2001-2014 RAPM (Updated now to 1997-2014 RAPM)

[colts18]

Do you have offense/defense per 100 for notable players and other guys in the top 10 of each?

Yes. Its listed on the website

offense/100
https://sites.google.com/site/rapmstats ... nsive-rapm

defense/100
https://sites.google.com/site/rapmstats ... nsive-rapm

overall/100
https://sites.google.com/site/rapmstats/01-14-rapm

offense:
Name Off/100
LeBron James 4.63
Dwyane Wade 4.31
Steve Nash 4.10
Manu Ginobili 3.79
Kobe Bryant 3.75
James Harden 3.53
Dirk Nowitzki 3.51
Shaquille O'Neal 3.36
Chris Paul 3.31
Baron Davis 2.80



Defense
Name Def/100
David Robinson 3.62
Dikembe Mutombo 3.40
Tim Duncan 3.28
Jason Collins 3.26
Shawn Bradley 3.24
Kevin Garnett 2.97
Tony Allen 2.67
Amir Johnson 2.67
Jorge Garbajosa 2.66
Omer Asik 2.66

[SideshowBob]

I'm still wary of even using the single-year NPI data. Someone correct me if I'm wrong, but doesn't it pose the same issue as single-year APM? And again, mixing that along with a set of years of prior-informed and then also including 2014 Vanilla RAPM AND a separate 2013 dataset is just a big red flag. I just don't understand how it makes sense to do what colts has done (great effort btw, I'm not questioning your intent at all), without separating out all of the sets.

[Doctor MJ]

I'm still wary of even using the single-year NPI data. Someone correct me if I'm wrong, but doesn't it pose the same issue as single-year APM? And again, mixing that along with a set of years of prior-informed and then also including 2014 Vanilla RAPM AND a separate 2013 dataset is just a big red flag. I just don't understand how it makes sense to do what colts has done (great effort btw, I'm not questioning your intent at all), without separating out all of the sets.

Hmm.

-RAPM has the same issues as APM period, just less so, but while adding other issues in. PI is going to have less noise than NPI, but it's also going to reward & penalize guys under the assumption that they'll perform like the previous year which won't always give the right result.

-Re: why combine the sets given they have different methods in them? Well I think ideally he should start doing some normalization between them if he hasn't already, but beyond stuff like that, if you want to do a 13 year study and this is the data that's available for those 13 years, to me it makes sense to work with what you got.

[MisterWestside]

To your last sentence: WAR is only useful for impact too, and there's certainly more to baseball than that, albeit not as much more as basketball has.

More interesting point: Would it be better to use XRAPM or something along those lines for this kind of analysis than RAPM which by definition is not the best indicator by itself?

It would certainly be good to see it done for XRAPM as well. On a simplistic level, what we are doing in this analysis is treating RAPM as an all-in-one stat, and XRAPM is actually the attempt to use RAPM to make an all-in-one stat. I'll be honest though: I don't know if I'd feel as comfortable analyzing those results as I do with a honed form of this analysis for the same reasons I'm reticent using XRAPM generally: I don't use anything as a pure all-in-one, and I think it's easier to mentally mod RAPM than XRAPM.

Yep, you're right on the money about WAR. RAPM does the exact same thing that all the other stats attempt to do though: summarize individual player value in a single number. It just doesn't use the box score. But the goal stays the same.

Personally, I'm past the all-in-one metric fad. I use so many sources of information for player evaluation that you'd be doing yourself a disservice to just use one stat. If I *have* to pick just one metric to use, it would be xRAPM. And that's if it's available for the given season. If it's not, then I'd use one of the box-score ASPM derivatives that are on the web. It's in the same ballpark as prior-informed RAPM while being less of a hassle to compute. It's a step-down from xRAPM, but whatever. Thankfully, I use other metrics as well as my own set of eyes and basketball common sense to complete the picture (instead of just using a box-score RAPM hybrid stat), which is what people should do anyway.

