LeBron James

[SideshowBob]

Actually his plus/minus is not so extreme. The reason it looks so high is that Blatt is playing Lebron with other starters, specifically Kevin Love, and putting out horrible bench lineups when Lebron is off court.

Yes but this is inherent, this is always generally the case, with almost every star player. And the fact that there are horrible bench lineups playing out there with no Lebron/Love is exactly the point.

Real adjusted plus-minus, which adjusts for the other players on the court with Lebron, doesn't look so extreme or really even all that great for him this year. Lebron's RAPM is 6.43 this year, 4th in the league. Last year it was 8.78 (2nd in the league) and the year before with Miami it was 9.08 (first in the league). So with proper adjustment Lebron's plus-minus is actually kind of low for him.

RPM is not the same as RAPM. Here is the explanation I posted on RealGM:

------------------------------------------------------------------------------------------

[#1]APM is Simple OLS, exactly as Spaceman described it. Set up every 5on5 matchup, set equal to the scoring margin and solve for each player across the league (I've run it for a few years, its like ~65000 lines of 5on5 matchups). The resulting coefficients (on each player) are the APM values. This needs a very large sample size to say anything of considerable meaning; a single-year APM has large error terms on each coefficient, multi-year (usually 2-year) studies are preferred.

RAPM is essentially the same thing (OLS) with one exception. It introduces what we'll call a "reference matrix", basically each player is given a baseline value, towards which their coefficient will be pulled. I believe this tries to reduce the multicollinearity problem.

In [#2A]vanilla/basic RAPM, every value in the reference matrix is set to 0. The greater the amount of games played, the less weight that reference of 0 has. It is almost the same as APM [#1], but the regression towards 0 in theory reduces the error within a single-season set. It's still fairly volatile, but it's better than APM [#1] is within a single year. There is also [#2B]multi-year RAPM, which just uses a larger number of seasons, with most weight given to the current year and less and less weight given to previous years, reference matrix of 0s.

[#3]Prior-informed RAPM is essentially the best (ITO out-of-sample prediction) version of this family without introducing the box-score. It's built the same way as RAPM, but the reference matrix uses RAPM values from the previous year, instead of all players being set at 0. Again, as the sample size of the season grows, the reference value holds less and less weight. Obviously this only works once we have multiple years of data, in the 1st year, there is no prior. In the 2nd, we have a prior but it is vanilla RAPM [#2A], but by the 3rd year we can use PI RAPM of the previous year to inform the current year.

[#4A]RPM is RAPM, but the reference matrix is made up of SPM values (SPM is again, regression of box-score metrics on a multi-year non-box-score model such as RAPM). There is also [#4B]multi-year RPM, which is the same as multi-year RAPM, except it presumably uses a reference matrix of multi-year SPM values.

There is also [#5]prior informed RPM. Again, same idea as PI RAPM [#3], single-year, reference matrix of prior-year's RPM values.

ESPN, is unfortunately poor at labeling and differentiating between what they put up on their site. Luckily for us, JE, one of the co-creators, often posts on APBRmetrics and has offered clarification.

ESPN's site has RPM for three seasons, 2014, 2015, 2016.

2014 is multi-year RPM [#4B]
2015 is single-year RPM [#4A]
2016 is single-year RPM [#4A]

It's silly, but they've given us 2 different models for 3 years and labelled them all the same.

Multi-year RPM [#4B] for 2015 was provided in an article last year before the playoffs. Other than that, everything they've posted is single-year. I asked Engelmann for multi-year RPM for this year on twitter, and he said that it was in the works. Where it will be posted I can't say, but I'm sure he'll clarify at some point.

------------------------------------------------------------------------------------------

The reason Lebron's RPM [#4A] (which is what's currently +6.43 on ESPN) is low right now is that the SPM prior from the reference matrix is holding most of the weight, which, for the sake of our conversation, is the equivalent of PER/BPM that we see on BBR. Essentially, his low box-score line is holding his RPM [#4A] rating down. As the sample size of 2015 games increases, his strong +/- performance will drive that up, as it will gain weight relative to his box-score line (which should also improve as his shooting normalizes).

ALSO, since I made that post, JE has updated with multi-year RAPM [#2B]. No prizes for guessing who's leading there. To reiterate, this set is data from the last few years, with most weight placed on the current season, and then consecutively less weight on each prior season (probably back to 2013 or 2014), with a reference matrix filled with values of 0 for each player.

