APBRmetrics

J.E. wrote:

Is it possible to run a 15 year playoff RAPM using 15 year RS+PS RAPM as the opponent and teammate adjustment?

I don't understand this, please elaborate further

Run a 15 year playoff RAPM. For example, lets look at LeBron

One key to RAPM is to adjust for your teammates and opponents. If you do a 15 year playoff RAPM, then those adjustment values will be the small sample size of playoffs for opponents and teammates. Lets say one of his teammates has a +15 Net in the playoffs. A playoff RAPM would punish LeBron for that even though the normal big sample size 15 year RAPM has that player as average. For the adjustment in the playoff RAPM, put in the teammates and opponents 15 year total RAPM number, not the 15 playoff RAPM number

I'm curious, did you use a different methodology for this NPI than the NPI you ran in the past? For example, when you initially did single year NPI RAPM, you had Shaq at #5 in 2006 with a +4.2 RAPM.
but in this version, he is in 40th place with a +1.92 RAPM. In 2001, you had Shaq with a +2.4 RAPM advantage over Kobe in NPI, but in this version Kobe is ahead of Shaq.

This is interesting and I'll have to look into it. Can you post one of those single year NPI RAPMs so I can compare/play around?
Although the simplest explanation could just be "the last version didn't include playoffs and this one does". Shaq had a -8.6 NET rating in the 2006 playoffs

Here is 2001 RAPM

https://sites.google.com/site/rapmstats/2001-npi-rapm

I'm guessing that it didn't include playoffs where Kobe was around +15 Net and Shaq was negative. That would explain why Kobe was able to catch up from a 2.4 RAPM and surpass Shaq.

Multiyear RAPM with lower weight to older seasons (those who requested PI RAPM should use this instead)

https://www.dropbox.com/sh/teutg7zvxudq ... wS2ja?dl=0

Some observations:

2002
- The Lakers of early 2000 are mostly seen as Shaq/Kobe's team, yet Derek Fisher is actually rated higher than Kobe
- Steve Francis, who many back then saw as a mindless chucker (that's at least how I remember it) ranks in the top 10
- Horace Grant, at age 36, ranks in the top 30. Stockton ranks 3rd at age 39

2003
- Duncan with with one of the highest ratings (in pure RAPM) ever, at 9.1. He was certainly seen as a 2-way player by RAPM, ranking 6th for offense, 1st for defense. Soon, his offense started to decline, but his defense has remained very good

2004
- KG with the highest rating in those years analysed with a +10, ranking 2nd on offense and 3rd on defense. The TWolves were playing like an elite team with him ON the court: +10, and like a lottery team with him OFF: -11
- A couple of not-so-well-knows/surprises near the top: Bo Outlaw, Najera, Eric Williams, Jason Collins, Desmond Mason, Derek Anderson, Brian Cardinal

2006
- Manu Ginobili comes in 1st for a change (from KG, Duncan). Ron Artest is right behind him
- Kirilenko 3rd. He had been near the top for many years in a row
- This year's finals MVP, Iguodala, comes in ~25th

2007
- Ginobili, again, 1st
- Nash near the top for the ~3rd year in a row, just due to his elite offense (his defense is slightly below average)
- Arenas makes an appearance in the top 20

2008
- KG's 1st year in Boston brings him back into form

2010
- The age of LeBron as undisputed #1 starts

2012
- Ekpeh Udoh with a surprise appearance in the top 10. His +/- numbers have since fallen off a cliff

2013
- Nowitzki takes the #1 spot from LeBron for a brief moment
- Chris Paul in the top 5
- Nick Collison in the top 10
- Asik in the top 15

2014
- Iguodala 3rd, Curry 4th

A note on 2015: these are a tiny bit "wider" than the ones I posted earlier. This is because I used a lower penalization value here (2500 v 3000), which I remembered gives slightly better OOS projections when there's more than 1 season of data. It's certainly not a big difference, by any means

J.E., thanks for your work, I really appreciate you are doing this. Two questions:

1. Would you be willing to share the used raw data; e.g. matchupfiles?
2. Could you add a column with the respective amount of possessions for each player?

Also, I just took a small glimpse at some of the results and I saw that the multiyear files contain more than those players for the respective season. Would it be possible to describe which years went into each of those season results with the respective weighting?

Thanks for doing this J.E.

A 2012-2015 NPI RAPM including playoffs and weighted heavily towards 2015 would be great to use for the Vantage data

Thank you J.E.!

Bo Outlaw did really well in the late 90's RAPM. He has the defensive reputation and some of the stats there, like shotblocking. He was really athletic too.

Any chance at doing one where 2014 and 2015 are weighed equally? I guess I could just combine them.

Any chance at doing one where 2014 and 2015 are weighed equally? I guess I could just combine them.

I asked for it too(playoffs included). Combining single-year NPI-RAPM of 2014 and 2015 shouldn't give the same result.

