2016 RAPM/RPM/etc. *Full 2016 RS + PS RPM & RAPM Updated 6/24*

[Knosh]

[tweet]https://twitter.com/jerryengelmann/status/670008697767264256[/tweet]

That's more like it.

I still just don't get what JE says about Deandre rating so high in RPM because of his individual stats resembling a good defender, it's not even true unless they weight volume rebounds and blocks more than anything else, in which case why even use RPM instead of PER or whichever straight boxscore stat. His defensive +/- is awful, his rebounding % this year is awful, specially his on/off DRB%, and he's not even getting close to as many volume rebounds or steals as last year. He's just getting more per-possession blocks, is that enough to make your RPM jump like that? if so, it's pretty useless a stat ain't it?

Yes, he is indeed talking about his volume blocks and rebounds. No, the fact that RPM doesn't completely eliminate all of the issues of boxscore stats isn't a good reason to not use it.

His rebounding% isn't awful. He leads the league. I doubt that defensive +/- or on/off DRB% are part of the prior.

[QRich3]

He does not lead the league in DRB%, he's a bit under last year's % actually, when his DRPM was considerably lower. And I meant the team's DRB% mostly, which is awful when he's on the court (and a tad less awful when he's out).

This is not RPM not eliminating the boxscore issues, it's a case of one single boxscore stat (BLK%) shooting up a player's numbers to the top of the league, even though everything else indicates a regression on his part.

[Dr Spaceman]

[tweet]https://twitter.com/jerryengelmann/status/670008697767264256[/tweet]

That's more like it.

I still just don't get what JE says about Deandre rating so high in RPM because of his individual stats resembling a good defender, it's not even true unless they weight volume rebounds and blocks more than anything else, in which case why even use RPM instead of PER or whichever straight boxscore stat. His defensive +/- is awful, his rebounding % this year is awful, specially his on/off DRB%, and he's not even getting close to as many volume rebounds or steals as last year. He's just getting more per-possession blocks, is that enough to make your RPM jump like that? if so, it's pretty useless a stat ain't it?

Statistical Plus Minus (exists in various forms including BPM on bball ref) is used as the prior in RPM. This means that with minute samples like 3 weeks, you're basically going to see the numbers default to what the prior is telling it because the actual plus minus data is pretty much all noise. By contrast, RAPM uses previous seasons' RAPM as a prior, and so the values this year won't be all that different from last year.

As the season goes on, the current results will outweigh the prior more and more until the influence of the prior is pretty negligible.

[Knosh]

He does not lead the league in DRB%, he's a bit under last year's % actually, when his DRPM was considerably lower. And I meant the team's DRB% mostly, which is awful when he's on the court (and a tad less awful when he's out).

This is not RPM not eliminating the boxscore issues, it's a case of one single boxscore stat (BLK%) shooting up a player's numbers to the top of the league, even though everything else indicates a regression on his part.

Yeah idk what I was looking at earlier. He is 2nd in DRB%, 3rd in total RB%:
http://www.basketball-reference.com/leagues/NBA_2016_leaders.html?lid=header_seasons

My point remains: His individual rebounding numbers are still very good.

And as I said, I don't think that team rebounding and on/off are part of the statistical prior used for RPM. Engelmann is saying that his (individual) rebounding and his blocks mean that he is rated very highly by the SPM used.

The actual SPM used is not publicly available afaik, but we can look at BPM from basketball-reference for example:

Code: Select all

Player           DBPM 
DeAndre Jordan    3.7
Josh Smith        3
Blake Griffin     0.2
Pablo Prigioni    0.2
Luc Mbah a Moute  0
Lance Stephenson -0.6
Paul Pierce      -1
Wesley Johnson   -1.2
Austin Rivers    -1.4
Chris Paul       -1.9
J.J. Redick      -3.6
Jamal Crawford   -3.9
Cole Aldrich     -5.1


Assuming RPM is using something that looks similar to this, it a priori assumes that Jordan is a very good defender and that all of his teammates (outside of Smith) are pretty bad. Then +/- data is used to update those ratings. In the past Jordan hasn't looked like a great defender based on +/- data, so you would expect his rating to go down, but it basically needs enough +/- evidence that shows Jordan isn't a great defender to overrule the initial SPM assumption that he actually is a great defender. Based on how RPM rated Jordan last season with a full season of data, I think this is mostly a sample size issue.

