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

[lorak]

was referring to multi-year RAPM

Which one exactly?

https://docs.google.com/spreadsheets/d/12rcLDCaJSh_b1hA3UyLG5-neayTBl_weqNVCGM0LheA/edit

So what's wrong with it? If it's multiyear, then AD's rookie year is included and he wasn't too good then, while Ellis for several years back then was quite valuable player. I think you just shouldn't use such studies (multiyear) to judge players in ONE year...

[Knosh]

Could someone make me a ELI5 (Explain Like I'm 5) on RAPM and RPM¿?

Thanks

How about an ELI10?

Spoiler:

[#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.

I don't want to open the can of worms that is the financial collapse, but as far as I understand, it's more along the lines of knowingly turning a blind-eye to what were known as poor models, rather than them just being present/utilized.

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.

I don't think you know about OLS when your 10

[cpower]

Which one exactly?

https://docs.google.com/spreadsheets/d/12rcLDCaJSh_b1hA3UyLG5-neayTBl_weqNVCGM0LheA/edit

So what's wrong with it? If it's multiyear, then AD's rookie year is included and he wasn't too good then, while Ellis for several years back then was quite valuable player. I think you just shouldn't use such studies (multiyear) to judge players in ONE year...

it uses mutiyear to eliminate noise but put more emphasis on recent possessions. Otherwise, Why do you think Curry has a better rating than Lebron?

[lorak]

https://docs.google.com/spreadsheets/d/12rcLDCaJSh_b1hA3UyLG5-neayTBl_weqNVCGM0LheA/edit

So what's wrong with it? If it's multiyear, then AD's rookie year is included and he wasn't too good then, while Ellis for several years back then was quite valuable player. I think you just shouldn't use such studies (multiyear) to judge players in ONE year...

it uses mutiyear to eliminate noise but put more emphasis on recent possessions. Otherwise, Why do you think Curry has a better rating than Lebron?

Because Curry is better than LeBron since long time.

[cpower]

So what's wrong with it? If it's multiyear, then AD's rookie year is included and he wasn't too good then, while Ellis for several years back then was quite valuable player. I think you just shouldn't use such studies (multiyear) to judge players in ONE year...

it uses mutiyear to eliminate noise but put more emphasis on recent possessions. Otherwise, Why do you think Curry has a better rating than Lebron?

Because Curry is better than LeBron since long time.

and Michael Jordan never played the game of basketball

[The-Power]

well, still way better than someone doesn't know how to play offense and defense at all.

RAPM isn't supposed to measure goodness, though. That's what we have to keep in mind and where I agree with what MyUniBroDavis was probably referring to. When we evaluate a player, we take his role into account. Whenever a player is being used the wrong way, we use this in our evaluation. Whenever a player with obvious deficiencies has a perfectly suited role that cannot or can only hardly be replicated, we use this in our evaluation. Whenever a player has a flawed approach to play the game in our opinion, but the team happens to rely heavily on what he does, we use this in our evaluation. You're going to see impact data that, even if we assume there's no noise and collinearity in it, doesn't necessarily capture exactly what you're looking for or trying to evaluate.

Re: Ellis and Davis. First of all, let me say that RAPM isn't gospel and we shouldn't treat it as such. Aside from what I wrote above, collinearity certainly is an issue and the sample size, although big enough to provide at least somewhat resilient information, still isn't ideal. Anyway, when we look at Ellis we see that he has a reduced role on offense this season. His USG% is at the lowest point since his rookie season. This means inefficient offense might very well be still existent but it doesn't carry as much weight since his inefficient approach is less emphasized when he's on the court. Davis, on the other hand, has a clearly higher USG% and his team's offense is built around him to a higher degree. When such a player, as it is the case with Davis, struggles to deal with his role on offense then his influence - negative or positive - is going to be more emphasized by impact stats regardless of whether he's a superior offensive player or not. On defense it's also not hard to see why Davis impact is only slightly positive according to DRAPM. Raw on/off-numbers suggest basically neutral impact on that end and while this is of course clearly insufficient to reach a resilient conclusion because it lacks any context we have to take into consideration and because it is completely unadjusted, hence sharing the time with certain players more or less is going to heavily influence your individual on/off numbers, it gives us a first idea of what to expect. Then we have scouting which tells us that his defense still has a long way to go to get where it could be and we can also see that the Pelicans with Davis on the court have a pretty bad DRTG that cannot fully be explained by Davis being surrounded by defenders so poor that his individually very impactful defense is being offset. Ellis DRAPM is hard to believe, yes, but we can try to explain it and put it in context. Keep in mind that RAPM only knows thousands of data points and nothing about reputation, scouting etc. pp. What we see first is that the raw on/off-numbers show no defensive improvement when Ellis is on the bench. WOWY data shows us that Ellis spends most of his time with Paul George on the court who is considered to be a good defender. It's entirely possible and plausible that Ellis - maybe not being as bad as his reputation would suggest this season - has a relatively high DRAPM because the lineup-data we have isn't able to share the correctly (yet), assuming Ellis in fact doesn't deserve as much credit as he gets for the defense of the line-ups he's in.

