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

[Atmanne]

RPM has just been updated: http://espn.go.com/nba/statistics/rpm/_/sort/RPM

Quite a bit different from last week's.

[MyUniBroDavis]

You guys think Curry Over/Under 11+?

[RxMidnight]

You guys think Curry Over/Under 11+?

+11 would be a ridiculous level to maintain over an entire season. He might peek above it for a short time but +10 is already impressive enough considering nobody has done it before.

[bballexpert]

RPM has just been updated: http://espn.go.com/nba/statistics/rpm/_/sort/RPM

Quite a bit different from last week's.

Duncan strait killing it still and it is not even close in the drpm man 39 going on 40 still one of the best defense players in the league. Kg, Robinson and Hakeem were never able to keep there defense this high at Duncans age it is nuts.

[SideshowBob]

RPM has just been updated: http://espn.go.com/nba/statistics/rpm/_/sort/RPM

Quite a bit different from last week's.

Yeah, sample size has increased, which means box-score weight has gone down and +/- weight has gone up. As a result, Curry's lead over Russ/Lowry/James has shrunk, they're incrementally closer to his value now than they were in the first iteration. Also the overall spread of the data is wider, haven't looked at the details yet, but SD for RPM in the first set was 2.25. I doubt Curry's 4.2 deviations above the mean in this set.

[spearsy23]

I was looking through the old APBmetrics topics (specifically the original RAPM one) and it made me wonder why this board seemed to have an affinity for RAPM over APM. It seems that RAPM has always been intended as a predictor more so than a summation of past events, so what about it makes it the preferred stat when comparing current/historical seasons? What exactly are the advantages of NPI-RAPM over APM?

[bondom34]

I was looking through the old APBmetrics topics (specifically the original RAPM one) and it made me wonder why this board seemed to have an affinity for RAPM over APM. It seems that RAPM has always been intended as a predictor more so than a summation of past events, so what about it makes it the preferred stat when comparing current/historical seasons? What exactly are the advantages of NPI-RAPM over APM?

I'd love a better explanation on that again. I know RAPM is more predictive, and APM varies way more. I know RPM isn't a predictor and is just a 1 year version of xRAPM, but forget some of the RAPM vs. APM stuff since I don't see much APM (non RAPM). I've always preferred a stat that's telling me more of "what's happining now or has happened" vs. "what will happen" which I think is why I like RPM more.

[spearsy23]

I was looking through the old APBmetrics topics (specifically the original RAPM one) and it made me wonder why this board seemed to have an affinity for RAPM over APM. It seems that RAPM has always been intended as a predictor more so than a summation of past events, so what about it makes it the preferred stat when comparing current/historical seasons? What exactly are the advantages of NPI-RAPM over APM?

I'd love a better explanation on that again. I know RAPM is more predictive, and APM varies way more. I know RPM isn't a predictor and is just a 1 year version of xRAPM, but forget some of the RAPM vs. APM stuff since I don't see much APM (non RAPM). I've always preferred a stat that's telling me more of "what's happining now or has happened" vs. "what will happen" which I think is why I like RPM more.

The way I understood it, which could be completely wrong , is that even NPI-RAPM (and thus RPM) is still intended to have predictive value in that it's predicting what would happen if you dropped player x into any lineup (obviously that's extremely generalized) and the only difference is that instead of using prior season's data it sets the limit at 0. If that's the case it still isn't telling you 'this is what player x did for his team', it's 'what player x could do for any team.' It may be that there is almost no difference in the end, but it seems to my limited understanding that it could lead to some minor differences in determining who had the best season vs who is the best player.

[bondom34]

I was looking through the old APBmetrics topics (specifically the original RAPM one) and it made me wonder why this board seemed to have an affinity for RAPM over APM. It seems that RAPM has always been intended as a predictor more so than a summation of past events, so what about it makes it the preferred stat when comparing current/historical seasons? What exactly are the advantages of NPI-RAPM over APM?

I'd love a better explanation on that again. I know RAPM is more predictive, and APM varies way more. I know RPM isn't a predictor and is just a 1 year version of xRAPM, but forget some of the RAPM vs. APM stuff since I don't see much APM (non RAPM). I've always preferred a stat that's telling me more of "what's happining now or has happened" vs. "what will happen" which I think is why I like RPM more.

The way I understood it, which could be completely wrong , is that even NPI-RAPM (and thus RPM) is still intended to have predictive value in that it's predicting what would happen if you dropped player x into any lineup (obviously that's extremely generalized) and the only difference is that instead of using prior season's data it sets the limit at 0. If that's the case it still isn't telling you 'this is what player x did for his team', it's 'what player x could do for any team.' It may be that there is almost no difference in the end, but it seems to my limited understanding that it could lead to some minor differences in determining who had the best season vs who is the best player.