[QRich3]

Cool, thanks for the explanation guys.

I still don't really get where they take some things from, let's compare him with Whiteside for instance, who has a DRB% just slightly worse than Deandre (29.3% to 32.1%) but has almost twice the BLK% (12.5% to 7.3%), and his bball-ref DBPM is much superior to Jordan's too (5.5 to 3.7), yet RPM has Jordan as a superior player and it's not particularly close either. I'm guessing Whiteside's teammates having superior SPM makes the supposition that Jordan is more important to his team than Whiteside, but then that raises the question of how reliable is that SPM in the first place.

I guess my main question would be how much different has to be the SPM they use instead, and how do they get there. And with results like these, how useful is it, even in bigger sample sizes.

[Dr Spaceman]

Cool, thanks for the explanation guys.

I still don't really get where they take some things from, let's compare him with Whiteside for instance, who has a DRB% just slightly worse than Deandre (29.3% to 32.1%) but has almost twice the BLK% (12.5% to 7.3%), and his bball-ref DBPM is much superior to Jordan's too (5.5 to 3.7), yet RPM has Jordan as a superior player and it's not particularly close either. I'm guessing Whiteside's teammates having superior SPM makes the supposition that Jordan is more important to his team than Whiteside, but then that raises the question of how reliable is that SPM in the first place.

I guess my main question would be how much different has to be the SPM they use instead, and how do they get there. And with results like these, how useful is it, even in bigger sample sizes.

At this point the best answer we can give is "we don't really know". As Knosh mentioned the SPM prior used is not BPM and is also not publicly available, but we do know that plus minus models have a notoriously difficult time parsing the offense/defense split, and in fact BPM punts on this problem by saying that DBPM is just the difference between total BPM and OBPM - ie. Blocks and steals and rebounds are not factored into the calculation for DBPM at all, counterintuitively. So it's possible Jordan's huge offensive impact is being bled into his defensive stats due to the prior assuming he is a beast defender.

Further it's possible that due to randomness Jordan just has a super inflated plus minus right now, due to lineups that perform well or whatever. Again RPM is going to be biased toward attributing this to his defense due to the prior.

Remember- RAPM is pretty agnostic toward offense/defense, the only actual data point used in the calculation is scoring margin. It's something else that tells the model "x fraction of this is offense, y fraction is defense" which again is a kind of black box but definitely includes things like height adjustments etc.

[Knosh]

in fact BPM punts on this problem by saying that DBPM is just the difference between total BPM and OBPM - ie. Blocks and steals and rebounds are not factored into the calculation for DBPM at all, counterintuitively.

I guess you are getting this from the explantion here: http://www.basketball-reference.com/about/bpm.html

But if you look at the actual model for BPM:

The actual coefficients for Box Plus/Minus are as follows:

Raw BPM = a*ReMPG + b*ORB% + c*DRB% + d*STL% + e*BLK% + f*AST% - g*USG%*TO% +
h*USG%*(1-TO%)*[2*(TS% - TmTS%) + i*AST% + j*(3PAr - Lg3PAr) - k] + l*sqrt(AST%*TRB%)

And the model for OBPM:

Here are the tables of coefficients:

Offensive BPM

Raw O/D BPM = a*ReMPG + b*ORB% + c*DRB% + d*STL% + e*BLK% + f*AST% - g*USG%*TO% +
h*USG%*(1-TO%)*[2*(TS% - TmTS%) + i*AST% + j*(3PAr - Lg3PAr) - k] + l*sqrt(AST%*TRB%)

So it looks like both formulas have the exact same explanatory variables, the only difference are the coefficients. So while DPBM is technically just the difference and I'm sure that's how they calculate it for efficiency reasons, the implicitly have a formula for DPBM and could calculate DPBM and define OPBM = BPM - DPBM instead.

