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

[CommonerCoffee]

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 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:

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

Sorry to nitpick but I believe the first statement is also invalid. Given that our sample size is so small, the +8.85 estimate
likely has an enormous confidence interval and the 8.85 itself is almost certainly not the true MOV parameter. In an extreme case, it may not even be significantly different from the 0 null (though in all likelihood, it is). I'm also not too familiar with the construction of the regression model but does anyone know if the model contain interaction terms?

Also the generalization here about any random team is somewhat dangerous given the importance of context and roles, which the regression model is unable to capture.

[SideshowBob]

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 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:

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

Sorry to nitpick but I believe the first statement is also invalid. Given that our sample size is so small, the +8.85 estimate
likely has an enormous confidence interval and the 8.85 itself is almost certainly not the true MOV parameter. In an extreme case, it may not even be significantly different from the 0 null (though in all likelihood, it is). I'm also not too familiar with the construction of the regression model but does anyone know if the model contain interaction terms?

That is correct, but I started by posting this on the general board and I did actually include the problem with errors at first but figured it'd just end up being more confusing. Unfortunately though, I don't think we've ever seen error terms listed for JE's RAPM or xRAPM or RPM, but I know he's run OOS tests for confirmation on predictive power.

As for interaction terms, the RPM regression would not have any, but the SPM used as a prior almost certainly does (which was the case for xRAPM, I'm not as familiar with RPM but I'm pretty certain its the same).

[bballexpert]

I am glad that Duncan is doing so well in defense still this year he is playing better d then last year. Maybe he can turn back the clock more and win a dopy which he should of won in 2005,2006,2007 and 2013 maybe he might get it.

[picc]

Not a RAPM fan. Saying that, its possible Curry's negative defensive rating has to do with his high number of turnovers and how they usually lead to immediate fastbreak points.

[ceiling raiser]

Thanks. Wonder how much of DAJ's high rating is due to the SPM prior.

[Texas Chuck]

Thanks. Wonder how much of DAJ's high rating is due to the SPM prior.

yeah Clipper fans have been pretty criticial of the guy(have intentionally not looked personally at how he is playing for a reason probably not hard to figure out. ) so I was stunned to see his early number here looking so good.

[SideshowBob]

Thanks. Wonder how much of DAJ's high rating is due to the SPM prior.

PCP asked JE about it.

[tweet]https://twitter.com/JerryEngelmann/status/669262181121265665[/tweet]

Also, I tweeted him for multi-year RPM/xRAPM and he just responded saying its in the works.

[Doctor MJ]

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 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:

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

Except of course that this is also a valid statement:

"If I put Anthony Davis on any random team, I expect the team performance (MOV), while Anthony Davis is on the court, to improve by 1.12 points per 100 possessions, given the the league-wide lineup data we have for the first 3 weeks of the 2016 season, while I expect if I put Clint Capella on any random team, I expect the team performance (MOV), while Clint Capella is on the court, to improve by 2.66 points per 100 possessions."

And if any human walks up to you and makes that statement you laugh in his face for being an idiot and tell him to go back to his mommy's basement and watch some more World Series of Poker or whatever it is he watches instead of basketball.

But now dress that dumbass up as a "metric" and ooh! What amazing insight he must have!

That's not what the metric says though, it simply gives a best-estimate correlation of impact based on the performance to this point using a standard approach used across science.

Folks like you tend to look at any new thing you don't understand, find something you see as imperfect about it, and use that as a reason to feel righteous about your rejection of it.

Folks like us are looking for additional tools to put in our analyst's arsenal. We expect all tools to have flaws and limits, the only question is how it can add to our understanding of what's happening. As such, at least for me personally, there are very few metrics that are bad enough I wish they weren't around, and the incorporation of regression data to the analysis is certainly not one of them. And the NBA obviously agrees.

But hey, I don't really expect to convince you Winsome, so unless you really think you're going to have a productive discussion here, you should probably avoid jumping into thread to hate-post on the concept it's discussing.

[Doctor MJ]

Thanks. Wonder how much of DAJ's high rating is due to the SPM prior.

PCP asked JE about it.

[tweet]https://twitter.com/JerryEngelmann/status/669262181121265665[/tweet]

Also, I tweeted him for multi-year RPM/xRAPM and he just responded saying its in the works.

I'm really glad he acknowledged this issue.