[giberish]

1361 Troy Murphy -808.4
1362 J.J. Hickson -811.0
1363 Hakim Warrick -910.7
1364 DeShawn Stevenson -941.4
1365 Juwan Howard -952.8
1366 Drew Gooden -963.5

Really get a kick out of this list.

Drew Gooden is the Least Valuable Player of the +/- era, and also a player that certain segments wondered for quite a long time why he kept getting traded despite decent stats. Coaches who had him knew what was up, and coaches who hadn't had him yet lived in hope.

There seems to be a clear weak spot NBA teams have for bigs who get rebounds without helping team rebounding and also play surprisingly bad defense (the type that teams often have to see for themselves to believe).

Gooden, Howard, Hickson and Murphy all fit that profile. Hickson just got a significant contract last summer (and is a serious threat to pass Gooden in a few years) so the issues hasn't gone away. Mo Speights as well, though he likely won't get the PT to climb that high on this list.

[Doctor MJ]

To your last sentence: WAR is only useful for impact too, and there's certainly more to baseball than that, albeit not as much more as basketball has.

More interesting point: Would it be better to use XRAPM or something along those lines for this kind of analysis than RAPM which by definition is not the best indicator by itself?

It would certainly be good to see it done for XRAPM as well. On a simplistic level, what we are doing in this analysis is treating RAPM as an all-in-one stat, and XRAPM is actually the attempt to use RAPM to make an all-in-one stat. I'll be honest though: I don't know if I'd feel as comfortable analyzing those results as I do with a honed form of this analysis for the same reasons I'm reticent using XRAPM generally: I don't use anything as a pure all-in-one, and I think it's easier to mentally mod RAPM than XRAPM.

Yep, you're right on the money about WAR. RAPM does the exact same thing that all the other stats attempt to do though: summarize individual player value in a single number. It just doesn't use the box score. But the goal stays the same.

Personally, I'm past the all-in-one metric fad. I use so many sources of information for player evaluation that you'd be doing yourself a disservice to just use one stat. If I *have* to pick just one metric to use, it would be xRAPM. And that's if it's available for the given season. If it's not, then I'd use one of the box-score ASPM derivatives that are on the web. It's in the same ballpark as prior-informed RAPM while being less of a hassle to compute. It's a step-down from xRAPM, but whatever. Thankfully, I use other metrics as well as my own set of eyes and basketball common sense to complete the picture (instead of just using a box-score RAPM hybrid stat), which is what people should do anyway.

The entire reason why I prefer RAPM to xRAPM is because I don't use it acontextually, and no one serious uses anything in such a way.

Look, not being that interested in a list that tallies up RAPM alone to glance at players is something I do understand. I feel I can get good information out of it, but it's not perfect. What I don't really understand from your description of a fad though is what you replace something like RAPM with. You can't seriously think it's better to use +/- stats without some form of APM to go with that, can you? You can't seriously think it's better to ignore +/- stats altogether, can you? I just see these as tools to use that give information you can't get elsewhere.

[MisterWestside]

The entire reason why I prefer RAPM to xRAPM is because I don't use it acontextually, and no one serious uses anything in such a way.

Look, not being that interested in a list that tallies up RAPM alone to glance at players is something I do understand. I feel I can get good information out of it, but it's not perfect. What I don't really understand from your description of a fad though is what you replace something like RAPM with. You can't seriously think it's better to use +/- stats without some form of APM to go with that, can you? You can't seriously think it's better to ignore +/- stats altogether, can you? I just see these as tools to use that give information you can't get elsewhere.

Oh no, I wasn't saying that I'd just use one stat. Since the topic was about all-in one metrics though, I was making my choices from the supposition that I could use only use one stat in player evaluation (based on the predictive power of the metrics). Some versions of SPM have done quite well in retrodiction stat-offs while being way easier to compute than APM or RAPM, which is useful especially since APM/RAPM isn't always available.

