APBRmetrics

Ingram with a modest fraction of a point positive on RPM and basically neutral on RAPM. But he is headed toward a max deal or very close we are told. Only slightly ahead of Hart. Modest positive on offense offset by modest negative on D. Helps team at getting to line but gives it back by being a weak offensive rebounder.

Giving him the max will probably mean trading Holiday or not retaining Ball and / or Hart. Luxury Tax? They probably want to avoid but maybe not. Depends how good they get or anticipate becoming.

If one ran RAPM for all plays and then RAPM except when a player did or did not do something and compared, you'd at least know the impact of that something on RAPM. Direction, relative indicated magnitude (subject to error).Something being: stand in corner, on fastbreak (or specific actions or not on fb), 2 plus dribbles, set or use screen, cut, pnr ballhandler, etc. Something can be anything. And I'd probably check everything. Not conclusive with small samples but it would add info on how you get to the RAPM estimate. Shifting to excluding something for plays at player type, team or league level would increase sample size and reduce noise for the estimates of the impact of that less black box something present and / or not.

It is long past time to COMBINE RAPM AND PLAYER TRACKING. And as more than an aftermarket blend of 2 separate metrics. Use of player tracking as a prior certainly should be done but I'd think split work would help too. I have suggested plenty of splits over the years. There is no limit. Are some teams doing it? I dunno. A few might be. Or not. Paying attention to / being influenced by these comments or coming to it on their own.

If you can with / without RAPM impacts on everything imaginable and divided by frequency, you'd have a first draft set of impacts of actions. Or do with / without to second or more levels to try to get an impact of sets of actions / choices. No limits. For the imagination. Or for implementation by the capable, paid and interested.

Evaluating Basketball Player Performance via
Statistical Network Modeling
James Piette∗
, Lisa Pham†
, Sathyanarayan Anand‡ 2011

Focuses on data from player pairs / networks.

Old article. Some sporadic public follow-up. I used pair and lineup RAPM data when available. The multi-level approach could help with guessing about lineup management compared to guessing without as many levels of estimates. Consistent data across levels would tend to increase confidence levels. Inconsistent data could help focus attention on potential errors in estimates.