2015 APM/RAPM/etc.

[SideshowBob]

fpliii has kindly gotten J.E. to post PI numbers for the 2015 season thus far.

http://stats-for-the-nba.appspot.com/ratings/RAPM_2015.html

Spoiler:

J.E. just posted vanilla RAPM for 2014-15 so far:

http://stats-for-the-nba.appspot.com/ra ... _2015.html

EDIT: This is multi-year, not NPI.

Here's the APBRmetrics thread.

EDIT: Note that the possessions refer to Off+Def. So Curry at 5199 refers to the roughly 2600 possessions he had played on each end through 38 games (last night is not included).

[Jaivl]

Awesome. Ricky stylin' and profilin' (minuscule sample size tho).

[Lost92Bricks]

How come KD is always lower than expected?

[SideshowBob]

How come KD is always lower than expected?

The expectations are too high? He doesn't seem lower than expected to me, he's seemed like a 6-6.5ish per 100 player in all of the last three years IMO. Filter out Rubio (only 5 games played) and Leonard's the only odd one out. The difference is Durant can sustain his play once he shifts back up to 38 MPG, and Leonard's not going to be that high, so Durant at +6 in 38 MPG is still more impressive than Leonard at +6.5 in 30 MPG. Also he's sharing the load with Westbrook who himself is right there as well.

[NinjaSheppard]

How come KD is always lower than expected?

The expectations are too high? He doesn't seem lower than expected to me, he's seemed like a 6-6.5ish per 100 player in all of the last three years IMO. Filter out Rubio (only 5 games played) and Leonard's the only odd one out. The difference is Durant can sustain his play once he shifts back up to 38 MPG, and Leonard's not going to be that high, so Durant at +6 in 38 MPG is still more impressive than Leonard at +6.5 in 30 MPG. Also he's sharing the load with Westbrook who himself is right there as well.

I am not so sure Leonard is the odd man out. Haralabos Voulgaris thinks he is a top 10 player in the league and he probably charts more defense than any non-front office person in the league. Leonard isn't the 4th best guy but calling him top 10 might not be a stretch.

[Dr Spaceman]

Since this isn't sortable, here are the top 10 splits for offense and defense

Code: Select all

Offense:                         Defense:
                                               Sanders 5.1
Curry 7.4                                  Green, Draymond 4.8
James 6.9                                 Leonard 4.6
Harden 6.4                                Noah 4.6
Paul 6.0
Durant 6.0                                   Bogut 4.6
Lillard 5.5                                Duncan 4.5
Lowry 5.2                                 Gobert 4.2
Lawson 5.2                               Hibbert 4.2
GOATBrook 5.1                        Garnett 4.2
                                                Splitter 4.0
Anthony 4.3                              Nene 3.8
Conley/Dragic 4.0                     M. Gasol/Asik 3.5             

Don't kill me if I missed someone, this is just from an initial scan

[colts18]

Since this isn't sortable, here are the top 10 splits for offense and defense

Code: Select all

Offense:                         Defense:
Curry 7.4                                  Green, Draymond 4.8
James 6.9                                 Leonard 4.6
Harden 6.4                                Noah 4.6
Paul 6.0                                    Bogut 4.6
Lillard 5.5                                Duncan 4.5
Lowry 5.2                                 Gobert 4.2
Lawson 5.2                               Hibbert 4.2
GOATBrook 5.1                         Splitter 4.0
Anthony 4.3                              Nene 3.8
Conley/Dragic 4.0                     M. Gasol/ Asik 3.5

Don't kill me if I missed someone, this is just from an initial scan

You missed KG. Plus you missed out on Larry Sanders who is #1 in defensive RAPM. KD is missing on offense

Name Defense per 100
Larry Sanders 5.1
Draymond Green 4.8
Kawhi Leonard 4.6
Joakim Noah 4.6
Andrew Bogut 4.6
Tim Duncan 4.5
Rudy Gobert 4.2
Roy Hibbert 4.2
Kevin Garnett 4.2
Tiago Splitter 4

[SideshowBob]

Spoiler:

GSW

Curry +7.7
Iggy +5.0
Draymond +5.4
Bogut +2.8
Klay +2.6
Speights -3.5
Lee -1.2
Barnes -0.6
Livingston -1.4
Barbosa -3.1

Starters really shining. Curry's on pace for the highest recorded raw +/- total ever right now.