For example the coefficient for BLK% in the formula for DBPM would be 0.584314.

[QRich3]

Thanks guys, that helps a lot in painting the picture of what it does for a guy with no math background like me.

At this point the best answer we can give is "we don't really know". As Knosh mentioned the SPM prior used is not BPM and is also not publicly available, but we do know that plus minus models have a notoriously difficult time parsing the offense/defense split, and in fact BPM punts on this problem by saying that DBPM is just the difference between total BPM and OBPM - ie. Blocks and steals and rebounds are not factored into the calculation for DBPM at all, counterintuitively. So it's possible Jordan's huge offensive impact is being bled into his defensive stats due to the prior assuming he is a beast defender.

Further it's possible that due to randomness Jordan just has a super inflated plus minus right now, due to lineups that perform well or whatever. Again RPM is going to be biased toward attributing this to his defense due to the prior.

Remember- RAPM is pretty agnostic toward offense/defense, the only actual data point used in the calculation is scoring margin. It's something else that tells the model "x fraction of this is offense, y fraction is defense" which again is a kind of black box but definitely includes things like height adjustments etc.

Yeah, it has to be some of his offense leaking into the defensive numbers, and then his blocks and rebounds help make the wrong assumption. With regards to raw +/-, that was one of the things that concerned me, his raw +/- numbers are not even good, he's the worst out of the 5 players that play the most minutes in the Clippers.

[SideshowBob]

By contrast, RAPM uses previous seasons' RAPM as a prior, and so the values this year won't be all that different from last year.

As the season goes on, the current results will outweigh the prior more and more until the influence of the prior is pretty negligible.

Just a note, this is multi-year, not prior-informed.

Multi-year: Ridge-regression, reference matrix of 0s, 3-4 years of data with less weight on each prior year.

Prior-informed: Ridge-regression, reference matrix of prior-year values.

Your bolded point still holds, but the model itself is slightly different.

[trex_8063]

He does not lead the league in DRB%, he's a bit under last year's % actually, when his DRPM was considerably lower. And I meant the team's DRB% mostly, which is awful when he's on the court (and a tad less awful when he's out).

This is not RPM not eliminating the boxscore issues, it's a case of one single boxscore stat (BLK%) shooting up a player's numbers to the top of the league, even though everything else indicates a regression on his part.

Where are you getting this information? The bolded/colored portion above in particular doesn't appear at all accurate according to the data on bbref:

Clippers DREB% with Jordan on the court: 74.5%
Clippers DREB% with Jordan off the court: 69.6%

Clippers OREB% with Jordan on the court: 21.9%
Clipper OREB% with Jordan off the court: 20.3%

Even with him on the court their TREB% is still only 49.1% (so yeah, not good); but it's an abysmal 42.7% when he sits. So while it's true they're not a good rebounding team so far this year, it's equally clear that they do get a big lift from him.

[QRich3]

heh I could swear it was 3% lower when I looked it a few days ago, which would speak of how random this small a sample size is, and kind of make part of my point moot.

[Doctor MJ]

Re: limitations on Harden's game. Yes and no. Harden has issues and shouldn't be above criticism. Additionally, I find myself laughing at the Players' Awards. They make this big stink about player's doing a better job than reporter's in evaluating players, and then they immediately side against Curry in favor of Harden. I'm a Harden fan, but to me it was no contest.

I don't see how Harden playing like crap this year somehow invalidates his legitimacy as an MVP candidate last year.

It's really more about Curry.

And I understand you could argue, "Why should Curry getting better this year matter to who was better last year?", but while I didn't predict this in particular for Curry, central to my entire argument last year was Curry's game scaled in a way that we couldn't expect of Harden's - at least the way Houston was using Harden. Any take this year saying "Who knew Curry could score more if it was so desired?" has me rolling my eyes because I thought that was pretty damn clear last year.

Put another way: Harden won that award because he scored a lot of points with a narrative that let people say "And he's doing it all by himself.", and I always took exception to that way of thinking.

