What's new
By:

Keyonte George’s Star Ascension

MTS said:
I guess I need to repeat my previous post: There's a giant difference between "it takes into account" and "it is not impacted by".

The stat doesn't do the latter. This isn't up for debate - they explicitly say "A different role, a different coaching scheme, different teammates, different match-ups, or different seasons affect APM big time."

If you have a problem with that statement, take it up with them.
Click to expand...

The adjustment in APM is because it adjusts for the teammates. I seriously don’t know how to make this more clear. It’s all about performance relative to who a player shares the court with. A more simple example would be a relative TS%. With relative true shooting, you compare true shooting to the league average of that year. Different eras have different TS%. So if you look at a raw TS% it’s biased towards the efficient years. But with relative TS% it doesn’t matter as much because you’re only comparing TS% to the same era.

Likewise, playing with Kevin Love is neutral when it comes to APM. Key’s minutes with Love are compared to other Love minutes. It is not inherently more beneficial to play with Love vs Gobert in APM because the outcome expectations are different.

Here is the description for RAPM which is the variant I have been referencing:

RAPM is a form of "Adjusted Plus Minus" (APM)

APM is an impact-metric analysis that treats players as influencing factors on the outcome of NBA possessions to determine a player's impact on his team's performance in a more unbiased way.

The "R" in RAPM stands for "Regularized", i.e. quadratic penalization for coefficients - a method to help deal with noise and high degrees of multi-collinearity in the data.

In the standard (non-"x") version of (R)APM, no individual player data is used
Instead, a player's impact is estimated solely using the points scored by his team - and the opponent - while he was on the floor, while adjusting for the quality of his teammates and the opponent players
To give a short example, the 2023/24 Grizzlies had an Offensive Rating of 107. An average MEM lineup would be expected to score 107 points in 100 offensive possessions. If a Grizzlies player managed to record a team ORtg of 115 - the team scored 115 points per 100 possessions with him on the floor (with average teammates, against average opponents) - his offensive (R)APM would be positive, though not necessarily equal to (115-107 = +8).
At the same time, a player recording an ORtg of 115 as a member of the Boston Celtics, who had an ORtg of 123 for the season as a team, would lead to a negative offensive (R)APM
Click to expand...

PS. I acknowledged those concerns earlier, but that is not an argument for Love tanking Key’s APM numbers. An argument for that would be something along the lines of…”Hardy sucks at coaching defense, so another coach with the same lineups would have produced better results and those shortcomings are passed on from the coach to the player.” That’s a sound argument, but what you’re arguing does not fall into the cons of this number.
 
If you're going to continue to close your eyes, plug your ears, and pretend that "it takes into account" = "it is not impacted by", then I don't know what more can be gained from this.

At the very least it'd be nice if you acknowledged the disclaimer that the literal creators of the statistic put front and center: Different teammates affect APM big time.
 
MTS said:
If you're going to continue to close your eyes, plug your ears, and pretend that "it takes into account" = "it is not impacted by", then I don't know what more can be gained from this.

At the very least it'd be nice if you acknowledged the disclaimer that the literal creators of the statistic put front and center: Different teammates affect APM big time.
Click to expand...

Different teammates affect APM in the same way that different league averages TS% affect relative TS%. That is to say that accounting for that difference is the entire point.

You are the only person who used the language “is not impacted by”. I never said teammates do not impact APM. I’ve been trying to tell you that the point of APM is to be effected by teammates!

My statement said that APM is not necessarily better or worse because of better or worse teammates. Key’s APM is not bad because he plays with Love because it adjusts for having minutes played with Loce. Again, the whole point is to adjust for that.

What you’re essentially doing is saying that Mitchell’s relative TS% is only bad because the league averages TS% was lower. That makes no sense because you’re using a relative measure! Likewise, saying Key’s APM is bad because he plays with Love doesn’t make a ton of sense because you are looking at +\- in relation to the players he shares the floor with.
 
KqWIN said:
Different teammates affect APM in the same way that different league averages TS% affect relative TS%. That is to say that accounting for that difference is the entire point.
Click to expand...
That is not what they mean. Did you even read your own link? The entire context of the "Different teammates affect APM big time" is that it is a major CON of the stat. Hence why it is listed under "Cons".

KqWIN said:
You are the only person who used the language “is not impacted by”. I never said teammates do not impact APM. I’ve been trying to tell you that the point of APM is to be effected by teammates!
Click to expand...

No. The stat is trying to adjust for the noise that other teammates bring to a player's individual on/off metrics. The creators also acknowledge that the stat doesn't actually fully do that, and why one of its biggest inherent flaws is the fact that different teammates affect it, in their words, "big time".
 
MTS said:
That is not what they mean. Did you even read your own link? The entire context of the "Different teammates affect APM big time" is that it is a major CON of the stat. Hence why it is listed under "Cons".



No. The stat is trying to adjust for the noise that other teammates bring to a player's individual on/off metrics. The creators also acknowledge that the stat doesn't actually fully do that, and why one of its biggest inherent flaws is the fact that different teammates affect it, in their words, "big time".
Click to expand...