Of course, I wouldn't be using one metric

[Doctor MJ]

The entire reason why I prefer RAPM to xRAPM is because I don't use it acontextually, and no one serious uses anything in such a way.

Look, not being that interested in a list that tallies up RAPM alone to glance at players is something I do understand. I feel I can get good information out of it, but it's not perfect. What I don't really understand from your description of a fad though is what you replace something like RAPM with. You can't seriously think it's better to use +/- stats without some form of APM to go with that, can you? You can't seriously think it's better to ignore +/- stats altogether, can you? I just see these as tools to use that give information you can't get elsewhere.

Oh no, I wasn't saying that I'd just use one stat. Since the topic was about all-in one metrics though, I was making my choices from the supposition that I could use only use one stat in player evaluation (based on the predictive power of the metrics). Some versions of SPM have done well in retrodiction stat-offs while being way easier to compute than APM or RAPM, which is useful especially since APM/RAPM isn't always available.

Of course, I wouldn't be using one metric

I'll clarify:

I know full well that you don't want to use just one metric, I'm rejecting your assertion that the goal of people using RAPM is different from yours, and questioning if you truly don't use RAPM at all.

[MisterWestside]

I'll clarify:

I know full well that you don't want to use just one metric, I'm rejecting your assertion that the goal of people using RAPM is different from yours, and questioning if you truly don't use RAPM at all.

I do use RAPM. My comment was based on what I've seen some people do with these stats (not on this board, but in some basketball circles) when they become available to the public - that is, they use a stat that summarizes player value as the be-all, end-all stat or something similar to it. I know you don't do that.

Whenever a new all-in-one metric hits the web, however, that's what happens sometimes with some fans or even analysts of the game. So while I find the idea of RAPM VORP neat, I've also seen this movie before. Hence the fad quip

[ceiling raiser]

I really dig RAPM, but until we have some uniformity (with regards to priors, weighting, lambdas, etc.) in each season for which it's possible (96-97 to the present), I'm skeptical of its use. Admittedly, I'm much more of a fan of dRAPM than oRAPM since it provides much more information.

As is the case with MisterWestside (I think from skimming, though I haven't read the last few pages too closely), in general I'm moving away from all-in-one or overly encompassing metrics, as an initial pass. I'm really trying to identify roles/skillsets first, and then delve into ascertaining impact from these sorts of stats afterwards.

[acrossthecourt]

Wow. So Dirk is underrated defensively or it is bcoz he played a defensive center his almost entire career?

Sent from my V790 using RealGM Forums mobile app

There's a reason why "serious" basketball analysts hate chopping up the game by offense/defense: there's an interaction between offense and defense that's hard to separate. For example, Dirk being an extremely low turnover player, especially for a big man, helps out his defense because liveball turnovers hurt the defense (transition opportunities.)

[O_6]

Just want to discuss LeBron's defense again. These are the top 10 perimeter defenders since 2001 according to RAPM. I included Josh Smith because he's kind of a hybrid and more of a PF but we've seen him guard guys like Kobe/Melo/Durant before so I think he deserves to be on this list.

Defensive RAPM Above Average: Since 2001
Player -------------------- RAPM -------- Mins
Metta World Peace ---- 1277.9 ------- 28,452
Andre Iguodala --------- 1202.6 ------- 27,520
Shane Battier ----------- 1087.3 ------- 29,554
Luol Deng ---------------- 1026.3 ------ 23,584
Bruce Bowen ------------- 948.4 ------- 21,374
Josh Smith ---------------- 887.2 ------- 25,119
Gerald Wallace ---------- 880.9 -------- 24,412
Jason Kidd ---------------- 879.9 ------- 34,651
LeBron James ------------ 713.7 ------- 32,316
Shawn Marion ------------ 702.5 ------- 36,964

LeBron is 9th on that list, despite playing the 3rd most minutes.

How many of those guys do you think were more valuable defenders than LeBron?

