An effort to create a efficiency/volume public-friendly stat

[v-zero]

Ok, well my opinion comes from my recent work on a predictive box-score metric, which suggests no strong usage-usage interaction, but a very strong usage-3pa interaction, which to me says wing scorers.

[DSMok1]

Ok, well my opinion comes from my recent work on a predictive box-score metric, which suggests no strong usage-usage interaction, but a very strong usage-3pa interaction, which to me says wing scorers.

I don't have a strong USG^2 interaction, either... I may have misunderstood what you were saying.

I think a usage threshold is a good idea. I think that an adjustment TO the threshold based on what the usage is is not necessarily a good idea, though I have that term in ASPM (but the coefficient is small).



One other point, crsofa: USG% includes Turnovers. As this stat currently stands, turnovers are good (!!). It would be good to subtract turnovers out of the usage rate.

[crsofa]

One other point, crsofa: USG% includes Turnovers. As this stat currently stands, turnovers are good (!!). It would be good to subtract turnovers out of the usage rate.

There is a mistake in the formula, it really is Total Points-(League Average Points per Possession*Number of possessions used*Threshold), not divided by "Number of possessions used". I will go to fix the OP.

In "variables" the formula is PTS-(League average points per possession*0.8945*(FGM+FGX*DEF.REB.RATE+0.44*FTA+SHOOTING TO)). Turnovers appear on the term that substracts, so turnovers are always bad.

About the usage/volume discussion what I meant to do here wasn't predictive but explanatory. For example Kyle Korver and Brook Lopez are tied at +3 POSRP/G, with one scoring 11 points in 8.1 possessions and the other 19.4 in 16.8. I don't mean to say that if Korver shot 16.8 times would score 19.4, I have no idea, his efficiency may go down the drain, what I would like to find is the point where the impact is similar.

If you have a Korver on your team making 11 in 8.1, compared to Brook doing 19.4 in 16.8, you only need to replace those 8.7 possessions at a 0.97 clip (average is 109.3) to break even. I think is more or less a reasonable trade-off: chances are that you are probably going to go over 19.4 points with Korver+8.7 possessions to any other player(s), but you might get worse too, whereas Lopez guarantees you that, so at the end they come out rated in a similar way.

Of course that there are a lot of considerations or complications and each scorer is different (some people add the value to expand the floor, others draw double teams, others have the range to score anywherem, others need someone to generate the shot for them...) but this falls out of the scope of this one.

[v-zero]

There is a mistake in the formula, it really is Total Points-(League Average Points per Possession*Number of possessions used*Threshold), not divided by "Number of possessions used". I will go to fix the OP.

In "variables" the formula is PTS-(League average points per possession*0.8945*(FGM+FGX*DEF.REB.RATE+0.44*FTA+SHOOTING TO)). Turnovers appear on the term that substracts, so turnovers are always bad.

About the usage/volume discussion what I meant to do here wasn't predictive but explanatory. For example Kyle Korver and Brook Lopez are tied at +3 POSRP/G, with one scoring 11 points in 8.1 possessions and the other 19.4 in 16.8. I don't mean to say that if Korver shot 16.8 times would score 19.4, I have no idea, his efficiency may go down the drain, what I would like to find is the point where the impact is similar.

If you have a Korver on your team making 11 in 8.1, compared to Brook doing 19.4 in 16.8, you only need to replace those 8.7 possessions at a 0.97 clip (average is 109.3) to break even. I think is more or less a reasonable trade-off: chances are that you are probably going to go over 19.4 points with Korver+8.7 possessions to any other player(s), but you might get worse too, whereas Lopez guarantees you that, so at the end they come out rated in a similar way.

Of course that there are a lot of considerations or complications and each scorer is different (some people add the value to expand the floor, others draw double teams, others have the range to score anywherem, others need someone to generate the shot for them...) but this falls out of the scope of this one.

I recently suggested, and maintain that the two do not really differ. Other than a little mean-reversion there is nothing different between prediction and explanation in science.

I'm not trying to be irritating, and I do quite like how your stat is developing.

[EvanZ]

FYI, yesterday I added my "distance metric" to nbawowy. The formula is kind of complicated, but basically 1 = all-time great scorers like Jordan and 0 = average scorers (defined as 20% USG, 54% TS).

The one thing I'm doing now that is slightly different from what I did before is to to put a sign on the distance so there can be players that are even better than one.

There are two players with SI ("scoring index" is what I'm calling it on nbawowy) greater than 1 this season. I assume everyone here can name them rather quickly.

[talkingpractice]

I would like to posit that the usage interaction only exists for wing scorers, and in general usage does not self-interact. I.e. guys who have high usage but play in the paint shouldn't get any kind of usage trade-off, but wing scorers should.

When you say usg interaction are you referring to usg/efficiency tradeoff? And when you say self-interact youre referring to a usg*usg term in a spm regression (ie you have usg/3pa and usg/ast as significant, but not usg/usg)?

[v-zero]

I'm really only saying that the usg*usg term isn't useful predictively in my findings, but that usg*3pa and usg*ast are.

[DSMok1]

I'm really only saying that the usg*usg term isn't useful predictively in my findings, but that usg*3pa and usg*ast are.

Sounds reasonable to me. I don't think crsofa is using any of those terms in his regression? In ASPM, I use usg*usg and usg*ast, but usg*ast is far more important. In fact, that is the only assist based term I use.

[v-zero]

Have you standardised the coefficients? I would hazard that usg*usg is a pretty big effect in ASPM.