I finally have the PA100 metric to a place where I am pretty happy with it. It isn't perfect. I think I need to account for diminishing returns in a few areas, namely shot creation and offensive rebounding, but I trust the results. Individual scores explain about 97-98% of team efficiency differential. This is higher than I thought I would get given that I am not including any "team" statistics, even unassigned rebounds and turnovers (like shot clock violations.) Furthermore, season to season PA100 scores are correlated at just around 80% for players logging at least 2000 possessions (.8 for differential and offensive and .75 for defensive.) Honestly, I don't know what I should be shooting for here, but not even controlling for age, I can't imagine reality is much better than an 80-90% correlation.
In addition to the basic PA100 numbers I am also including another new gizmo I have been working on. The "Help" defensive statistics are not factored into PA100 scores, but offer a handy diagnostic tool for identifying players whose defensive scores owe something to cross-matching or who cover the whole floor rather than focusing on locking their man down.
Points added per 100 offensive possessions above the average player at his position. Gives credit for creating better offensive situations (i.e. rim or three for himself or others). Also gives credit for outperforming the situation of his shots by shooting with accuracy above expectation. For example, when Ricky passes to an open Kevin Love three-pointer, Ricky gets credit for the expected value of an assisted three above and beyond the lowest value shot (an unassisted midrange jumper). Love then earns points for making that shot above average, and loses points for making it below average.
Mostly the PA100.off score of player's positional counterparts. However, credit for blocks and steals is given to the blocker/stealer. For example, when Nikola Pekovic blocks Steve Nash, Pekovic gets credit for a missed rim shot, while Rubio is debited for an allowed rim shot.
PA100.off - PA100.def
HLP and component numbers:
These numbers show the difference in how non-counterpart opposing players perform when ego is in the game. Do opposing shooting guards perform worse with Ricky is in the game, or on the bench? What about opposing centers? This measure gives players a quarter share of responsibility for the performance of each non-counterpart opposing player. The HLP measure simply combines the player's impact on the four other positions over an average 100 possession period.
The biggest concern to keep in mind when interpreting these numbers is the cohort effect. Some players are almost always on the court together, and this makes the HLP measure for that position largely useless.