I think I'd take World Peace, Iggy, Battier, Deng, and Bowen over LeBron. LeBron's offensive responsibility forced him to conserve more energy on D than those 5 guys. I think Deng is the weakest of that group though, I could see an argument for LeBron over him.

Smith/Wallace/Marion are kind of hybrid defenders. I don't know how to rank LeBron among these guys. I think I'd take LeBron over all of them because he's a better and more versatile man defender imo. But those guys were all vicious help defenders and better rebounders than Lebron. And due to less offensive responsibility they probably played 100 on D more often.

Kidd is the other guy. Just weird to compare him to LeBron as a defender. Kidd was an incredible defender with God-like hands, but I just don't think he was as valuable as LeBron on D. But again, hard for me to compare them as defenders.

[lorak]

^
We have Engelmann's '02-'11 RAPM study and the best perimeter defenders (+ KG and Duncan for reference point) are:

Code: Select all

rank   player   DRAPM
 1   Garnett   5,0
 4   Duncan   3,2
 5   Artest   3,2
13   Bo Outlaw   2,4
15   T. Allen   2,3
16   Christie   2,3
18   LeBron   2,2
32   Deng   1,8


Looks legit for me. Artest was definitely the best perimeter defender during last decade and the rest is a level below him and LeBron among them.

[Quotatious]

It's amazing how much better Garnett is than Duncan by that metric. I still think they're basically equal on defense, but there's some evidence that KG may be better, and perhaps the most impactful post-Russell...Or at the very least, on par with Hakeem and D-Rob.

LeBron's score is still very impressive when you consider that he's been considered as a great defender only since the 2008-09 season, and pretty much average before that.

[Doctor MJ]

colts18,
could you explain how you did that? Use Durant as example and show step by step how you get his total value, please.

Here is the Kevin Durant data. Above average is RAPM / 100 * Possessions played

Code: Select all

year   poss   Off per 100   Def per 100   Ovr per 100  off   def   above avg   above rep
2008   5581.5   -1.5   -3   -4.5   -83.7   -167.5   -251.2   -195.4
2009   5656   -1.8   -2.8   -4.6   -101.9   -158.2   -260.2   -203.6
2010   6777.5   2.9   1.2   4.1   196.4   81.4   277.9   345.7
2011   7315.5   3.9   0.6   4.5   286.0   43.8   329.2   402.4
2012   6585   3.5   -0.1   3.4   230.9   -6.6   223.9   289.7
2013   5475   3.0   0.0   3.0   161.9   1.8   163.7   218.5
2014   3925.5   3   0.8   3.8   117.8   31.4   149.2   188.4


It occurred to me to add a thing: When I evaluate players, years like Durant's first two literally mean 0 to me. I don't penalize Durant for the negative value added.

Two components of this:

1) I emphasize the best years more than the worst years, so on that alone padding longevity with meh years is pretty ignorable.

2) OKC literally wasn't trying to win with this approach. Durant wasn't getting in their way of winning, getting him ready was actually their goal because they knew trying to "win now" at the time made no sense.

Take these years out and his cume RAPM*possession figures go up quite a bit. The above average cumes literally go up from around 600 to around 1100. Guys like Tony Parker & Pau Gasol with far more experience only rank up around 1300 with that metric. Still, in comparison the heavy hitters on the list, Durant remains pretty far down. There's just no substitute for longevity.

[Doctor MJ]

It's amazing how much better Garnett is than Duncan by that metric. I still think they're basically equal on defense, but there's some evidence that KG may be better, and perhaps the most impactful post-Russell...Or at the very least, on par with Hakeem and D-Rob.

LeBron's score is still very impressive when you consider that he's been considered as a great defender only since the 2008-09 season, and pretty much average before that.

It's telling to me that Garnett dwarfs Duncan in a 10-year-together study but Duncan has the clear edge taking it year by year. I'd say it's without question that if you agree to take a guys defensive achievement by summing up each year, colts' approach gives the more accurate results.