[Doctor MJ]

He does not lead the league in DRB%, he's a bit under last year's % actually, when his DRPM was considerably lower. And I meant the team's DRB% mostly, which is awful when he's on the court (and a tad less awful when he's out).

This is not RPM not eliminating the boxscore issues, it's a case of one single boxscore stat (BLK%) shooting up a player's numbers to the top of the league, even though everything else indicates a regression on his part.

Where are you getting this information? The bolded/colored portion above in particular doesn't appear at all accurate according to the data on bbref:

Clippers DREB% with Jordan on the court: 74.5%
Clippers DREB% with Jordan off the court: 69.6%

Clippers OREB% with Jordan on the court: 21.9%
Clipper OREB% with Jordan off the court: 20.3%

Even with him on the court their TREB% is still only 49.1% (so yeah, not good); but it's an abysmal 42.7% when he sits. So while it's true they're not a good rebounding team so far this year, it's equally clear that they do get a big lift from him.

I'm not going to get into the specifics of these numbers, but I was pretty vocal last year emphasizing the issues with defensive rebounding statistics particularly with Jordan. You can check out details here:

http://www.gotbuckets.com/2015-ffapm-defense/

Basically according to this data, Jordan actually had a slightly negative impact on defensive rebounding. To be clear, the important point to me wasn't that Jordan had a less than 0 impact there because a positive impact certainly wouldn't have surprised me, it's simply that when you see guys chase boards as hard as Jordan did, it's pretty much a given in my experience that there defensive rebounding gets overrated because that's not actually optimal strategy for team defensive rebounding most of the time.

[spearsy23]

Re: limitations on Harden's game. Yes and no. Harden has issues and shouldn't be above criticism. Additionally, I find myself laughing at the Players' Awards. They make this big stink about player's doing a better job than reporter's in evaluating players, and then they immediately side against Curry in favor of Harden. I'm a Harden fan, but to me it was no contest.

I don't see how Harden playing like crap this year somehow invalidates his legitimacy as an MVP candidate last year.

It's really more about Curry.

And I understand you could argue, "Why should Curry getting better this year matter to who was better last year?", but while I didn't predict this in particular for Curry, central to my entire argument last year was Curry's game scaled in a way that we couldn't expect of Harden's - at least the way Houston was using Harden. Any take this year saying "Who knew Curry could score more if it was so desired?" has me rolling my eyes because I thought that was pretty damn clear last year.

Put another way: Harden won that award because he scored a lot of points with a narrative that let people say "And he's doing it all by himself.", and I always took exception to that way of thinking.

Most valuable player isn't the same as best though.

[trex_8063]

He does not lead the league in DRB%, he's a bit under last year's % actually, when his DRPM was considerably lower. And I meant the team's DRB% mostly, which is awful when he's on the court (and a tad less awful when he's out).

This is not RPM not eliminating the boxscore issues, it's a case of one single boxscore stat (BLK%) shooting up a player's numbers to the top of the league, even though everything else indicates a regression on his part.

Where are you getting this information? The bolded/colored portion above in particular doesn't appear at all accurate according to the data on bbref:

Clippers DREB% with Jordan on the court: 74.5%
Clippers DREB% with Jordan off the court: 69.6%

Clippers OREB% with Jordan on the court: 21.9%
Clipper OREB% with Jordan off the court: 20.3%

Even with him on the court their TREB% is still only 49.1% (so yeah, not good); but it's an abysmal 42.7% when he sits. So while it's true they're not a good rebounding team so far this year, it's equally clear that they do get a big lift from him.

I'm not going to get into the specifics of these numbers, but I was pretty vocal last year emphasizing the issues with defensive rebounding statistics particularly with Jordan. You can check out details here:

http://www.gotbuckets.com/2015-ffapm-defense/

Basically according to this data, Jordan actually had a slightly negative impact on defensive rebounding. To be clear, the important point to me wasn't that Jordan had a less than 0 impact there because a positive impact certainly wouldn't have surprised me, it's simply that when you see guys chase boards as hard as Jordan did, it's pretty much a given in my experience that there defensive rebounding gets overrated because that's not actually optimal strategy for team defensive rebounding most of the time.