[E-Balla]

I'm still more of a fan of NPI RAPM. Still nice to see these published. I honestly think no one stood out or was unexpectedly high outside of Nerlens Noel being a -6.1 on offense (next worst is -4.5). Even Dante Exum being the highest ranked rookie didn't phase me.

[ceiling raiser]

I'm still more of a fan of NPI RAPM. Still nice to see these published. I honestly think no one stood out or was unexpectedly high outside of Nerlens Noel being a -6.1 on offense (next worst is -4.5). Even Dante Exum being the highest ranked rookie didn't phase me.

He offered to post either, so I asked for prior informed RAPM because I wanted to see how it compares with his RPM/xRAPM on ESPN. I'm going to ask him for both prior informed and NPI after the playoffs are finished, just didn't want to be too greedy.

[Ascrilas]

Interesting that Mirotic is seen as a (significant) plus. Even though he's not the classical rookie, that's not an easy feat for a first-year player.

[Doctor MJ]

So a few thoughts:

I'm at a point right now where I just don't have trust in any particular maker of RAPM stats, and it's tough.
When JE says "vanilla RAPM", I swear I've looked at data he's given that label too and seen it not match earlier data I had from him.
So I'm just really hesitant to say much specific.

I will say the first thing I looked at is the Hawks' numbers. By Real Plus Minus, Millsap is seen as the top player, however by any actual +/- metric I can find, it seems like it's been Korver...and now Vanilla RAPM agrees. So in the team that's truly shocked the league this and forced us to really think beyond the basic stats that were available before the datable age, Real Plus Minus' existence makes it harder to glean insight into the intangibles.

But that sounds bitter. Really I'm just fascinated by the Hawks and by Korver. What I feel like we're seeing here is a kind of video game extreme in which you can have a really limited player who is just so damn effective at what he does that you can build a scheme that just spreads the defense too far to really work.

The GB has a thread saying Vegas has put the Hawks as NBA champion favorites. I can't imagine they actually win the thing - and in fairness to Vegas, those odds must have everything to do with how unpredictable the West is - but if it did happen based on the Hawks doing their thing as they are now, it would mean that Korver was the best player on an NBA champion. The mind reels...

Speaking more generally my take is always this:

PI RAPM is the more important thing to have when trying to understand years and years of NBA basketball.
But NPI RAPM of the current season plus old PI RAPM data is the best judge of the MVP race.

LeBron ranking #1 in the PI data here is just a clear case of the prior causing misinformation. It's not a big deal as long as you realize it though.

It weirds me out to see Leonard so high, but I can grant it to be not utterly absurd.

Rubio's high number is basically meaningless. To me his number was inflated last year due to Love's rage quit toward the end of last year which made it unclear by season's end who was the more valuable player by +/- data. You add that in there, plus Love's struggles in Cleveland, and the fact that Rubio's largely not played in the greasier on a Timberwolves season, and Rubio's basically being given the statistical benefit of the doubt. "Simplest explanation: It was Rubio all along." We humans know better though. Rubio is a good player, but he's not, well, what this stat says he is.

I saw some people talk about Durant always being underrated by stats in the +/- favorite. To me that's basically 2 things:

1) He actually has a tendency to be inconsistent. He's not playing his best right now, and so that gets factored in, as is any prior stretches like this.
2) Westbrook. Westbrook. Westbrook. That's what happens when you've got a teammate capable of doing great things with or without you, but who you don't synergies with that well. Those 2 will continue to do great things together, and maybe they win titles, but I think at this point it's probable they'll just keep getting in each other's way when it comes to RAPM stats.

Also, let's note Anthony Davis: His numbers have skyrocketed compared to JE's "pure" RAPM last year. Up from a negative value to clear cut all-star levels despite the fact that he's still being hindered by last year's prior, and this with a stat that has nothing to do with his best-in-game box score numbers. The narrative that he's not actually have huge impact on the court literally has no basis at all now.

[lorak]

I will say the first thing I looked at is the Hawks' numbers. By Real Plus Minus, Millsap is seen as the top player, however by any actual +/- metric I can find, it seems like it's been Korver...and now Vanilla RAPM agrees. So in the team that's truly shocked the league this and forced us to really think beyond the basic stats that were available before the datable age, Real Plus Minus' existence makes it harder to glean insight into the intangibles.