[Doctor MJ]

#16 SF Jae Crowder RPM=3.52
#26 SF Kevin Durant RPM=2.94

#20 C Mason Plumlee RPM=3.34
#37 C DeMarcus Cousins RPM=2.33

#21 PF Marvin Williams RPM=3.31
#78 PF Anthony Davis RPM=1.12

#36 PG George Hill RPM=2.38
#64 PG Chris Paul RPM=1.41

I've mentioned before my belief that its a stat's responsibility to conform to reality. It can't create reality. And if what it measures doesn't conform to reality then what is its use?

If anyone uses RPM based on a 12-14 game sample size that is their fault and not the stat's fault.

That would imply that things will balance out, but that short term those numbers reflect some significant reality.

Do you understand why people talk about the concerns of small sample size?

In any study without sufficient sample size it's basically like having a big randomness factor built in because of the inherent variance to the data itself. The data we have to this point will be legit and useful as things go forward, but the clear way to use anything like this is with common sense and caution. You don't try to make bold statements simply based on numbers like this, but on the other hand, looking at the data to this point might give you an idea for something to pay attention to that could be useful, or vice versa.

[day1086]

Interesting to see Curry as a negative defensively, on-court he's clearly been better on that end and a small positive at the least.

I feel like it could be a product of his high turnovers recently. Turnover -> dunk isn't really about you playing defense, but it's gonna hurt your defensive metrics, right?

[Texas Chuck]

Interesting to see Curry as a negative defensively, on-court he's clearly been better on that end and a small positive at the least.

I feel like it could be a product of his high turnovers recently. Turnover -> dunk isn't really about you playing defense, but it's gonna hurt your defensive metrics, right?

Uh yeah catastrophic turnovers definitely should impact defensive metrics. Its why high-volume low-turnover guys are really valuable. One of the ways Rick Carlisle has kept the Mavericks in the playoff mix post-title with relatively little talent is by a strong emphasis on taking care of the ball. Live-ball turnovers are the absolute worst thing you can do offensively.

It's one of the few flaws in Curry and Westbrook's games right now.

[ceiling raiser]

Interesting to see Curry as a negative defensively, on-court he's clearly been better on that end and a small positive at the least.

I feel like it could be a product of his high turnovers recently. Turnover -> dunk isn't really about you playing defense, but it's gonna hurt your defensive metrics, right?

Uh yeah catastrophic turnovers definitely should impact defensive metrics. Its why high-volume low-turnover guys are really valuable. One of the ways Rick Carlisle has kept the Mavericks in the playoff mix post-title with relatively little talent is by a strong emphasis on taking care of the ball. Live-ball turnovers are the absolute worst thing you can do offensively.

It's one of the few flaws in Curry and Westbrook's games right now.

Exactly right. From APBRmetrics awhile back, ORtg of next possession by origin (from the Spanish league, but still valuable info):

Effic. Possession start
126.8 Steal
124.5 (2p FG missed, off. reb.)

115.5 (3p FG missed, off. reb.)
112.5 2p FG missed, def. reb.
111.6 Non-steal turnover

110.4 Average

108.4 FT made
107.3 3p FG missed, def. reb.
107.0 2p FG made
106.9 3p FG made

104.7 Start of quarter
103.2 FT missed, def. reb.

[day1086]

Interesting to see Curry as a negative defensively, on-court he's clearly been better on that end and a small positive at the least.

I feel like it could be a product of his high turnovers recently. Turnover -> dunk isn't really about you playing defense, but it's gonna hurt your defensive metrics, right?

Uh yeah catastrophic turnovers definitely should impact defensive metrics. Its why high-volume low-turnover guys are really valuable. One of the ways Rick Carlisle has kept the Mavericks in the playoff mix post-title with relatively little talent is by a strong emphasis on taking care of the ball. Live-ball turnovers are the absolute worst thing you can do offensively.

It's one of the few flaws in Curry and Westbrook's games right now.

Yeah I was just explaining why Curry's improved defense via the eye-test versus his defensive metrics don't match up.

Turnovers are an offensive mistake, which hurts your defense. But it's different from watching the guy play halfcourt defense.

[The-Power]

Yeah I was just explaining why Curry's improved defense via the eye-test versus his defensive metrics don't match up.

Turnovers are an offensive mistake, which hurts your defense. But it's different from watching the guy play halfcourt defense.

On the other hand, Curry's efficiency (or rather the offensive efficiency of the whole team when he plays) as well as the kind of shots he takes (from beyond the arc, allowing him to get back on defense early and also statistically accompanied by a higher share of long rebounds) should improve his defensive rating. That was one of my explanations for his high defensive ratings in some +/- data last season. So maybe we should wait until the sample size increases. Could also be noise and we see a more neutral rating in the near future.