 

[Adam]

love that game winner, nice aggressive move and an example of the fact that there is not a shot in the game he cant make

 

[cavsfever]

Yes but this is inherent, this is always generally the case, with almost every star player. And the fact that there are horrible bench lineups playing out there with no Lebron/Love is exactly the point.



RPM is not the same as RAPM. Here is the explanation I posted on RealGM:

------------------------------------------------------------------------------------------

[#1]APM is Simple OLS, exactly as Spaceman described it. Set up every 5on5 matchup, set equal to the scoring margin and solve for each player across the league (I've run it for a few years, its like ~65000 lines of 5on5 matchups). The resulting coefficients (on each player) are the APM values. This needs a very large sample size to say anything of considerable meaning; a single-year APM has large error terms on each coefficient, multi-year (usually 2-year) studies are preferred.

RAPM is essentially the same thing (OLS) with one exception. It introduces what we'll call a "reference matrix", basically each player is given a baseline value, towards which their coefficient will be pulled. I believe this tries to reduce the multicollinearity problem.

In [#2A]vanilla/basic RAPM, every value in the reference matrix is set to 0. The greater the amount of games played, the less weight that reference of 0 has. It is almost the same as APM [#1], but the regression towards 0 in theory reduces the error within a single-season set. It's still fairly volatile, but it's better than APM [#1] is within a single year. There is also [#2B]multi-year RAPM, which just uses a larger number of seasons, with most weight given to the current year and less and less weight given to previous years, reference matrix of 0s.

[#3]Prior-informed RAPM is essentially the best (ITO out-of-sample prediction) version of this family without introducing the box-score. It's built the same way as RAPM, but the reference matrix uses RAPM values from the previous year, instead of all players being set at 0. Again, as the sample size of the season grows, the reference value holds less and less weight. Obviously this only works once we have multiple years of data, in the 1st year, there is no prior. In the 2nd, we have a prior but it is vanilla RAPM [#2A], but by the 3rd year we can use PI RAPM of the previous year to inform the current year.

[#4A]RPM is RAPM, but the reference matrix is made up of SPM values (SPM is again, regression of box-score metrics on a multi-year non-box-score model such as RAPM). There is also [#4B]multi-year RPM, which is the same as multi-year RAPM, except it presumably uses a reference matrix of multi-year SPM values.

There is also [#5]prior informed RPM. Again, same idea as PI RAPM [#3], single-year, reference matrix of prior-year's RPM values.

ESPN, is unfortunately poor at labeling and differentiating between what they put up on their site. Luckily for us, JE, one of the co-creators, often posts on APBRmetrics and has offered clarification.

ESPN's site has RPM for three seasons, 2014, 2015, 2016.

2014 is multi-year RPM [#4B]
2015 is single-year RPM [#4A]
2016 is single-year RPM [#4A]

It's silly, but they've given us 2 different models for 3 years and labelled them all the same.

Multi-year RPM [#4B] for 2015 was provided in an article last year before the playoffs. Other than that, everything they've posted is single-year. I asked Engelmann for multi-year RPM for this year on twitter, and he said that it was in the works. Where it will be posted I can't say, but I'm sure he'll clarify at some point.

------------------------------------------------------------------------------------------

The reason Lebron's RPM [#4A] (which is what's currently +6.43 on ESPN) is low right now is that the SPM prior from the reference matrix is holding most of the weight, which, for the sake of our conversation, is the equivalent of PER/BPM that we see on BBR. Essentially, his low box-score line is holding his RPM [#4A] rating down. As the sample size of 2015 games increases, his strong +/- performance will drive that up, as it will gain weight relative to his box-score line (which should also improve as his shooting normalizes).

ALSO, since I made that post, JE has updated with multi-year RAPM [#2B]. No prizes for guessing who's leading there. To reiterate, this set is data from the last few years, with most weight placed on the current season, and then consecutively less weight on each prior season (probably back to 2013 or 2014), with a reference matrix filled with values of 0 for each player.

Thanks, but just to clarify -- is SPM the same as 'statistical plus minus'?

This is all very interesting but I always thought that the point of both plus-minus and adjusted plus-minus was to *get away* from box score stats and try to build a metric of how a player influences team performance based on purely team performance. I get that there's a major sample size issue for any metric that tries to isolate a single player's correlation with team plus-minus, but it seems like the way to handle that is by weighting by previous season plus-minus, not getting back to individual box score stats which is what you were trying to get away from in the first place. So #3, prior-informed RAPM, seems like the way to go.