But that sounds bitter. Really I'm just fascinated by the Hawks and by Korver. What I feel like we're seeing here is a kind of video game extreme in which you can have a really limited player who is just so damn effective at what he does that you can build a scheme that just spreads the defense too far to really work.

The GB has a thread saying Vegas has put the Hawks as NBA champion favorites. I can't imagine they actually win the thing - and in fairness to Vegas, those odds must have everything to do with how unpredictable the West is - but if it did happen based on the Hawks doing their thing as they are now, it would mean that Korver was the best player on an NBA champion. The mind reels...

I'm not sure where is the problem with Korver as the best Hawks player. You are big Reggie fan and know how valuable he was, so imagine that Korver is just like Miller - both extremely good shooters and because of that VERY valuable offensive players. In fact Pacers were kind of similar with a lot of good defenders and Reggie as no. 1 offensive player.

PI RAPM is the more important thing to have when trying to understand years and years of NBA basketball.
But NPI RAPM of the current season plus old PI RAPM data is the best judge of the MVP race.

LeBron ranking #1 in the PI data here is just a clear case of the prior causing misinformation. It's not a big deal as long as you realize it though.

I think you are overrating impact of prior years. For example look how good defensively are some young players like Gobert or Nurkic. Or how much Harden improved on defense. I think prior isn't a problem here, it actually helps, because data is more accurate. Look at this that way - NPI has bigger "standard error" than PI and that's why even for single year PI is better.

fpliii,
if you can, please ask JE about NPI right now. He probably has the data if he calculated PI, so for him it's just copy & paste. And for us it would be interesting to compare differences between PI and NPI.

[Doctor MJ]

I will say the first thing I looked at is the Hawks' numbers. By Real Plus Minus, Millsap is seen as the top player, however by any actual +/- metric I can find, it seems like it's been Korver...and now Vanilla RAPM agrees. So in the team that's truly shocked the league this and forced us to really think beyond the basic stats that were available before the datable age, Real Plus Minus' existence makes it harder to glean insight into the intangibles.

But that sounds bitter. Really I'm just fascinated by the Hawks and by Korver. What I feel like we're seeing here is a kind of video game extreme in which you can have a really limited player who is just so damn effective at what he does that you can build a scheme that just spreads the defense too far to really work.

The GB has a thread saying Vegas has put the Hawks as NBA champion favorites. I can't imagine they actually win the thing - and in fairness to Vegas, those odds must have everything to do with how unpredictable the West is - but if it did happen based on the Hawks doing their thing as they are now, it would mean that Korver was the best player on an NBA champion. The mind reels...

I'm not sure where is the problem with Korver as the best Hawks player. You are big Reggie fan and know how valuable he was, so imagine that Korver is just like Miller - both extremely good shooters and because of that VERY valuable offensive players. In fact Pacers were kind of similar with a lot of good defenders and Reggie as no. 1 offensive player.

But Korver is NOT the #1 offensive player by traditional measures.

He's only the 4th leading scorer on his team.
His FGA/36 is 13th on his team.

We're literally talking about a guy who does one thing so well...that he barely does that thing. An offense based not on "You can't stop our best shooter" but rather based on "Our shooter is so good, you simply have to commit to stop him, and when you do you can't cover the rest of our guys well."

Reggie, though he seems unusual for an offensive star, was still your #1 scorer, and hence far more of a normal situation.

What I will say is that Atlanta with Korver can be seen as a descendant of Indiana with Miller in terms of a "How far can we push this?" type thing.

PI RAPM is the more important thing to have when trying to understand years and years of NBA basketball.
But NPI RAPM of the current season plus old PI RAPM data is the best judge of the MVP race.

LeBron ranking #1 in the PI data here is just a clear case of the prior causing misinformation. It's not a big deal as long as you realize it though.

I think you are overrating impact of prior years. For example look how good defensively are some young players like Gobert or Nurkic. Or how much Harden improved on defense. I think prior isn't a problem here, it actually helps, because data is more accurate. Look at this that way - NPI has bigger "standard error" than PI and that's why even for single year PI is better.