We know this possibility, and we also have other tools in our evaluation-box, so that we don't have to look only at RAPM and shouldn't do it either. At the same time, we shouldn't disregard the tool RAPM is just because we don't agree with particular numbers because it definitely has its value. Like always, we must look at everything valuable in order to evaluate a player as good as possible with the limited information we have. Then, after taking everything into consideration and explaining discrepancies with rational reasoning, we might be able to get a pretty accurate idea of what a player contributes to his team. A different approach, like trusting a single stat - probably because it ranks the players closest to one's own personal and prejudiced opinion - is doomed to fail if it's supposed to evaluate a player as a whole (and create a ranking).

if you don't agree on this particular numbers, would you trust other numbers? It is very hard to tell which number has noise and which ones do not. From a non-basketball fan's perspective , player A has more impact than player B. It just happens to be player A is Monta Ellis and player B is Davis.

The same is true for every single number out there. RAPM is measuring impact in the sole valid way, so what I certainly agree with is the approach. That's what matters the most, first of all. There is some noise because of sample size issues but it doesn't mean we shouldn't use the numbers RAPM delivers - that's why I talked about the necessity of adding context as well as other stats, numbers, data and scouting. As I said, if you disregard something because there are some unlikely results then a) you can disregard any single stat out there and b) you simply picked the wrong way to evaluate a player because you don't add other information available. It's not the fault of the stat, though.

[MyUniBroDavis]

On a pure impact basis it's understandable why Davis ranks so low.

His offense is not being used right

And his defense, which started out well, has progressively gotten worse as the season goes on.

well, still way better than someone doesn't know how to play offense and defense at all.

RAPM isn't supposed to measure goodness, though. That's what we have to keep in mind and where I agree with what MyUniBroDavis was probably referring to. When we evaluate a player, we take his role into account. Whenever a player is being used the wrong way, we use this in our evaluation. Whenever a player with obvious deficiencies has a perfectly suited role that cannot or can only hardly be replicated, we use this in our evaluation. Whenever a player has a flawed approach to play the game in our opinion, but the team happens to rely heavily on what he does, we use this in our evaluation. You're going to see impact data that, even if we assume there's no noise and collinearity in it, doesn't necessarily capture exactly what you're looking for or trying to evaluate.

Re: Ellis and Davis. First of all, let me say that RAPM isn't gospel and we shouldn't treat it as such. Aside from what I wrote above, collinearity certainly is an issue and the sample size, although big enough to provide at least somewhat resilient information, still isn't ideal. Anyway, when we look at Ellis we see that he has a reduced role on offense this season. His USG% is at the lowest point since his rookie season. This means inefficient offense might very well be still existent but it doesn't carry as much weight since his inefficient approach is less emphasized when he's on the court. Davis, on the other hand, has a clearly higher USG% and his team's offense is built around him to a higher degree. When such a player, as it is the case with Davis, struggles to deal with his role on offense then his influence - negative or positive - is going to be more emphasized by impact stats regardless of whether he's a superior offensive player or not. On defense it's also not hard to see why Davis impact is only slightly positive according to DRAPM. Raw on/off-numbers suggest basically neutral impact on that end and while this is of course clearly insufficient to reach a resilient conclusion because it lacks any context we have to take into consideration and because it is completely unadjusted, hence sharing the time with certain players more or less is going to heavily influence your individual on/off numbers, it gives us a first idea of what to expect. Then we have scouting which tells us that his defense still has a long way to go to get where it could be and we can also see that the Pelicans with Davis on the court have a pretty bad DRTG that cannot fully be explained by Davis being surrounded by defenders so poor that his individually very impactful defense is being offset. Ellis DRAPM is hard to believe, yes, but we can try to explain it and put it in context. Keep in mind that RAPM only knows thousands of data points and nothing about reputation, scouting etc. pp. What we see first is that the raw on/off-numbers show no defensive improvement when Ellis is on the bench. WOWY data shows us that Ellis spends most of his time with Paul George on the court who is considered to be a good defender. It's entirely possible and plausible that Ellis - maybe not being as bad as his reputation would suggest this season - has a relatively high DRAPM because the lineup-data we have isn't able to share the correctly (yet), assuming Ellis in fact doesn't deserve as much credit as he gets for the defense of the line-ups he's in.

