Characterization of Player Tracking Plus Minus composition
Posted: Mon Jun 13, 2016 8:08 am
This statement / question is largely for the author of PT-PM to comment on, if Andrew J. wants to:
PT-PM is some % RPM (which is box score informed RAPM with some twists,where the RAPM tries to account for everything including all non box score impacts) and % value estimate of a handful judged very important non box score impacts. So the box score impacts are "in there" via the prior and via RPM's implicit consideration of everything and the judged most important non box score impacts are "in there" via RPM's implicit consideration of everything and via the PT component and the non box score, non PT impacts are just "in there" once via RPM? And that single counting puts them in distant third to the boxscore and player tracking data because... of our understanding of what's most important and without enumerating or analyzing these non PT, non box score impacts? Or because this is a defacto method to try to reduce the influence of presumed noise / error, believed to be coming mainly or entirely from somewhere beyond the boxscore and PT impacts?
It seems to work well. But do we stop there? Or backtrack to some version of pure RPM where everything is in there once? Or go to a RAPM version where everything is more evenly in there by putting everything in there twice (explicitly as best we are able, including more enumerated non box score, non PT impacts and then implicitly via the RPM)?
PT-PM is some % RPM (which is box score informed RAPM with some twists,where the RAPM tries to account for everything including all non box score impacts) and % value estimate of a handful judged very important non box score impacts. So the box score impacts are "in there" via the prior and via RPM's implicit consideration of everything and the judged most important non box score impacts are "in there" via RPM's implicit consideration of everything and via the PT component and the non box score, non PT impacts are just "in there" once via RPM? And that single counting puts them in distant third to the boxscore and player tracking data because... of our understanding of what's most important and without enumerating or analyzing these non PT, non box score impacts? Or because this is a defacto method to try to reduce the influence of presumed noise / error, believed to be coming mainly or entirely from somewhere beyond the boxscore and PT impacts?
It seems to work well. But do we stop there? Or backtrack to some version of pure RPM where everything is in there once? Or go to a RAPM version where everything is more evenly in there by putting everything in there twice (explicitly as best we are able, including more enumerated non box score, non PT impacts and then implicitly via the RPM)?