 

[cavsfever]

Yes but this is inherent, this is always generally the case, with almost every star player. And the fact that there are horrible bench lineups playing out there with no Lebron/Love is exactly the point.



RPM is not the same as RAPM. Here is the explanation I posted on RealGM:

------------------------------------------------------------------------------------------

[#1]APM is Simple OLS, exactly as Spaceman described it. Set up every 5on5 matchup, set equal to the scoring margin and solve for each player across the league (I've run it for a few years, its like ~65000 lines of 5on5 matchups). The resulting coefficients (on each player) are the APM values. This needs a very large sample size to say anything of considerable meaning; a single-year APM has large error terms on each coefficient, multi-year (usually 2-year) studies are preferred.

RAPM is essentially the same thing (OLS) with one exception. It introduces what we'll call a "reference matrix", basically each player is given a baseline value, towards which their coefficient will be pulled. I believe this tries to reduce the multicollinearity problem.

In [#2A]vanilla/basic RAPM, every value in the reference matrix is set to 0. The greater the amount of games played, the less weight that reference of 0 has. It is almost the same as APM [#1], but the regression towards 0 in theory reduces the error within a single-season set. It's still fairly volatile, but it's better than APM [#1] is within a single year. There is also [#2B]multi-year RAPM, which just uses a larger number of seasons, with most weight given to the current year and less and less weight given to previous years, reference matrix of 0s.

[#3]Prior-informed RAPM is essentially the best (ITO out-of-sample prediction) version of this family without introducing the box-score. It's built the same way as RAPM, but the reference matrix uses RAPM values from the previous year, instead of all players being set at 0. Again, as the sample size of the season grows, the reference value holds less and less weight. Obviously this only works once we have multiple years of data, in the 1st year, there is no prior. In the 2nd, we have a prior but it is vanilla RAPM [#2A], but by the 3rd year we can use PI RAPM of the previous year to inform the current year.

[#4A]RPM is RAPM, but the reference matrix is made up of SPM values (SPM is again, regression of box-score metrics on a multi-year non-box-score model such as RAPM). There is also [#4B]multi-year RPM, which is the same as multi-year RAPM, except it presumably uses a reference matrix of multi-year SPM values.

There is also [#5]prior informed RPM. Again, same idea as PI RAPM [#3], single-year, reference matrix of prior-year's RPM values.

ESPN, is unfortunately poor at labeling and differentiating between what they put up on their site. Luckily for us, JE, one of the co-creators, often posts on APBRmetrics and has offered clarification.

ESPN's site has RPM for three seasons, 2014, 2015, 2016.

2014 is multi-year RPM [#4B]
2015 is single-year RPM [#4A]
2016 is single-year RPM [#4A]

It's silly, but they've given us 2 different models for 3 years and labelled them all the same.

Multi-year RPM [#4B] for 2015 was provided in an article last year before the playoffs. Other than that, everything they've posted is single-year. I asked Engelmann for multi-year RPM for this year on twitter, and he said that it was in the works. Where it will be posted I can't say, but I'm sure he'll clarify at some point.

------------------------------------------------------------------------------------------

The reason Lebron's RPM [#4A] (which is what's currently +6.43 on ESPN) is low right now is that the SPM prior from the reference matrix is holding most of the weight, which, for the sake of our conversation, is the equivalent of PER/BPM that we see on BBR. Essentially, his low box-score line is holding his RPM [#4A] rating down. As the sample size of 2015 games increases, his strong +/- performance will drive that up, as it will gain weight relative to his box-score line (which should also improve as his shooting normalizes).

ALSO, since I made that post, JE has updated with multi-year RAPM [#2B]. No prizes for guessing who's leading there. To reiterate, this set is data from the last few years, with most weight placed on the current season, and then consecutively less weight on each prior season (probably back to 2013 or 2014), with a reference matrix filled with values of 0 for each player.

Thanks, but just to clarify -- is SPM the same as 'statistical plus minus'?

This is all very interesting but I always thought that the point of both plus-minus and adjusted plus-minus was to *get away* from box score stats and try to build a metric of how a player influences team performance based on purely team performance. I get that there's a major sample size issue for any metric that tries to isolate a single player's correlation with team plus-minus, but it seems like the way to handle that is by weighting by previous season plus-minus, not getting back to individual box score stats which is what you were trying to get away from in the first place. So #3, prior-informed RAPM, seems like the way to go.