I'm fuzzy on how to interpret the FFAPM data. Is his Reb rating of 42 bad? I get the gist of the "Four Factors", but I don't quite understand what those numbers are telling me (and his defensive FFAPM is still good---though not elite).

At any rate.....
I agree---and have stated elsewhere, I think in relation to a Kevin Love discussion---that chasing boards may increase the individual's chance of obtaining the board, but it may have a neutral (or even negative) effect on his team's chance of obtaining the board. It also often comes at the expense of shot contesting.
However, the data on bbref is again at odds with your above claim that Jordan had a NEGATIVE impact on his team's defensive rebounding last year:

'15 Clips DREB% with Jordan on court: 76.2%
'15 Clips DREB% with Jordan off court: 74.7%

[andyhop]

I'm fuzzy on how to interpret the FFAPM data. Is his Reb rating of 42 bad? I get the gist of the "Four Factors", but I don't quite understand what those numbers are telling me (and his defensive FFAPM is still good---though not elite).

However, the data on bbref is again at odds with your above claim that Jordan had a NEGATIVE impact on his team's defensive rebounding last year:

'15 Clips DREB% with Jordan on court: 76.2%
'15 Clips DREB% with Jordan off court: 74.7%

The 42 is his defensive rebounding FF score which they then convert into percentile rankings.

Your on court/off court numbers don't show that Jordan didn't have negative impact on the Clippers defensive rebounding they show that the team got a higher percentage of defensive rebounds when he was on court than when he was off court which isn't the same thing.

[CBA]

Lawson/Harden is such a horrible fit.

To be fair, Lawson/anyone is a terrible fit so far this season. Every one on the Rockets plays better with him off the court. James Harden, for example, is 63%TS on 34%USG (61%/31% last year). Eliminating (this version of) Lawson from the rotation should help them.

[Doctor MJ]

I don't see how Harden playing like crap this year somehow invalidates his legitimacy as an MVP candidate last year.

It's really more about Curry.

And I understand you could argue, "Why should Curry getting better this year matter to who was better last year?", but while I didn't predict this in particular for Curry, central to my entire argument last year was Curry's game scaled in a way that we couldn't expect of Harden's - at least the way Houston was using Harden. Any take this year saying "Who knew Curry could score more if it was so desired?" has me rolling my eyes because I thought that was pretty damn clear last year.

Put another way: Harden won that award because he scored a lot of points with a narrative that let people say "And he's doing it all by himself.", and I always took exception to that way of thinking.

Most valuable player isn't the same as best though.

I'm not sure what precisely your point is here. To me in this particular case Curry had the lead in both questions by a good margins, and were I asked to elaborate I'd largely give the same answer.

[Doctor MJ]

Where are you getting this information? The bolded/colored portion above in particular doesn't appear at all accurate according to the data on bbref:

Clippers DREB% with Jordan on the court: 74.5%
Clippers DREB% with Jordan off the court: 69.6%

Clippers OREB% with Jordan on the court: 21.9%
Clipper OREB% with Jordan off the court: 20.3%

Even with him on the court their TREB% is still only 49.1% (so yeah, not good); but it's an abysmal 42.7% when he sits. So while it's true they're not a good rebounding team so far this year, it's equally clear that they do get a big lift from him.

I'm not going to get into the specifics of these numbers, but I was pretty vocal last year emphasizing the issues with defensive rebounding statistics particularly with Jordan. You can check out details here:

http://www.gotbuckets.com/2015-ffapm-defense/

Basically according to this data, Jordan actually had a slightly negative impact on defensive rebounding. To be clear, the important point to me wasn't that Jordan had a less than 0 impact there because a positive impact certainly wouldn't have surprised me, it's simply that when you see guys chase boards as hard as Jordan did, it's pretty much a given in my experience that there defensive rebounding gets overrated because that's not actually optimal strategy for team defensive rebounding most of the time.