NPI from my understanding is mostly that. RPM is from what I can gather informed with box score data (no prior, but box score). Both intend to do the same thing, just different ways of calculating it from my understanding, though someone else could probably put it better.

[trex_8063]

I was looking through the old APBmetrics topics (specifically the original RAPM one) and it made me wonder why this board seemed to have an affinity for RAPM over APM. It seems that RAPM has always been intended as a predictor more so than a summation of past events, so what about it makes it the preferred stat when comparing current/historical seasons? What exactly are the advantages of NPI-RAPM over APM?

I'd love a better explanation on that again. I know RAPM is more predictive, and APM varies way more. I know RPM isn't a predictor and is just a 1 year version of xRAPM, but forget some of the RAPM vs. APM stuff since I don't see much APM (non RAPM). I've always preferred a stat that's telling me more of "what's happining now or has happened" vs. "what will happen" which I think is why I like RPM more.

The way I understood it, which could be completely wrong , is that even NPI-RAPM (and thus RPM) is still intended to have predictive value in that it's predicting what would happen if you dropped player x into any lineup (obviously that's extremely generalized) and the only difference is that instead of using prior season's data it sets the limit at 0. If that's the case it still isn't telling you 'this is what player x did for his team', it's 'what player x could do for any team.' It may be that there is almost no difference in the end, but it seems to my limited understanding that it could lead to some minor differences in determining who had the best season vs who is the best player.

I don't think the bolded portions are accurately stating what NPI RAPM (or similar metrics) is attempting to define. If that description of its intent appeared somewhere at some time, it seems wholly incomplete.

More accurately I think it's attempting to predict "what would happen if you dropped player x into any lineup, provided he's allowed to play the exact same role with similar(ish) support pieces around him".

I'll use a couple of extreme hypotheticals to illustrate the point......
Suppose peak Ben Wallace is played at center (just like he was in actuality), but he's forced to be the offensive focal point for the team, shouldering ~30% usage......what do you suppose his offensive impact would look like? I'm guessing it would be godawful in an historic sense (whereas in actuality, his teams utilized him in such a way----extreme low usage, rare opportunistic uses of him----that he basically managed to be [roughly] an offensive neutral according to impact data).

Or suppose we drop Steph Curry on to a different team......but force him to play center. i.e. he doesn't get to be the primary ball-handler on offense, he doesn't get to operate from the outside (except of rare floor-stretching sets or pick n' pops----and how effective do you think a screen set by the 6'3" 180 lb Curry would be, btw?----and similar; he's otherwise positioned in the post or maybe high post on offense). Defensively he's forced to guard one of the bigs. How do you suppose his impact would look in that circumstance? I would say he likely goes from being best in the league down to a net NEGATIVE impact player, because this is an obvious and exaggerated mis-use of his abilities.

Obviously, it's an unrealistic scenario (no one would ever utilize him in this fashion); but it hopefully illustrates the point that how a player is utilized effects his impact on the scoring margin. In an exaggerated case it can even turn the league MVP into a negative-impact player. Smaller [and more realistic] changes to a player's utilization and circumstance can naturally produce smaller changes in their impact potential.

[bondom34]

I'd love a better explanation on that again. I know RAPM is more predictive, and APM varies way more. I know RPM isn't a predictor and is just a 1 year version of xRAPM, but forget some of the RAPM vs. APM stuff since I don't see much APM (non RAPM). I've always preferred a stat that's telling me more of "what's happining now or has happened" vs. "what will happen" which I think is why I like RPM more.

The way I understood it, which could be completely wrong , is that even NPI-RAPM (and thus RPM) is still intended to have predictive value in that it's predicting what would happen if you dropped player x into any lineup (obviously that's extremely generalized) and the only difference is that instead of using prior season's data it sets the limit at 0. If that's the case it still isn't telling you 'this is what player x did for his team', it's 'what player x could do for any team.' It may be that there is almost no difference in the end, but it seems to my limited understanding that it could lead to some minor differences in determining who had the best season vs who is the best player.

I don't think the bolded portions are accurately stating what NPI RAPM (or similar metrics) is attempting to define. If that description of its intent appeared somewhere at some time, it seems wholly incomplete.

More accurately I think it's attempting to predict "what would happen if you dropped player x into any lineup, provided he's allowed to play the exact same role with similar(ish) support pieces around him".

I'll use a couple of extreme hypotheticals to illustrate the point......
Suppose peak Ben Wallace is played at center (just like he was in actuality), but he's forced to be the offensive focal point for the team, shouldering ~30% usage......what do you suppose his offensive impact would look like? I'm guessing it would be godawful in an historic sense (whereas in actuality, his teams utilized him in such a way----extreme low usage, rare opportunistic uses of him----that he basically managed to be [roughly] an offensive neutral according to impact data).

