Quality of defense played & value in adjusted defensive +/-

[rebirthoftheM]

Might be coming from a place of ignorance so please be nice

Adjusted defensive +/_ sort through different teams lineups viz-a-viz opponent offensive production to best isolate a player's individual defensive *value*.

But does it holistically capture the quality of defense a player displays on the court? His actual value tied to his actual defense?

1) Consider the situation of Player X, a wing guarding a corner 3 player on 1 side of the court. On the other side of the court, Player Y is going 1 on 1 against Player X's teammate, Z. Player Y scores.

In my understanding, the data from this made FG by Player Y would *inform* the adjusted D +/_ for Player X (and Z), though it will of course feed into an adjusted total calculation.

But should this data be at all relevant to assessing what defense Player X played & what good/bad value he brought? After all, he was simply not involved in that defensive possession at all beyond standing near his own man on the other side of the court, a feat all players can do with ease. He neither played good or bad defense on that posession. He neither provided good or bad defensive value. His defense was simply irrelevant in that possession due to his location & yet the FG would seemingly inform his adjusted +/_.

Is this an issue of junk data that should be excluded from analysis?


2) Consider player X is mostly tasked with guarding weaker perimeter players whose offensive value is low & their abiliy to pressure defenses low. Such players tend to not play a meaningful offensive pressure role for their teams.

Consider too that Player X generally reasonably helps out on D when his location on the court calls for it. He is not agent of defensive breakdowns

However, due to the lack of pressure & activity of who he is guarding & their relative irrelevance to their team's offense, Player X is often out of the action location wise when it comes defense. He isn't engaged much defensively. Most of the made FG's made & FTA attempts while on court have very little do with his actual defense, good or bad. Yet all this data would seemingly inform his adjusted D +/-.

In the posessions where he is not involved (including not being responsible for any defensive breakdowns) he neither plays meaningfully good or bad defense. And neither provides meaningful good or bad defensive value. His defense on such posessions once again is just irrelevant since he is not meaningfully engaged in the action on FGA's and FTA's by opponents.

Is this an issue of junk data that should be excluded from analysis?


My overall q is: Do adjusted defensive +/- accurately capture the defensive quality & actual value of the inconsequential defender? Or is the data on them half junk? And this is not wholly inaccurate since some examples over the years come to mind.

[Owly]

Also not an expert.

I think what you're saying is ... does/doesn't the impact family of stats come with a bunch of noise. And the answer is yes. That might not be enough to make RAPM type stuff the least worst defensive measure, especially over a larger sample.

I would argue back a little on "In the possessions where he is not involved" ... I would imagine if I or presumably just about anyone here were guarding an NBA player, the ball would quickly go to "my" man. And they would score. I would compromise the defense. "Not involved" then ... I suppose if there were multiple of us out there there would be plays on which we would concede points in which an individual might not be culpable. But offenses can attack weak links so "not involved" plays would typically indicate a level of competence. A good defense might stop the ball moving, cause a bad shot and most of them be perceived to be "not involved."

Say we imagine watching a game of any semi-competent players play 5(O)-4(D) ... from that, for me, it's hard to imagine a scenario where a players "defense was simply irrelevant in that possession" on a significant scale (maybe in transition off a guard getting their pocket picked at halfcourt).

There isn't going to be one perfect number. We don't know what instructions coaches have given and players probably have to mostly follow those instructions. A player doesn't control whether an opponent happens to make their free throws when he's on (or off). There's noise from the other players and luck etc ... but ... there's chance for the signal to get through. The 76ers were way better on D (and worse on O) when Manute Bol was on the court. That's an extreme example from an extreme player. Over time though if your teams keep doing bad and/or worse than otherwise defensively you're probably not helping defenses and the inverse would also apply. Then too people have always favored players on good teams. This is just much more closely aligning the credit with those who were on court for the periods of goodness (being more likely to have caused it).

[Doctor MJ]

Might be coming from a place of ignorance so please be nice

Adjusted defensive +/_ sort through different teams lineups viz-a-viz opponent offensive production to best isolate a player's individual defensive *value*.

But does it holistically capture the quality of defense a player displays on the court? His actual value tied to his actual defense?

1) Consider the situation of Player X, a wing guarding a corner 3 player on 1 side of the court. On the other side of the court, Player Y is going 1 on 1 against Player X's teammate, Z. Player Y scores.

In my understanding, the data from this made FG by Player Y would *inform* the adjusted D +/_ for Player X (and Z), though it will of course feed into an adjusted total calculation.

But should this data be at all relevant to assessing what defense Player X played & what good/bad value he brought? After all, he was simply not involved in that defensive possession at all beyond standing near his own man on the other side of the court, a feat all players can do with ease. He neither played good or bad defense on that posession. He neither provided good or bad defensive value. His defense was simply irrelevant in that possession due to his location & yet the FG would seemingly inform his adjusted +/_.

Is this an issue of junk data that should be excluded from analysis?


2) Consider player X is mostly tasked with guarding weaker perimeter players whose offensive value is low & their abiliy to pressure defenses low. Such players tend to not play a meaningful offensive pressure role for their teams.

Consider too that Player X generally reasonably helps out on D when his location on the court calls for it. He is not agent of defensive breakdowns

However, due to the lack of pressure & activity of who he is guarding & their relative irrelevance to their team's offense, Player X is often out of the action location wise when it comes defense. He isn't engaged much defensively. Most of the made FG's made & FTA attempts while on court have very little do with his actual defense, good or bad. Yet all this data would seemingly inform his adjusted D +/-.

In the posessions where he is not involved (including not being responsible for any defensive breakdowns) he neither plays meaningfully good or bad defense. And neither provides meaningful good or bad defensive value. His defense on such posessions once again is just irrelevant since he is not meaningfully engaged in the action on FGA's and FTA's by opponents.

Is this an issue of junk data that should be excluded from analysis?


My overall q is: Do adjusted defensive +/- accurately capture the defensive quality & actual value of the inconsequential defender? Or is the data on them half junk? And this is not wholly inaccurate since some examples over the years come to mind.

I tend to talk about +/- as a stand with superior validity to box score stat (with worse reliability).

Worse reliability means more noise, which means you need more sample.
With more sample, the noise decreases, and you get get a more and more accurate estimation of impact.

Or to put another way: Those possessions where a player's play didn't actually impact the results are effectively random results with regards to that player, and so with sample, random data tends to cancel out.

In theory you can do a better job by identifying possessions where the player is less involved and discounting or disconnecting the player from that possession. The biggest picture issue there is in how to identify this in a way that isn't arbitrary. I think with sufficient player tracking we could do something based on the average distance of the player from the ball, and so a player gets more credit/blame when he's more involved in the play.

But of course, the whole "It wasn't my fault, I was hiding as far away as possible form the ball!" thing is certainly not something we want to be incentivizing by our metric. It's just really tricky, and this relates to why I always think it's important to have a final step to your analysis that involves interpolating the disparate elements of data through your own thought rather than through an algorithm that can be automated.

[An Unbiased Fan]

+/- stats were never meant for player comparisons. It wasn't at first, but then a few start to use it in that way to promote themselves into jobs. Defensive +/- really tells you little other than the type of lineups players were in

For example, "big" lineups will do better on defense, and small lineups will be better on offense.