I'm fuzzy on how to interpret the FFAPM data. Is his Reb rating of 42 bad? I get the gist of the "Four Factors", but I don't quite understand what those numbers are telling me (and his defensive FFAPM is still good---though not elite).

At any rate.....
I agree---and have stated elsewhere, I think in relation to a Kevin Love discussion---that chasing boards may increase the individual's chance of obtaining the board, but it may have a neutral (or even negative) effect on his team's chance of obtaining the board. It also often comes at the expense of shot contesting.
However, the data on bbref is again at odds with your above claim that Jordan had a NEGATIVE impact on his team's defensive rebounding last year:

'15 Clips DREB% with Jordan on court: 76.2%
'15 Clips DREB% with Jordan off court: 74.7%

As mentioned, it's a percentile based rating.

50 would be average.
42 is slightly below average.
DeAndre Jordan led the entire league in DREB% so he should have had 100.

It's the gap between the second and the third numbers that are such a big deal.

Re: Love. Actually if you go back to '13-14, Love's rating with this stat was a 96, and to my knowledge Love never showed any major trend of similar "stat padding". People only think he had issues because his team sucked.

Re: at the expense of shot contesting. Well, Love has defensive weaknesses to be sure. It's crucial to understand though that the reason why Jordan's defensive rebounding is overrated is basically the same reason why his blocks are overrated: He chased the individual action rather than taking an approach that works more with the team. This is also noteworthy because if the reasoning is "well he can get those defensive rebounds because he isn't trying to disrupt shots", well, oftentimes the worst abusers of this type of play seem to have zero problem both overplaying for blocks and overplaying for rebounds.

Re: bbref is at odds. No it isn't. Look, if I'm on LeBron's team and I play with LeBron whenever I play, there's a good chance my team's numbers are going to look great when I'm on the court compared to win I'm off it. The reason isn't because of me, it's because of LeBron. This is the entire reason why when we first started getting access to +/- stats everyone recognized it wasn't enough to simply count what happens when players are on the court, you need more advanced statistical techniques to adjust for context.

[trex_8063]

I'm fuzzy on how to interpret the FFAPM data. Is his Reb rating of 42 bad? I get the gist of the "Four Factors", but I don't quite understand what those numbers are telling me (and his defensive FFAPM is still good---though not elite).

However, the data on bbref is again at odds with your above claim that Jordan had a NEGATIVE impact on his team's defensive rebounding last year:

'15 Clips DREB% with Jordan on court: 76.2%
'15 Clips DREB% with Jordan off court: 74.7%

The 42 is his defensive rebounding FF score which they then convert into percentile rankings.

OK, so the 42 means 42nd percentile (like better than 42% of league, worse than 58%), is that correct?

Your on court/off court numbers don't show that Jordan didn't have negative impact on the Clippers defensive rebounding they show that the team got a higher percentage of defensive rebounds when he was on court than when he was off court which isn't the same thing.

I'm having a hard time reconciling this statement. From a defensive rebounding standpoint, the goal is to obtain the defensive rebound/prevent the opposing team from getting the offensive rebound. So getting a higher % of the defensive rebounds is, in essence, "doing better" on the defensive boards.
Thus, the way I'm interpreting your above statement is: Jordan has a negative impact on their defensive rebounding, but their defensive rebounding is better when he's on the court.
These assertions seem at odds with each other.


Now it could be that the rebounding rates of his four teammates on the court go down when he steps on to the court (i.e. his rebounds are coming mostly at the expense of his teammates). That is often the case of many (if not most) high-volume rebounders, I should think. Perhaps more so with Jordan than with others, idk.
But still, if he's grabbing so many boards as an individual (even if many of them are at the expense of his teammates) that the team defensive rebounding rate goes up, I don't understand how that can be labeled a net negative impact on their defensive rebounding.

It could be that if he was blocking out instead of "chasing" boards, that the TEAM defensive rebounding rate would be even better (even if his individual defensive rebounding rate was a little worse) when he's on the court; but that's not the same thing as having a negative impact.


So I'm still kinda confused by your above statement. Perhaps you could elaborate for me.