Or suppose we drop Steph Curry on to a different team......but force him to play center. i.e. he doesn't get to be the primary ball-handler on offense, he doesn't get to operate from the outside (except of rare floor-stretching sets or pick n' pops----and how effective do you think a screen set by the 6'3" 180 lb Curry would be, btw?----and similar; he's otherwise positioned in the post or maybe high post on offense). Defensively he's forced to guard one of the bigs. How do you suppose his impact would look in that circumstance? I would say he likely goes from being best in the league down to a net NEGATIVE impact player, because this is an obvious and exaggerated mis-use of his abilities.

Obviously, it's an unrealistic scenario (no one would ever utilize him in this fashion); but it hopefully illustrates the point that how a player is utilized effects his impact on the scoring margin. In an exaggerated case it can even turn the league MVP into a negative-impact player. Smaller [and more realistic] changes to a player's utilization and circumstance can naturally produce smaller changes in their impact potential.

Thanks!

That was my take on it too, but wasn't thinking of the contextual/role portion of it.

[spearsy23]

I'd love a better explanation on that again. I know RAPM is more predictive, and APM varies way more. I know RPM isn't a predictor and is just a 1 year version of xRAPM, but forget some of the RAPM vs. APM stuff since I don't see much APM (non RAPM). I've always preferred a stat that's telling me more of "what's happining now or has happened" vs. "what will happen" which I think is why I like RPM more.

The way I understood it, which could be completely wrong , is that even NPI-RAPM (and thus RPM) is still intended to have predictive value in that it's predicting what would happen if you dropped player x into any lineup (obviously that's extremely generalized) and the only difference is that instead of using prior season's data it sets the limit at 0. If that's the case it still isn't telling you 'this is what player x did for his team', it's 'what player x could do for any team.' It may be that there is almost no difference in the end, but it seems to my limited understanding that it could lead to some minor differences in determining who had the best season vs who is the best player.

I don't think the bolded portions are accurately stating what NPI RAPM (or similar metrics) is attempting to define. If that description of its intent appeared somewhere at some time, it seems wholly incomplete.

More accurately I think it's attempting to predict "what would happen if you dropped player x into any lineup, provided he's allowed to play the exact same role with similar(ish) support pieces around him".

I'll use a couple of extreme hypotheticals to illustrate the point......
Suppose peak Ben Wallace is played at center (just like he was in actuality), but he's forced to be the offensive focal point for the team, shouldering ~30% usage......what do you suppose his offensive impact would look like? I'm guessing it would be godawful in an historic sense (whereas in actuality, his teams utilized him in such a way----extreme low usage, rare opportunistic uses of him----that he basically managed to be [roughly] an offensive neutral according to impact data).

Or suppose we drop Steph Curry on to a different team......but force him to play center. i.e. he doesn't get to be the primary ball-handler on offense, he doesn't get to operate from the outside (except of rare floor-stretching sets or pick n' pops----and how effective do you think a screen set by the 6'3" 180 lb Curry would be, btw?----and similar; he's otherwise positioned in the post or maybe high post on offense). Defensively he's forced to guard one of the bigs. How do you suppose his impact would look in that circumstance? I would say he likely goes from being best in the league down to a net NEGATIVE impact player, because this is an obvious and exaggerated mis-use of his abilities.

Obviously, it's an unrealistic scenario (no one would ever utilize him in this fashion); but it hopefully illustrates the point that how a player is utilized effects his impact on the scoring margin. In an exaggerated case it can even turn the league MVP into a negative-impact player. Smaller [and more realistic] changes to a player's utilization and circumstance can naturally produce smaller changes in their impact potential.

That's why I put 'obviously an extreme generalization' in there. I just didn't feel like making a 5 sentence post a multiple paragraph one

[Atmanne]

[tweet]https://twitter.com/JerryEngelmann/status/672533537053138945[/tweet]
[tweet]https://twitter.com/JerryEngelmann/status/672534079489843200[/tweet]

[PCProductions]

Any Grizzlies fans around to explain why Chalmers' impact is so high?

[Texas Chuck]

Any Grizzlies fans around to explain why Chalmers' impact is so high?

the one thing that jumps out is how often he is getting to the FT line. this is a guy who shoots 4 FT per 100 possessions for his career but in his limited time in Memphis he is shooting almost 15!?!?! And he's making 86% again a pretty nice bump from his career mark.

[SideshowBob]

Any Grizzlies fans around to explain why Chalmers' impact is so high?

Still early in the season, so variance is playing a part here.

