Stalwart wrote:But don't you agree that impact metrics, by themselves, don't accurately reflect the full extent of a players impact? So, again, there is only so far we can go in isolating and measuring an individual players impact. So it may give you a good idea depending on your perspective. What we can say is, by itself, will give you an inaccurate or incomplete idea.
Stalwart wrote:The same impact metrics that don't accurately capture a players full impact? Why would you do that?
So, I've been avoiding getting into the back and forth, but I was reading this post and saw the questions here. They aren't addressed to me, and I'm surely missing context, but I feel like I should try to put into words how I approach use of these metrics, and of all pieces of evidence, be they other metrics or qualitative data.
I approach basketball similar to how I think about fields in more traditional science:
There is reality, and then there are our models of reality, which we can never expect them to perfectly match reality.
We use the best model we can, and as new evidence comes to mind, we adjust. Sometimes the adjustment just about filling a known lacuna. Sometimes it requires a minor tweaking of a type that we expect to tweak in the details of the model, and sometimes we abandon a model and try to create a new one - albeit likely with many shared characteristics of the model we had before.
Of course, the thing is, calling everything one model isn't quite right. You end up with one ontology, but it's build with many stacking models that feedback into each other. I have a model for the mechanics of neurological structure that I think about when I think about neuroscience, and then I make use of that model when I think about psychological phenomena which then surely informs my neuroscience gaze.
In basketball conversation, similarly, there is the understanding of the overall shape of the player - tendencies, roles, strengths, weaknesses - on a qualitative level, and then there's the forced-choice of player ranking comparisons. While I value the former more, and always try to go back to that, I find that putting myself in the latter ends up enriching the former in a really powerful way.
Taking this as a starting point, the reason why I care about +/- stats is because I care about impact. I say that not to be tautological, but say that I want to be able to quantify impact, and +/- stats are the best way I happen to have found to do it.
Consider the dartboard analogy:
If statistically matching impact is the goal, then I would argue that +/- stats have unique validity (aka accuracy) but relatively weak reliability (aka precision), because they are the stats that go directly based on team vs team scoreboard, which is what determines who wins the game.
Hence, because impact matters, and I don't have better stats for it than those in the +/- family, I use +/- stats. I don't use them exclusively - I use them in conjunction with all the other tools at my disposal - but I do feel a need to use, and a key question for me is always:
What explains the +/- data I'm seeing? Where that explanation isn't simply about allocating credit, but understanding how a team is achieving advantage with the players that are out there working together. It also certainly recognizes that noise is a thing, that that means I need to be cautious in jumping to conclusions, and willing to reconsider the best-I-could-do-at-the-time conclusions I made previously.
Last:
I want to be clear that when I rank players, impact is not typically the last thing I'm looking at. I don't consider raw impact to be the same thing as player achievement or player goodness. What is the case though is that achievement is dependent on impact, and goodness does represent a general capacity for impact, hence impact estimation - which includes +/- stats wherever possible, but always has a holistic mechanism to it - is vital for all of these specific criteria.