Well LeBron is pretty far down the list with Real Plus Minus, which is a stat that include a prior as well, so truthfully, it's hard for me to know what exactly is going on, and I should hedge more in my language to make that clear.

Re: Gobert. Dude played 9 MPG last year, and he played only 21 this year. That means:

1) Just on this year, he's playing low minutes, which means I'm hesitant to say much about his numbers.

2) That last year is basically like a single raindrop in a heavy wind. Hell of a lot easier to erase that than erase LeBron's previous year.

Re: Harden's defense. Harden's improvement is driving him toward the mean, not away from it. That essentially means that his defense is simply being given less weight by the algorithm as his results disagree with each other. Doesn't seem weird to me.

[fuzzy_dunlop]

Well LeBron is pretty far down the list with Real Plus Minus, which is a stat that include a prior as well

This year's version of RPM uses in-season information only.

[Knosh]

Well LeBron is pretty far down the list with Real Plus Minus, which is a stat that include a prior as well, so truthfully, it's hard for me to know what exactly is going on, and I should hedge more in my language to make that clear.

As far as I can tell we are looking at multi-year PI RAPM here. I was under the impression that RPM is only one year? Maybe that explains the difference for Lebron.

Imo it's pretty hard to keep up with all those +/- stuff... I would wish that those stats came with an explanation of what we are actually looking it attached to it. All I find on ESPN on RPM is this for instance:
"RPM is based on Engelmann's xRAPM (Regularized Adjusted Plus-Minus)"
So it's not xRAPM? Just based on it? I was also under the impression that RPM includes a prior. But now fuzzy_dunlop is saying this year's version doesn't? Does that mean last year's version did? Maybe that's a change worth noting somewhere on those RPM pages.

[ceiling raiser]

fpliii,
if you can, please ask JE about NPI right now. He probably has the data if he calculated PI, so for him it's just copy & paste. And for us it would be interesting to compare differences between PI and NPI.

It probably would be easy, though he might have to run the regression over (since the priors are an argument in the function used, or part of the matrix algebra if he computed directly, and NPI is not produced as an intermediate result). Still though, he would just have to remove that argument from his script, or just set the priors vector = 0.

I asked him in the thread just now. Hopefully he responds.

EDIT: Also that link *was* incorrect: http://www.apbr.org/metrics/viewtopic.p ... 827#p22626

Evidently it was multi-season RPM, not RAPM. It should be correct now at the same link according to J.E.

[SideshowBob]

Top 10 Offense

Lebron +5.1
Curry +4.8
Harden +4.1
Paul +3.9
Durant +3.8
Lillard +3.5
Westbrook +3.4
Lawson +3.2
Patterson +3.0
Lowry +2.9

Top 10 Defense

Leonard +4.0
Iggy +3.7
Bledsoe +3.5
Draymond +3.4
Splitter +3.2
Sanders +3.1
Howard +2.9
Nene +2.8
Dudley +2.7
Allen +2.6

[PCProductions]

Top 10 Offense

Lebron +5.1
Curry +4.8
Harden +4.1
Paul +3.9
Durant +3.8
Lillard +3.5
Westbrook +3.4
Lawson +3.2
Patterson +3.0
Lowry +2.9

Top 10 Defense

Leonard +4.0
Iggy +3.7
Bledsoe +3.5
Draymond +3.4
Splitter +3.2
Sanders +3.1
Howard +2.9
Nene +2.8
Dudley +2.7
Allen +2.6

Top 4 defensive players being perimeter players is fascinating to me.

[SideshowBob]

Spoiler:

Top 10 Offense

Lebron +5.1
Curry +4.8
Harden +4.1
Paul +3.9
Durant +3.8
Lillard +3.5
Westbrook +3.4
Lawson +3.2
Patterson +3.0
Lowry +2.9

Top 10 Defense

Leonard +4.0
Iggy +3.7
Bledsoe +3.5
Draymond +3.4
Splitter +3.2
Sanders +3.1
Howard +2.9
Nene +2.8
Dudley +2.7
Allen +2.6

Top 4 defensive players being perimeter players is fascinating to me.

Horizontal game is getting more and more important as the trend towards shooting 3s continues. That and the fact that these long/athletic/mobile wings are uber-useful in shutting down PnR schemes.