Memphis Pre-Chalmers: 3-5, -9.8 MOV, -10.4 eDiff (GSW blowout is influencing heavily here)
Memphis With Chalmers: 8-3, +2.5 MOV, +3.7 eDiff

Single-year RAPM is noisy in full-year samples, so in ~20 games we can observe stuff like this (example: Delly's getting some of Lebron's defensive credit in RAPM/RPM right now, even multi-year). There's a clear shift in Memphis's play if you split the season in half, and right now the model probably thinks Chalmers is the easiest answer as to who's driving that shift (even though there's obviously more factors than just him). He does bring something they desperately needed in 3PT shooting though. Of their best spaced lineups, he's probably featured in most of them.

[bballexpert]

http://espn.go.com/nba/statistics/rpm/_/sort/DRPM Tim Duncan still leading the drpm 20 games in and top 10 over all. I think he may have the best defense longevity there has ever beeen. I have Duncan high on the top 10 defense players and Hakeem and DRob peaks i may put over him although they are close but Longevity of his defense can not even be touched really.

[Doctor MJ]

I was looking through the old APBmetrics topics (specifically the original RAPM one) and it made me wonder why this board seemed to have an affinity for RAPM over APM. It seems that RAPM has always been intended as a predictor more so than a summation of past events, so what about it makes it the preferred stat when comparing current/historical seasons? What exactly are the advantages of NPI-RAPM over APM?

I'd love a better explanation on that again. I know RAPM is more predictive, and APM varies way more. I know RPM isn't a predictor and is just a 1 year version of xRAPM, but forget some of the RAPM vs. APM stuff since I don't see much APM (non RAPM). I've always preferred a stat that's telling me more of "what's happining now or has happened" vs. "what will happen" which I think is why I like RPM more.

I don't actually tend to see it in terms of what is happening vs what will happen, so much as that judging a metric by its predictive value gives you a sense for how much noise there is a given metric. To be clear: That's not how others would see it, but what these folks are doing is essentially demonstrating reliability and asserting validity from that, whereas I tend to roll my eyes at any box score all-in-one metric claiming much validity and I'll continue to do so until player tracking data gets really good.

So yeah in terms of these stats, let me see if I can boil it down:

Validity is how closely the results are inherently tied to helping a team win presuming you have ample signal/noise.
Reliability is how consistent the results are, meaning you won't need to worry about signal/noise as much.

In terms of validity: APM > PI RAPM > NPI RAPM > RPM
In terms of reliability: RPM > PI RAPM > NPI RAPM > APM

You'll note that PI beats NPI by both methods the way I've ranked them, so what's the benefit of NPI? Well PI beats NPI based on using stuff from previous years to shape the results, so any time you're judging a player whose prior years aren't a good way to understand his current play, NPI >> PI.

I'll note one other thing: Despite my last paragraph, I can understand someone saying that on average NPI > PI in terms of validity. It's actually pretty debatable. What's not debatable are the specific issues involved. NPI has more noise issues in general, PI is more problematic with rapidly developing players. But there's a 3rd issue involving NPI RAPM in the sense that it tends distrust outliers even when the outlier is legit. As such while it may actually be that a detailed study would rate the validity of NPI higher on average, given that we're disproportionately using this data on outlier players, I'll end up using PI more than NPI presuming I have both available to me.

[JLei]

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

Updated.

Curry still +11.
Lowry at +10 moves into 2nd.

Lowry is having a ridiculously good season. Proud to have him as a Raptor and have been on the bandwagon of this dude is good since we acquired him, never thought he'd be this good though. But he always had the ability to pull from 3 (and not be a terrible shot) which due to Curry changing the way coaches think has become a more acceptable shot however due to his height can't get it off the same way (he needs space). His weight loss (micro booty Lowry!) and extra speed gear has made teams scared of his penetration and closing out weaker and giving him more cushion as well as going under screens which has allowed him to hit 3 3's a game on 42%.

Westbrook, Kawhi, Bron, Draymond round out the top 6 which is not surprising to anyone.

[dontcalltimeout]

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

Updated.

Curry still +11.
Lowry at +10 moves into 2nd.

Lowry is having a ridiculously good season. Proud to have him as a Raptor and have been on the bandwagon of this dude is good since we acquired him, never thought he'd be this good though. But he always had the ability to pull from 3 (and not be a terrible shot) which due to Curry changing the way coaches think has become a more acceptable shot however due to his height can't get it off the same way (he needs space). His weight loss (micro booty Lowry!) and extra speed gear has made teams scared of his penetration and closing out weaker and giving him more cushion as well as going under screens which has allowed him to hit 3 3's a game on 42%.

Westbrook, Kawhi, Bron, Draymond round out the top 6 which is not surprising to anyone.

As someone who was not been able to watch many Raptors games this year, I'm wondering if you think Lowry may actually be this good? Is he actually in the top tier of players this year? And if so, what / how has he improved?

From what I can tell he's never really approached this level.