NBA analyst!) but those are just some thoughts. Jon,
Really interesting stuff.

I’m pretty sure that what you did here was compute 24 different simple bivariate correlations (or in other words, a regression with only one right-hand-side variable) — i.e. you correlated MinutesPerGame against 3PA, then you correlated MinutesPerGame against 3P%, then you correlated PointsPer40 against 3PA, then you correlated MinutesPerGame against 3P%, etc.

The problem with this is that the “true” model of player performance should involve all 12 explanatory RHS variables simultaneously. That’s why your R-squared values are so low for each of those bivariate correlations.
…And in turn, the problem with the scenario I just described is (i) all the RHS variables [particularly PER, which is an artificial construct from other stats] would be mutually correlated, which the econometricians call “multicollinearity”, and (ii) you would have a reverse causality problem, which the econometricians call “endogeneity”.

So I don’t have any good solutions (that’s why I’m not a professional statistician, or a professional NBA analyst!) but those are just some thoughts.

]]> By: Mid-Afternoon Milk Mustache, featuring the secret Sports Guy phone call | Stacheketball, an NBA Blog http://www.hardwoodparoxysm.com/2009/11/05/predicting-a-player%e2%80%99s-impact-on-teammates%e2%80%99-three-point-shooting/comment-page-1/#comment-27204 Mid-Afternoon Milk Mustache, featuring the secret Sports Guy phone call | Stacheketball, an NBA Blog Thu, 05 Nov 2009 21:49:18 +0000 http://www.hardwoodparoxysm.com/?p=3637#comment-27204 [...] The Lampshade: Nichols on predicting a player’s impact on his teammates’ three-point shooting [Hardwood Paroxysm] [...] [...] The Lampshade: Nichols on predicting a player’s impact on his teammates’ three-point shooting [Hardwood Paroxysm] [...]

]]> By: Matt Nolan http://www.hardwoodparoxysm.com/2009/11/05/predicting-a-player%e2%80%99s-impact-on-teammates%e2%80%99-three-point-shooting/comment-page-1/#comment-27167 Matt Nolan Thu, 05 Nov 2009 17:02:40 +0000 http://www.hardwoodparoxysm.com/?p=3637#comment-27167 An interesting approach to things. One stat that I think could provide some sort of better explanation would be how often said player takes shots at the basket (not makes) as this would mean that defenses are more likely to shade off players and collapse in on that player leading to more open three point attempts if the player is able to get the ball to an outside shooter. An interesting approach to things. One stat that I think could provide some sort of better explanation would be how often said player takes shots at the basket (not makes) as this would mean that defenses are more likely to shade off players and collapse in on that player leading to more open three point attempts if the player is able to get the ball to an outside shooter.

]]>