We know this possibility, and we also have other tools in our evaluation-box, so that we don't have to look only at RAPM and shouldn't do it either. At the same time, we shouldn't disregard the tool RAPM is just because we don't agree with particular numbers because it definitely has its value. Like always, we must look at everything valuable in order to evaluate a player as good as possible with the limited information we have. Then, after taking everything into consideration and explaining discrepancies with rational reasoning, we might be able to get a pretty accurate idea of what a player contributes to his team. A different approach, like trusting a single stat - probably because it ranks the players closest to one's own personal and prejudiced opinion - is doomed to fail if it's supposed to evaluate a player as a whole (and create a ranking).

The biggest thing with his offense is that (since I believe it's lineup adjusted) the fact is that atm, anderson is a far bette choice as the go to ISO scorer than Davis is.

In the post, Anderson scores 0.99 PPP. Davis is at 0.74 PPP. Anderson is 8th with players with more than 100 possessions.

Davis is 5th last in that regard.

Neither is very good from ISO scoring, but davis is clearly worse.

The biggest reason is gentry. Apparently, to him, a pf that is probably the quickest at his position in the league, the most athletic player in the nba according to most, and with "guard" skills (dribbling has improved) means "lol he 7ft so all he should do is post up.

Essentially, the difference is a elite level ISO player to among, if not the worst, in the league. (He frequently gets the ball off the ISO, nearly takes his opponent off the dribble, only to start posting him up right before he gets past his opponent)

I can't be bothered to talk about the pick and roll. Someone like Davis can shoot, I get it, but him taking as many jumpers as he does off the p and r is ridiculous. Imo, it should be a 65-70:35-30 split. It's 55:45 now.


And holy crap his defense fell off a cliff since new year.

[ceiling raiser]

[tweet]https://twitter.com/JerryEngelmann/status/698836320668352512[/tweet]
[tweet]https://twitter.com/JerryEngelmann/status/698837211404435456[/tweet]

[kayess]

[tweet]https://twitter.com/JerryEngelmann/status/698836320668352512[/tweet]
[tweet]https://twitter.com/JerryEngelmann/status/698837211404435456[/tweet]

Again, this is really x-RAPM right? Thanks fpliii!

Wasn't there a blog (practice something) that had an RAPM calc that was closest to the original or something though?

[ceiling raiser]

[tweet]https://twitter.com/JerryEngelmann/status/698836320668352512[/tweet]
[tweet]https://twitter.com/JerryEngelmann/status/698837211404435456[/tweet]

Again, this is really x-RAPM right? Thanks fpliii!

Wasn't there a blog (practice something) that had an RAPM calc that was closest to the original or something though?

No, xRAPM evolved into RPM (on ESPN). This is RAPM (no box score SPM).

J.E. (same guy) had xRAPM on his site (from 00-01 on; prior to that they're fake RAPM, from a regression based on wuartery team scores and simulated rotations) but it's down now. Can still access via archive.org though. URL was stats-for-the-nba.appspot.com.

[NinjaSheppard]

It really seems like someone is going to poach Rubio from Minnesota and reap massive benefits

[Doctor MJ]

[tweet]https://twitter.com/JerryEngelmann/status/698836320668352512[/tweet]
[tweet]https://twitter.com/JerryEngelmann/status/698837211404435456[/tweet]

For the single year:

So, 4 of the top 5 guys are Warriors, that's a huge problem. The Warriors are in one of those rare situations I think where they are playing so good that it essentially breaks a stat that relies on regression. And by "rare", the last time I remember seeing anything like this was the '05-06 Pistons, and back then it was pre-RAPM. RAPM should be much more resilient to an issue like this, but apartly the Warriors are just that out there.