[Winsome Gerbil]

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 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:

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

Except of course that this is also a valid statement:

"If I put Anthony Davis on any random team, I expect the team performance (MOV), while Anthony Davis is on the court, to improve by 1.12 points per 100 possessions, given the the league-wide lineup data we have for the first 3 weeks of the 2016 season, while I expect if I put Clint Capella on any random team, I expect the team performance (MOV), while Clint Capella is on the court, to improve by 2.66 points per 100 possessions."

And if any human walks up to you and makes that statement you laugh in his face for being an idiot and tell him to go back to his mommy's basement and watch some more World Series of Poker or whatever it is he watches instead of basketball.

But now dress that dumbass up as a "metric" and ooh! What amazing insight he must have!

That's not what the metric says though, it simply gives a best-estimate correlation of impact based on the performance to this point using a standard approach used across science.

Folks like you tend to look at any new thing you don't understand, find something you see as imperfect about it, and use that as a reason to feel righteous about your rejection of it.

Folks like us are looking for additional tools to put in our analyst's arsenal. We expect all tools to have flaws and limits, the only question is how it can add to our understanding of what's happening. As such, at least for me personally, there are very few metrics that are bad enough I wish they weren't around, and the incorporation of regression data to the analysis is certainly not one of them. And the NBA obviously agrees.

But hey, I don't really expect to convince you Winsome, so unless you really think you're going to have a productive discussion here, you should probably avoid jumping into thread to hate-post on the concept it's discussing.

I think I'll continue to hop into any thread that I feel like to poke giant holes into its shoddy premises or logic. Which is to say a whole bunch of them around here. When I start resorting to "this stoopid!" posts, feel free to re-disinvite me.

[MyUniBroDavis]

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 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:

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

I like RAPM and RPM as a statistic (prefer the former) but just wondering, what would this mean for small sample sizes, and how it effects players?

I mean, Obviously im a davis fan, but I honestly can understand his offensive rapm being bad so far (his offense has, to be blunt, not been up to par this year, though I expect that to change)

otoh, I dont understand how his defense has been so bad. is this just due to small sample size? I mean, his net rating on that end is 9.5, which looks really good looking at the teams lineups. though granted, he was utter trash teh first 3 games on that end, which would effect the small sample size

where can we find a list of SPM stats (im not sure what that is, but its a type of plus minus right?)

[Jaivl]

where can we find a list of SPM stats (im not sure what that is, but its a type of plus minus right?)

BRef's BPM is a type of SPM (Statistical Plus Minus) stat, you can start there.

[MyUniBroDavis]

where can we find a list of SPM stats (im not sure what that is, but its a type of plus minus right?)

BRef's BPM is a type of SPM (Statistical Plus Minus) stat, you can start there.

Definately explains alot, tbh.

Gonna wait a bit before I take it seriously, sample size is way too small.

[SideshowBob]

I like RAPM and RPM as a statistic (prefer the former) but just wondering, what would this mean for small sample sizes, and how it effects players?

I mean, Obviously im a davis fan, but I honestly can understand his offensive rapm being bad so far (his offense has, to be blunt, not been up to par this year, though I expect that to change)

otoh, I dont understand how his defense has been so bad. is this just due to small sample size? I mean, his net rating on that end is 9.5, which looks really good looking at the teams lineups. though granted, he was utter trash teh first 3 games on that end, which would effect the small sample size

where can we find a list of SPM stats (im not sure what that is, but its a type of plus minus right?)

It's not stable enough - because of variance present in such a small sample, the error terms on most of the coefficients (which are the RPM values) is so large that we can't really say anything certain with any acceptable level of confidence.

So consider that and combine it with Davis having a few underwhelming games (which weigh heavily in the sample) and its not too crazy for him to have a lower than typical defensive value.

Another thing to keep in mind; typically in the Stat/Econometrics field, predictive power is what we're looking for. This leans more on the side of descriptive, a multi-year or prior-year-informed RPM (see 2014 RPM on ESPN) will outperform this metric in predicting how the rest of the season plays out. I tweeted Engelmann on the availability of that and its on its way as well. Davis will rank higher in that metric because, in a manner of speaking, includes a bit of context - a baseline of each player's previous year's performance to work with, rather than each player starting at 0.