APBRmetrics

Suppose the username is hint enough.

About to embark on a safari through "Short list of past threads worth reading" and "Guides to Creating RAPM". Are they any other reads that are essential, or is the literature listed there all canon? I think there are some "far out there" techniques that would significantly improve a prior, and the sooner I can calculate APM/RAPM in r -- Or MATLAB? Or Mathematica? -- the sooner I can work on what brings me here.

My deepest apologies should this thread be unnecessary, no offence will be taken should it be locked or deleted.

Hi and welcome to APBR.

If you are going to try and replicate RAPM, can I suggest that you document your process to some extent in the "guide to creating RAPM" thread to help others in a similar position going forward?

I look forward to your contributions.

The short list is rough and not well organized. I threw it together for a request and haven't spruced it up. But it is a start.

I can't guess identity from the hint in username, yet. Unless you are in the U.K.

Did you find enough info? Are you going to eventually share the work with the far out techniques publicly?

Crow wrote:Did you find enough info? Are you going to eventually share the work with the far out techniques publicly?

To a degree, and yes.

ampersand5 wrote:Hi and welcome to APBR.

If you are going to try and replicate RAPM, can I suggest that you document your process to some extent in the "guide to creating RAPM" thread to help others in a similar position going forward?

I look forward to your contributions.

I'm essentially using the Hickory High guide as a jumping off point. But I have question:

For splitting the RAPM into oRAPM and dRAPM, is the accepted practice splitting each player in half? Player[1] = Player₁O + Player₁D? This would seem like it adds more noise.

I tweeted a couple of people who probably can advise you.

Definitely a good question for J.E. Not sure if it would be "noisier", considering it basically is the equivalent number of samples in each case. I guess the question is whether adding O+D together is a better or worse predictor than the "total" RAPM case (i.e. fitting point differential). I haven't looked at that myself.

Crow wrote:I tweeted a couple of people who probably can advise you.

Thank you.

EvanZ wrote:Definitely a good question for J.E. Not sure if it would be "noisier", considering it basically is the equivalent number of samples in each case. I guess the question is whether adding O+D together is a better or worse predictor than the "total" RAPM case (i.e. fitting point differential). I haven't looked at that myself.

I can intellectually appreciate an RAPM number without O and D splits, but not knowing how a player's performance on each end of the court affects that number would bother me. And, for the purposes of what I'm trying to do, having the O/D split is essential.

Has there been any work done on if there's an... an inflection point (?) for minutes played cut offs? Or are we just using ~250 minutes?

Are we using the same variable for ALL of those players? Is it simple algebra if two of those players are against each other, canceling each other out?

This is my hurdle right now.

Too bad there hasn't been more public response. I guess there is expertise to protect against competition. I dunno if private inquiry would fair better. Hope to see or hear something of your work eventually.