Basketball-Reference adds BPM, a few questions

[Dr Spaceman]

http://www.sports-reference.com/blog/20 ... nus-bpm-2/

Noticed today there are a few new stats in the 'Advanced' section of player pages. Box (+/-), splits for offense and defense, and Value over replacement player. I have a few questions:

1. It seems like there are a litany of box score-informed plus/minus statistics. How does this stat differ from xRAPM, SPM, etc.?
2. How much value would you place in this stat? It seems like another "all-in-one" type effort; Is there any inherent advantage (or disadvantage) using it as opposed to simply looking at RAPM, WS, Box Score, play-by-play etc. separately?
3. Why was this stat, specifically, the first of its kind to show up on Bballref?

[blabla]

It's just another type of SPM. It differs from xRAPM because xRAPM uses lineup data *on top* of the SPM.
It's decent, but I don't understand why they took this step. It offers almost nothing over the already existing ORtg/Drtg except that it's on a different scale

[Doctor MJ]

http://www.sports-reference.com/blog/20 ... nus-bpm-2/

Here's there explanation, from what I see it's just the latest version of SPM. I'm a little annoyed that they gave it a new label given that it is just a form of SPM, and it's not like they need to give it a new label for us to pay attention, but oh well.

My response remains: Eh, it's still so far away from what it should ideally be that I'm not dying to switch over from other stats.

In a nutshell: The issue I have with all SPM can be seen from how heavily steals get weighted given that guys who get a lot of steals typically don't show up as strong in DRAPM. What that tells us then is that the regressions are essentially saying "Guys who try to get steals are way more valuable when they actually succeed, but they fail so much they aren't that valuable."

It's not a fundamental issue with the SPM approach, it's just that there still isn't a good enough box score to make the SPM stat all that worthwhile. And by that I mean: The weights determined by regressing steals without factoring in the unrecorded failed steal attempts produce something more out of whack than what common sense gives us, and such is true in other areas to.

Still, a worthwhile goal, and maybe I'll end up incorporating it as we go along.

[ElGee]

It's a cool addition to analyzing the box score, in the same way ASPM was. There are very real limits on the box score though, and so painting wings on the car might make it look cooler, but it still won't fly. If you were trying to take a swag at the box, I'd definitely have a look at it -- it incorporates calculations that involve the team, and that's always a nice cognitive aid. I also thinks it's nice to have things translated into points and VORP. I'm all for that. So nice job Daniel.

On the other hand, SPM (BPM) is not only limited by the box but in this case, by the RAPM data set. He does some cool and creative stuff to help remove some of these problems, but he's still bound by his dataset. No matter how hard he tries, he'll get some wonky results. Here's a good example I just spotted:

-Magic Johnson's VORP is +7.5 in his 3rd year, and then according to that stat he gets worse and worse until 1986. He finally returns code to his 1982 form in 1987, only to again be worse than he was as a rookie in 1988. By raw BPM, 87 is his 6th-best season. I suppose it's possible this is accurate, but it's an incredibly strange chronology. Additionally, BPM says young Magic was a better defender than peak Jordan. Weird. (Hello "positional" DREB% bias.)

Another example is Steve Nash, always underrated via the box score. BPM says he was better in 02 and 03 than in 05. He then barely sets his peak in 06 and 07...but by 2010, he's no better than he was 2001. Again, I suppose this is possible...it's just fairly strange and somewhat challenging to explain. You'll always have results like this with this kind of method.

[DQuinn1575]

-Magic Johnson's VORP is +7.5 in his 3rd year, and then according to that stat he gets worse and worse until 1986. He finally returns code to his 1982 form in 1987, only to again be worse than he was as a rookie in 1988. By raw BPM, 87 is his 6th-best season. I suppose it's possible this is accurate, but it's an incredibly strange chronology. Additionally, BPM says young Magic was a better defender than peak Jordan. Weird. (Hello "positional" DREB% bias.)

.

I don't see a positional bias built in the stat; am I missing something?

I'm not a big fan of the stat, but I guess no one (yet) has come up with anything better off of box scores to use for history.




Sent from my iPhone using RealGM Forums

[ElGee]

-Magic Johnson's VORP is +7.5 in his 3rd year, and then according to that stat he gets worse and worse until 1986. He finally returns code to his 1982 form in 1987, only to again be worse than he was as a rookie in 1988. By raw BPM, 87 is his 6th-best season. I suppose it's possible this is accurate, but it's an incredibly strange chronology. Additionally, BPM says young Magic was a better defender than peak Jordan. Weird. (Hello "positional" DREB% bias.)

.

I don't see a positional bias built in the stat; am I missing something?

I'm not a big fan of the stat, but I guess no one (yet) has come up with anything better off of box scores to use for history.




Sent from my iPhone using RealGM Forums

No there's no actual positional adjustment at all. I spoke hastily. I was thinking more along the lines of the interaction variable between assists and reb% (which Magic shines in)...which very well may be a valid indicator of something good going on. Then again, I'm also looking at that regression and wondering if it's literally being influenced by the following high ast% / trb% players:

-Grant Hill
-Jason Kidd
-LeBron James

If we opened it up to a wider pool of players -- say, over 9% trb and 25% sat -- the regression would still be influenced by:

Andre Iguodala 1
Antoine Walker 1
Clyde Drexler* 1
Grant Hill 4
Jamal Mashburn 1
Jason Kidd 12
Joakim Noah 1
Kevin Durant 1
Kevin Garnett 2
Kobe Bryant 1
Kyle Lowry 1
LeBron James 10
Manu Ginobili 1
Michael Carter-Williams 1
Michael Jordan* 1
Paul Pierce 2
Rajon Rondo 2
Steve Francis 2
Toni Kukoc 1
Tracy McGrady 2
Tyreke Evans 1
Vince Carter 1
Vlade Divac 1

No doubt some good players, and some RAPM monsters. But I'm always skeptical of interactions that could be influenced by sample size and that people have a hard time explaining.

[E-Balla]

It's just another type of SPM. It differs from xRAPM because xRAPM uses lineup data *on top* of the SPM.
It's decent, but I don't understand why they took this step. It offers almost nothing over the already existing ORtg/Drtg except that it's on a different scale

It measures something completely different than the ORTG and DRTG they had listed. Individual ORTG/DRTG are completely different and aren't found in the same way at all.

[blabla]

Saw that basketball-reference actually replaced ORtg and DRtg with BPM in the "Advanced" tab. With this, it went from a "meh" decision on bbr's part to a downright terrible one

[Dr Spaceman]

Saw that basketball-reference actually replaced ORtg and DRtg with BPM in the "Advanced" tab. With this, it went from a "meh" decision on bbr's part to a downright terrible one

ORTG/DRTG are found under Per 100 Possessions