 

[Chris Parker]

that's an issue and why RAPM in season is pretty skronky, signal to noise-wise

 

[SideshowBob]

Thanks, but just to clarify -- is SPM the same as 'statistical plus minus'?

Correct. And SPM can be modeled in a variety of ways as well; it's just a general term for regression of box-score stats on an APM/RAPM model. For example BPM from Basketball-Reference is a type of SPM. RPM incorporates an SPM that was created specifically for it, and adds variables beyond the box-score, such as age and height.

By adding this into the RAPM model [to create RPM], it adds a degree of stability within a single-season sample (but not as soon as 15 games into the season), and allows us to make use of it (only as a tool, nothing definitive).

This is all very interesting but I always thought that the point of both plus-minus and adjusted plus-minus was to *get away* from box score stats and try to build a metric of how a player influences team performance based on purely team performance. I get that there's a major sample size issue for any metric that tries to isolate a single player's correlation with team plus-minus, but it seems like the way to handle that is by weighting by previous season plus-minus, not getting back to individual box score stats which is what you were trying to get away from in the first place. So #3, prior-informed RAPM, seems like the way to go.

The bolded precisely mirrors my own thoughts. I don't want to dismiss any new tools, but yeah, the whole point was to get away from the box-score.

The creators of these stats though, are typically just trying to outperform whatever previous iteration of the stat in terms of explaining and predicting the season, thus they are understandably just willing to tweak them by any means necessary to do that. Fortunately, Engelmann (RPM&RAPM guy) seems to be at least willing to provide as many models as he can. While we haven't seen prior-informed RAPM [#3] in a while, he has posted multi-year RAPM [#2B] (the link at the end of my previous post), and that, in theory, should produce results most similar to [#3].

 

[SideshowBob]

Couple other posts on RPM:

RPM is a regression based model that uses SPM as a prior, it is similar to other box&+/- hybrid metrics such as IPV, but it is not a blend in the same manner. This kind of stat needs a considerably larger sample size for stability; I wish ESPN had held out, just for the sake of irrational/ignorant responses that we'll now see.

RPM (in this case) is meant to say the following:

"All else held equal, if [Player X] is in a lineup, the lineup's performance (MOV per 100 possesions) is expected to change by [Rating] per 100 possessions."

It is NOT a catch-all player rater.

-----------------------------------------------------------------------------------------------------------------------

A valid statement I can make based on RPM:

"If I put Steph Curry on any random team, I expect the team performance (MOV), while Steph Curry is on the court, to improve by 8.85 points per 100 possessions, given the the league-wide lineup data we have for the first 3 weeks of the 2016 season."

An invalid statement based on RPM:

"Kyle Lowry is the 3rd best player in the league."

Also, again to reiterate it is role based. Its what Sideshow said, with the caveat of "if I put Steph Curry on a random team to play the exact same role he's in" they'd improve by X. I wouldn't swap Curry for Dwight Howard and expect the same effect (honestly thinking I don't know if Curry works on Houston, which is bizarre to say but I don't think a Curry/Harden backcourt can exist).

So keep in mind, role is not something that is controlled for, we must account for that ourselves.

Also, something to keep in mind. The goal of a stat like this is to reduce error on a macro-level. That is to say, it is not trying to simply gauge each individual as accurately as possible; rather it wants to gauge everyone simultaneously as accurately as possible.

A good analyst will almost certainly do a better job of rating/grading an individual player, or 2, or 10, or even 20. But the larger that number grows, the more accurate RPM becomes and the less accurate the person becomes. When focusing on a single player, an analyst has access to terabytes of lineup/box-score/team data, game footage, team schemes/buildup, etc. RPM does not use nuance in looking at individuals. On the other hand, a single analyst simply does not have the resources to be able to breakdown all 400+ players in the league at once; its simply too many variables to consider in any meaningful amount of time - whereas something like RPM can provide (in a sense) a "best-fit" for all the players at once.

This is always something to keep in mind IMO - too many folks will just take an extreme side, one way or the other.

 

[nahmean]

this is fucking cool
View: https://www.youtube.com/watch?v=mKVzzZH9iAM&feature=youtu.be

 

[Zeus69]

this is fucking cool
https://streamable.com/ln0x

Good stuff. I like how JR was ready on the perimeter; just in case...

 

[priceFTW]

It's sometimes astounding how many times James can be fouled without any call. Just because he's built like Suoerman doesn't mean it's not a foul. Through his career, I'm willing to bet that had he been a skinnier, weaker player, he would have generated about 2 to 3 if not 4 more FTs a game. If that were the case, his advanced stats (BPM, VORP, TS%, RAPM, PER, WS/48) which already empirically place him at top 2 all-time would be that much better.

