APBRmetrics

I am constantly curious about using rpm approach in additional ways. In addition to many previous ideas / opportunities I now wonder about:

APM for PGs individually and as league composite when with and without another 20-25 plus assist rate guy. With and without 3 plus qualifying "competent 3 pt shooters" including themselves.

And there are probably other cuts that could yield results that give hints / maybe illumination that is not common knowledge. Maybe one could be a apm run for off. Rebounds - opp. Fast break buckets. Maybe the strategy of favoring transition defense over offensive rebounds could be better optimized for lineup and / or time and situation.

What would you find if you compiled the lineup apm for all lineups where the top 2 / bottom 3 combined player seasonal avg. usage ratios were high, medium and low, adjusted by the avg. apms of who was on the court to confer whether the usage ratio was possibly a low threshold significant interaction term for offensive apm or total apm?

Would it lead to more understanding than possible with just the data and analysis in http://bleacherreport.com/articles/2167 ... ba-forever ?

Not much shared interest in this line of thinking but one could also look at offensive apm for players who were +1 or better on offense for just when they played against a +1 or better apparent counterpart or -1 or worse defender. Could also take it down to factor level and see who does well or poorly when they face strength or weakness at the factor level. Smaller sample size to be somewhat offset by using 2-4 yrs of data. May not give definitive answers but surely it would raise more questions and suggest answers for further study & refinement.

I continue to think that situational apm data would provoke interesting questions, tentative answers and additional research. Apm runs for player performance against well above average, near average and well below team offensive and defensive efficiency would seem worthwhile to look at me and you could take it down to the same tiers of opposing team factor level performance too. You could look at not only the individual data readouts but also the readout for player type clusters too. Which player type do best / worst against the various best teams of these stripes. Hal Brown recently talked about performance against best defenses by team shooting styles. You could add player level information using this approach for the different shy distribution styles and maybe better understand who is and isn't succeeding against the best and maybe go on to learn something about why with further stat and video work.

I've long thought about taking a similar approach to APM, though frankly I don't have the statistical chops at this point to actually go and do so.

To me, the question is both what role a player is playing and what roles the guys around him are playing in a given lineup. I feel like the biggest blind spot of APM is that it can't or doesn't really account for a guy playing an entirely different role when he's on the court with certain players. Consider a guy like Gerald Green who goes from being a stationary spot up shooter in a slow, inside-out, offense in Indiana to a free reign, launch shots as you please, run and gun player in Phoenix. It was as if his role went from being Courtney Lee to being J.R. Smith. Ariza's another guy who strikes me as someone who's had their role changed drastically and had wildly varying results because of it. Those may be extreme examples, but obviously the role of a player often isn't static across lineups or teams.

My biggest concern would be that once you start breaking down guys into a bunch of roles, your sample size for an APM type approach might became too small pretty rapidly.

Splits that divide the data into 2 or no more than 3 piles and using 2-3 years of data would keep the concern from being huge imo. Multi-year has its own issues but the point is to do the best you can. With careful review it might help more net than "hurt", though there is risk of being misled, with this or just about any data.

knarsu recently produced charts showing usage, ts% and overall rapm of players. It might get closer to the specific direct implications of high, medium and low usage in NBA by those particular players and player classes or roles to look at offensive rapm of players when they actually qualify for usage vs plays were they do not. The overall view should not be lost but it doesn't finely answer the limited question that we might be trying to isolate. Looking at knarsu's charts is proper for evaluating players as wholes but not I think for assessing the efficacy of certain usage levels as practiced by those players (and not by possible others).

It would be worth computing the rapm for shot usage separate from assist usage. And with the video day available from sportsvu and vantagesports one could try to estimate the particular value of particular picksettees objectively rather than subjectively and the same for off the ball positioning by relatively static "spacing" in last few seconds before the shot or by overall space-motion throughout play rather using a subjective value derived from 3 pt shot rate or power impact or perhaps a yet more comprehensive, sophtiscated but still box score based proxy estimate for unique patterns of space-motion.