The annual variance of Defensive Real +/- allocations to some players is considerable

[Double Helix]

I intend to dig into this a little deeper out of curiosity but it seems to me that the team "impact" that's allocated negatively or positively through regression is least accurate on the defensive end, annually resulting in wild variations between players who one year look like First or Second Team all defence types and absolute pylons defensively another. It's of course possible that these players truly did have wild swings on impact year to year for various reasons but because of the ridge regression on the defensive end I'm more inclined to think the math simply allocated too much good some years, and too much bad in others to various players. I think we all watch enough basketball to know that Middleton didn't drop from being a Draymond-type to bench player defender and that Biyombo didn't go from being a poor man's Ben Wallace last year to simply being solid this year. Something's not right with the regression that's occuring on many of these defensive allocations.

Here are just a few players that I've seen go through some pretty substantial changes positively or negatively within Real +/- on the defensive end over the past 3 seasons. Whether it's going from looking like an elite defender, to a medicore, or worse one, or the exact opposite of that.

Cole Aldrich
Demarcus Cousins
Khris Middleton
Jusuf Nurkic
Darryl Arthur
Tim Mozgov
Robert Covington
CJ Miles
Steven Adams
Bizmak Biyombo
Damian Lillard
Lou Williams
Terrence Jones
Isaiah Thomas
Enes Kanter
Lucas Nogueira
Terrence Ross
Patrick Patterson

And this is just the ones I've visually caught myself visually over the years from players I've also watched regularly enough to wonder what's up. Guys you'll be tempted to label as being among the best defenders at their positions one year flip to become average or worse in others. Guys you think are only slight net negatives become massive negatives. Guys you think are among the worst defenders in the league one year are allocated average the next. We're talking swings of 3.5 or more in some cases. In others it's closer to 1.5-2.0. That's big and can mislead fans and journalists.

Has anyone else noticed this on the defensive end within Real +/- year-to-year? What's your take on this and how much faith should we put into the defensive predictions of Real +/- knowing this? For example, an anlytics-heavy GM who loved Real +/- might have given Biyombo, or Aldrich much bigger contracts in any one of the previous 2 seasons because of their respective Real +/- in comparison to what Real +/- shows from either today. Have those two dudes really changed that much or is it simply the allocation that was credited to them has changed for other reasons? Same for Middleton who's went from being seen as a possible DPOY type of wing that was ranked in the top 10 in the league in 2014/2015 to now 15th at his respective position and 158th in the league. Patrick Patterson in 2014 was seen as one of the worst defenders in the entire league via Real +/- and look where he ranks today. Some swings are wilder than others but they give me pause when I see people making bold assertions on either combined Real +/- (which incorporates the defensive end, which may be wildly incorrect) or simply defensive Real +/-.

Are these wild swings related mostly to the "known" informed priors of these players, their teammates, and the performance of the teams themselves, resulting in some seeing hugely undeserved boosts and others being unjustly punished?

[CptCrunch]

I read a comment somewhere on RealGM that RPM might be overrating low usage players, especially ones on the defensive side. Their usage/mpg scaling doesn't seem exactly right with high efficiency at medium-high usage not being awarded 'as much' as low usage at high efficiency.

And perhaps unsurprisingly, most of your list is composed of PF and C's.

Do you have the actual changes? (Numeric values)

[Double Helix]

I read a comment somewhere on RealGM that RPM might be overrating low usage players, especially ones on the defensive side. Their usage/mpg scaling doesn't seem exactly right with high efficiency at medium-high usage not being awarded 'as much' as low usage at high efficiency.

And perhaps unsurprisingly, most of your list is composed of PF and C's.

Do you have the actual changes? (Numeric values)

I don't yet. Something I'd like to do but those players all moved around year to year. Sometimes dramatically and all on the defensive end.


Sent from my iPhone using RealGM mobile app

[blabla]

Middleton isn't a great example because he has barely played and just came back from a major injury.
Some variance is to be expected, as players obviously don't stay the same over their careers. That said, defense is simply harder to measure than offense, and a lot of it is depending on scheme/coaching/motivation. When there are 450 players playing each season, it's no wonder that there'll be large swings for some of them
I'd expect you'd find something similar in the other metrics, as well. That's not to say the swings are a good thing. You want to "see things coming", if possible, but that's easier said than done