This year, he is averaging the fewest FTs per FG attempt since his first two years in the league. He is averaging almost 9 drives per game compared to 7.6 in his last year in Miami when his FT rate was .432, 23rd in the league. This year, his free-throw rate is .392, 55th in the league. Here are some of the players who are getting more FT attempts per FG attempt (some it is understandable as they don't take many shots and get fouled inside, but look at some of those players:


Sessions
Lou Williams
Nene
Derrick Williams
Alex Len
Gallinari
Rubio
Hickson
Kevin Martin
Jerian Grant
Jerami Grant
Jakarr Sampson
Covington
Teague

Look at these egregious non-calls.

Here's the vicious no regard for human life dunk on which Garnett shoves him with two hands:

Here's Ibaka breaking his nose without a call (and a reach in by Westbrook before that):


Lastly, the absolute bullshit refereeing we got at Golden State in games 1 and 2. I got pissed watching these again.

1:00 Iggy grabs LBJ with 2 hands as he goes to the rim
1:37 Iggy foul winds up as a "shot clock violation"
1:53 shove by Iggy which leads to the jump-ball
2:00 Draymond pulls him down by the shoulder on the jump-ball

 

[scopedog]

snip

And to make matters worse, LBJ was openly mocked by the Warriors for complaining about foul calls and of course there was that thing with Bogut telling everyone LeBron slammed himself head first into a fucking camera on purpose.

One reason I don't like them is because they have no appreciation or respect for their opponent. They were poor sports about losing 2 games to us and even after winning they showed absolutely no class. That's why I don't pity them all the trash talk about how they are the luckiest team in history. Nobody can deny they are good and that they won, but it also can't be denied that every team they faced was completely hamstrung.

Anyway the whole LeBron foul situation has become awkward and frustrating. On one hand he's the unguardable monster at times, and he also complains more and more frequently, but all his detractors just chalk it up to him whining without admitting he probably gets fouled harder and more often than anyone in fucking NBA history and gets no respect for it. I can't blame him for throwing fits to be honest, though it would obviously be better to just let it go and focus on what you can do rather than what you can't change.

 

[Flight479]

I think the lack of respect has caused LeBron to reevaluate his status in the league. Refs don't treat him like a superstar. They haven't for a while but now they ignore him as if he's some watered down star. LeBron seems to be saying I'm still here, I'm still great, give me my calls. You blow air around Curry and he's at the line. They did this with KD too. Ridiculous.

 

[cavsfever]

Blatt or somebody in team management should do a Mark Cuban and speak up, with video evidence, about the double standard. Do it going into the playoffs. Sure he'll be fined but it will be worth it, that kind of thing can work.

 

[Sticker114]

It's sometimes astounding how many times James can be fouled without any call. Just because he's built like Suoerman doesn't mean it's not a foul.

To your point Price, about two weeks ago I had the pleasure of meeting a med tech that works at LeBron's cryotherapy center near his Bath home. Told me it was absolutely unbelievable how much abuse his body takes each game, with a lot of it being in his upper torso and arms. I guess his recovery routine is absolutely crazy

 

[Pioneer10]

Lebron's defensive Synergy numbers are eye-popping:

Isolation: 69% percentile
PnR defense on the ballhandler: 99% percentile
PnR defense on the roll man: didn't qualify
Post Up: 98% percentile
Spot Up: 97.5% percentile
Hand Off: 79% percentille
Off Screen: 69% percentile

Lebron's basically in the 70% percentile or higher. You might as well not bother attacking him on the PnR or posting him up as he just destroys those plays.

Comparison
Butler
Isolation: 19% percentile
Ball Handler: 15% percentile
Roll man: didn't qualify
Post Up: 6 % percentile
Hand Off: 64%
Off Screen: 39% percentile

Leonard
Isolation: 65% percentile
Ball Handler: 75% percentile
Roll Man: didn't qualify
Post Up: 90% percentile
Spot up: 67% percentile
Hand Off: none
Off Screen: 33% percentile

It's basically no contest: the only argument maybe is that we don't use Lebron against the opponents best till the 4th quarter? Don't watch SAS and Chicago to know if this is true. Someone post this on RealGm as well as they really nuthug Butler and Leonard while dismmissing Lebron's DPOY numbers