Beyond that, most interesting thing: Durant still has the edge over Westbrook. It's not big enough I'd make a huge deal about it, but consensus that Westbrook has cleanly surpassed Durant is something that to me the jury is still out on.

[thekdog34]

Both Harden and Anthony Davis are below average players by single year.

Harden is easier to explain. Horrible defense and the worst kind of turnovers

[bondom34]

Good read on RPM vs. RAPM and the effect of the box score.

http://nyloncalculus.com/2016/02/19/freelance-friday-deconstructing-rpm-the-mighty-prior/

[Dr Spaceman]

Good read on RPM vs. RAPM and the effect of the box score.

http://nyloncalculus.com/2016/02/19/freelance-friday-deconstructing-rpm-the-mighty-prior/

Great article, and the Jordan thing as well as other stuff in this article has led me to the conclusion the box score influence here might be way bigger than I thought. Given that, I'm not much sure what the value of RPM is as opposed to just weighing all this divergent stuff in your head.

I really liked the point about the Warriors' offensive success being basically needlessly split up. Ironically gnus makes RPM actually better at grasping Curry's impact, but now looking back at previous years I think it's worth questioning whether we've actually been under rating Curry for a while. I kind of hinted at this in a recent thread, but didn't have the ammo to really dig in.

[bondom34]

Good read on RPM vs. RAPM and the effect of the box score.

http://nyloncalculus.com/2016/02/19/freelance-friday-deconstructing-rpm-the-mighty-prior/

Great article, and the Jordan thing as well as other stuff in this article has led me to the conclusion the box score influence here might be way bigger than I thought. Given that, I'm not much sure what the value of RPM is as opposed to just weighing all this divergent stuff in your head.

I really liked the point about the Warriors' offensive success being basically needlessly split up. Ironically gnus makes RPM actually better at grasping Curry's impact, but now looking back at previous years I think it's worth questioning whether we've actually been under rating Curry for a while. I kind of hinted at this in a recent thread, but didn't have the ammo to really dig in.

Yeah, was literally in this discussion on teh trade board citing Steven Adams vs. Pau Gasol on defense. They entirely flip based on RPM vs. RAPM in single and multi year, and checking BPM Gasol is runaway in the lead while Adams leads in RAPM. Glad to see it was researched more.

[colts18]

Anyone notice that Kawhi Leonard has surpassed Curry in RPM.

http://espn.go.com/nba/statistics/rpm/_/sort/RPM

[SideshowBob]

Anyone notice that Kawhi Leonard has surpassed Curry in RPM.

http://espn.go.com/nba/statistics/rpm/_/sort/RPM

Given Curry's massive lead in the box-score that might be suggestive of Leonard passing Curry in the next pass of NPI 1-year RAPM.

[Doctor MJ]

Anyone notice that Kawhi Leonard has surpassed Curry in RPM.

http://espn.go.com/nba/statistics/rpm/_/sort/RPM

Wow, so, good as time as any I suppose to ask what are the odds Kawhi wins POY. Thoughts?

[trex_8063]

[tweet]https://twitter.com/JerryEngelmann/status/698836320668352512[/tweet]
[tweet]https://twitter.com/JerryEngelmann/status/698837211404435456[/tweet]

Again, this is really x-RAPM right? Thanks fpliii!

Wasn't there a blog (practice something) that had an RAPM calc that was closest to the original or something though?

No, xRAPM evolved into RPM (on ESPN). This is RAPM (no box score SPM).

J.E. (same guy) had xRAPM on his site (from 00-01 on; prior to that they're fake RAPM, from a regression based on wuartery team scores and simulated rotations) but it's down now. Can still access via archive.org though. URL was stats-for-the-nba.appspot.com.

Hmm.....Avery Bradley is a slight negative defensively according to the single-year data. That's........well, counter-intuitive.

Also, def seeing what Doc has worried over (that Curry/GS has "broken" RAPM). Does seem some colinearity (sp?) concerns are appropriate given the 1, 2, 3 finish of Curry, Green, Thompson.