In the cons section, it’s referring to the synergistic effects of two players. If you had 5 clones of Rudy Gobert together it would likely not perform as well as 5 Rudy level players of different skillets. If you’re making the argument that playing with Love is uniquely bad for Keyonte and nobody else, whatever. I think Love is just bad whoever he plays with.

How many times can say this. The point is to adjust for the relative strength of teammates and opponents. What are you actually trying to say about Love and Key’s APM? If you are saying Key’s APM is worse because of Love, you are misguided. It does not work that way. Having good or bad teammates does not inherently make APM worse like on/off because everything is relative to how good or bad those teammates are. That’s the purpose.

Like do you not understand the example of how a player with a 115 rating can be positive or negative based on his teammates? It’s clear as day how the adjustment works and if you refuse to acknowledge this you’re either willfully ignorant or so dense it’s not worth discussing. We’re done here.
 
KqWIN said:
How many times can say this. The point is to adjust for the relative strength of teammates and opponents.
Click to expand...

How many times can I say this: "it takes into account" = "it is not impacted by".

You're treating this stat like some magical mystical stat that can completely account for the impact of other teammates on performance and erase it, leaving you only with a number that evaluates that one single player in a vacuum regardless of circumstance, even though the stat creator's left you a big, shiny disclaimer telling you, in the Cons section, that you can't do that.


A recent study by sport analytics professors shows the Adjusted Plus-Minus (APM) statistic used to evaluate the performance of NBA players is sometimes misleading because it does not accurately account for the quality of a player's teammates.

"We find evidence of complementarity effects: The better are your teammates, the better you will look according to adjusted plus-minus," Ehrlich said. "This is interesting because adjusted plus-minus purportedly controls out teammate and opponent effects. However, it does not have the benefit of out-of-sample information about how a player would play when put in other lineups and on other teams."

The results provide strong evidence that regularized adjusted plus minus player productivity measures are not, in fact, "teammate-independent." Rather, we find evidence that lineup-teammate productivity positively influences a given player's real plus minus value. As this result is conditional upon a given player's baseline productivity via player fixed effects and age, we interpret this as a significant and fairly strong complementarity effect that is uncontrolled in adjusted plus minus measures such as real plus minus.
 
I think RAPM tries to quantify a players impact on team defense but its still totally teammate/context dependant. Like if you watch Giannis career curve, its a perfect reflection of his team and how good they have been defensively, not how good he has been individually.

For instance in 2016-2017 Giannis finished the season 11th in SPG, 9th in BPG and 18th in RPG yet his defensive RAPM was pretty much 0.
 
MTS said:
How many times can I say this: "it takes into account" = "it is not impacted by".

You're treating this stat like some magical mystical stat that can completely account for the impact of other teammates on performance and erase it, leaving you only with a number that evaluates that one single player in a vacuum regardless of circumstance, even though the stat creator's left you a big, shiny disclaimer telling you, in the Cons section, that you can't do that.


A recent study by sport analytics professors shows the Adjusted Plus-Minus (APM) statistic used to evaluate the performance of NBA players is sometimes misleading because it does not accurately account for the quality of a player's teammates.

"We find evidence of complementarity effects: The better are your teammates, the better you will look according to adjusted plus-minus," Ehrlich said. "This is interesting because adjusted plus-minus purportedly controls out teammate and opponent effects. However, it does not have the benefit of out-of-sample information about how a player would play when put in other lineups and on other teams."

The results provide strong evidence that regularized adjusted plus minus player productivity measures are not, in fact, "teammate-independent." Rather, we find evidence that lineup-teammate productivity positively influences a given player's real plus minus value. As this result is conditional upon a given player's baseline productivity via player fixed effects and age, we interpret this as a significant and fairly strong complementarity effect that is uncontrolled in adjusted plus minus measures such as real plus minus.
Click to expand...

Self report with that article lol. That was a brilliant study design and great execution that led to this conclusion. I love to draw causation from R2’s of 0.12 or less. The article speaks for itself.

I never said it was perfect or that it was gospel, I simply tried to explain what it is and you refuse to learn. If you don’t trust it, you don’t trust, but stop ignoring the actual purpose of it and what adjustment is going on. I’m done trying to explain things to you.

I’m guilty of locking up this thread, carry on everyone.
 
KqWIN said:
Self report with that article lol. That was a brilliant study design and great execution that led to this conclusion. I love to draw causation from R2’s of 0.12 or less.
Click to expand...
What?

R^2 has nothing to do with causation. It’s a measure of how much variance the regressors explain.

In the study, they're using a fixed-effects design and IV robustness checks to argue causation.
 
MTS said:
What?

R^2 has nothing to do with causation. It’s a measure of how much variance the regressors explain.

In the study, they're using a fixed-effects design and IV robustness checks to argue causation.
Click to expand...

I mean….you're making the argument for me given their study design. Beyond the obvious selection bias you are proving my point by defining what r2 is.

If you or anyone else actually want to discuss why this is a horrible study, PM me. I’m not gonna flood the thread anymore but I am happy to talk about it.
 
Last edited:
You must log in or register to reply here.

Share this page
Back
Top