I'm trying to think how to put this: In my experience, in any given regression analysis there are essentially "anchors" the develop while the players who played with them fall in line based on how they relate to the anchor. In the case of prior-informed studies this is directly tied to the prior itself but it will happen regardless. So for example, Dirk really seems to be hurt by the fact that Nash went on to produce astronomical offensive impact in Phoenix in any study that takes that shift into account, and he does much better in those studies which just don't happen to include that sample.

The way that Boston's defense had a GOAT turnaround with Garnett's arrival seems to me like it caused everything else to be re-calibrated and given the difference I see in a study like colts', I think those re-calibrations go too far. Now to be fair, any way you look at it Garnett's impact on Boston's defense was tremendous year after year on a level that I believe surpasses Duncan's defensive impact. My statement here isn't "So Duncan clearly was truly the better defender" so much that Garnett may have been the more able defender given his Boston focus but this doesn't reflect his cumulative defensive impact over his whole career to anywhere near the extent the 10-year-together study indicates.

[lorak]

It's amazing how much better Garnett is than Duncan by that metric. I still think they're basically equal on defense, but there's some evidence that KG may be better, and perhaps the most impactful post-Russell...Or at the very least, on par with Hakeem and D-Rob.

LeBron's score is still very impressive when you consider that he's been considered as a great defender only since the 2008-09 season, and pretty much average before that.

It's telling to me that Garnett dwarfs Duncan in a 10-year-together study but Duncan has the clear edge taking it year by year. I'd say it's without question that if you agree to take a guys defensive achievement by summing up each year, colts' approach gives the more accurate results.

I disagree. Even if we ignore issues with colts' approach (different RAPM models treated as the same, why PI over NPI?), then it's like saying that for example KAJ is better scorer than MJ, because he has more total points.

[Doctor MJ]

It's amazing how much better Garnett is than Duncan by that metric. I still think they're basically equal on defense, but there's some evidence that KG may be better, and perhaps the most impactful post-Russell...Or at the very least, on par with Hakeem and D-Rob.

LeBron's score is still very impressive when you consider that he's been considered as a great defender only since the 2008-09 season, and pretty much average before that.

It's telling to me that Garnett dwarfs Duncan in a 10-year-together study but Duncan has the clear edge taking it year by year. I'd say it's without question that if you agree to take a guys defensive achievement by summing up each year, colts' approach gives the more accurate results.

I disagree. Even if we ignore issues with colts' approach (different RAPM models treated as the same, why PI over NPI?), then it's like saying that for example KAJ is better scorer than MJ, because he has more total points.

I didn't say it said Duncan was a better defender.

[colts18]

I disagree. Even if we ignore issues with colts' approach (different RAPM models treated as the same, why PI over NPI?), then it's like saying that for example KAJ is better scorer than MJ, because he has more total points.

Duncan has KG beat in defensive RAPM stats even by using NPI. Duncan has a huge lead when they were in their primes by NPI RAPM.

[Doctor MJ]

I disagree. Even if we ignore issues with colts' approach (different RAPM models treated as the same, why PI over NPI?), then it's like saying that for example KAJ is better scorer than MJ, because he has more total points.

Duncan has KG beat in defensive RAPM stats even by using NPI. Duncan has a huge lead when they were in their primes by NPI RAPM.

Eh,

First, you might want to make sure you answer his question. My assumption is that you used PI over NPI in part to have more consistent data since PI is available for more years, and possibly in part because you (like many) think using priors is a good things.

Second, given that we only have NPI in Garnett's pre-Boston years, this doesn't actually address Garnett's defensive "prime" in the sense of when he was having the most impact.

Third, Duncan did not have a huge NPI RAPM lead over Garnett in the years where there was serious debate between the two. Their '03 & '04 numbers are basically identical. I agree if what you want to say is that by any cume measure Duncan's going to come out ahead, but there are real indicators saying that Garnett in a right situation is right up there with Duncan...who of course lucked into having the "most right of a situation" one could realistically imagine.