APBRmetrics

So I've had these two ideas bouncing around in my head for a while now, and I'm curious as to your opinions about them.

RAWPA is Regularized Adjusted Win Probability Added, where instead of tracking score differential as you would for RAPM, you track the total Win Probability Added, along the lines of what you find at inpredictable.com -- and then run the same regressions on that data. It seems to me that such a metric, together with RPM, could be used to assess "clutch" play, or just provide another interesting data point to look at. Would such a metric be useful, or at least interesting?

The other idea is an individual elo rating, which would need to be tracked possession by possession, but could provide more interesting information of a different sort. I'm quite a bit fuzzier on how this would actually work, but it could, for example, adjust each player's rating based on the sum of the opponent's rating minus the sum of the teammates' ratings.

What are your thoughts on these? Their feasibility? Usefulness? Interestingness?

If you had RAWPA, you may not need
to add back RAPM if your focus is clutchness or gong beyond what RAPM already does; but I can support blending on general grounds.

Not sure I follow individual elo scores; but if I do, you'd have to use minutes weighted player plus minus and recognize properly that you have 5 opponents and 4 teammates. Can be done but need to write it out to make it right and express it in the simplest way.

mzellman wrote:...
RAWPA is Regularized Adjusted Win Probability Added, where instead of tracking score differential as you would for RAPM, you track the total Win Probability Added, along the lines of what you find at inpredictable.com

Win probability added doesn't really make sense as a way to measure player quality unless you make some pretty odd assumptions. So, if you make a player stat that's not expected to measure something about the player, I've got to wonder what question you're trying to answer.

The other idea is an individual elo rating, which would need to be tracked possession by possession, but could provide more interesting information of a different sort. I'm quite a bit fuzzier on how this would actually work, but it could, for example, adjust each player's rating based on the sum of the opponent's rating minus the sum of the teammates' ratings.

That could certainly work, but you'll end up with a lot of the same issues that the plus-minus stats do in the sense that the same players spend a lot of time on the floor together. You also have to make assumptions about how player qualities interact.