2013-14 + 2014-15 RAPM, equal weights to both years

https://www.dropbox.com/s/ju637ylhtsrjt ... s.txt?dl=0

Lots of young Spurs at the top: Leonard, Green, Mills

Was a little surprised to see Tolliver so high, but OK

J.E. wrote:2013-14 + 2014-15 RAPM, equal weights to both years

https://www.dropbox.com/s/ju637ylhtsrjt ... s.txt?dl=0

Lots of young Spurs at the top: Leonard, Green, Mills

Was a little surprised to see Tolliver so high, but OK

Awesome job J.E. This is great work. How does that 2 year RAPM work? Is it considered one big year or is it 2 separate RAPM's weighed equally? If its the former, I would assume there would be a slight age adjustment needed.

I'm curious about this tweet you had a while back:

Jeremias Engelmann retweeted
Steve Ilardi ‏@dr_ilardi Jun 22

Predictive RPM for next season: #1 LeBron (9.74), #2 Curry (+8.50), #3 Paul (7.83), #4 Kawhi (7.37), #5 Draymond (6.83). h/t @JerryEngelmann

Do you have a list of your predictive xRAPM/RPM for past seasons? I would be interested in a true talent RAPM/xRAPM. That would probably need to not only weigh past seasons but future seasons also (ex: 1x of 2010, 1/2x of 2009 and 2011, 1/4x of 2008 and 2012).

J.E. wrote:2013-14 + 2014-15 RAPM, equal weights to both years

https://www.dropbox.com/s/ju637ylhtsrjt ... s.txt?dl=0

Lots of young Spurs at the top: Leonard, Green, Mills

Was a little surprised to see Tolliver so high, but OK

Thank you very much. Two questions:

1. Does it include playoffs?
2. One big regression or the other approach?

permaximum wrote:

Any chance at doing one where 2014 and 2015 are weighed equally? I guess I could just combine them.

I asked for it too(playoffs included). Combining single-year NPI-RAPM of 2014 and 2015 shouldn't give the same result.

What I meant was multiyear with 2014 and 2015 weighed equally.

AcrossTheCourt wrote:

permaximum wrote:

Any chance at doing one where 2014 and 2015 are weighed equally? I guess I could just combine them.

I asked for it too(playoffs included). Combining single-year NPI-RAPM of 2014 and 2015 shouldn't give the same result.

What I meant was multiyear with 2014 and 2015 weighed equally.

I see. I meant NPI one big regression which treats all possessions in 2014 and 2015 regular/playoffs equally without any adjustments such as age. I guess that's exactly what J.E. shared recently.

J.E. wrote:Trying to answer a couple of questions in this post

Also I don't want to cause any confusion since this thread isn't about RPM/xRAPM, but are these sets (http://stats-for-the-nba.appspot.com/ratings/2001.html through
http://stats-for-the-nba.appspot.com/ratings/2013.html) comparable to the RPM/xRAPM hosted on ESPN?

Yes. Not exactly the same, but comparable

Sorry to drag this further into xRAPM land, but what is this data for years before 2001? I was under the impression the necessary play-by-play data wasn't available to generate APM/RAPM/etc... for 1995 for instance (http://stats-for-the-nba.appspot.com/ratings/1995.html).

J.E. wrote:

colts18 wrote:-15 year RAPM (no age adjustment)

https://docs.google.com/spreadsheets/d/ ... sp=sharing

This one's got a couple of interesting names here and there.

Draymond Green is ranked 8th, Amir Johnson 11th, Baron Davis 12th, Jeff Foster 29th

Zach LaVine and Anthony Bennett in the bottom 20 out of 1467. Jeff Green as well, who was discussed recently.

Not many in the bottom 20 are still in the league, besides those already mentioned: Tony Wroten, JJ Hickson, Norris Cole and Tim Hardaway Jr.

On that list you have a player named "30340" and "30588". Which players are that?

DSMok1 wrote:I'm going to ask a hard one first:

I would like to see an RAPM that is informed by a prior containing only age and MPG information.

EDIT: Perhaps Team Margin should also be included, but not age?

A regression I ran recently on BPM, since 2000:

Code: Select all

SUMMARY OUTPUT

Regression Statistics
Multiple R        0.748050082
R Square          0.559578925
Adjusted R Square 0.559226926
Standard Error    1.927022923
Observations             6262

ANOVA
                      df          SS           MS          F     Significance F
Regression                  5  29516.42523 5903.2850471589.717637           0
Residual                 6256  23231.13891 3.713417345
Total                    6261  52747.56414

                 CoefficientsStandard Erro   t Stat     P-value   Lower 95%   Upper 95%  Lower 95.0% Upper 95.0%
Intercept              -6.938        0.434     -15.990 1.940E-56       -7.789      -6.088      -7.789      -6.088
RegressedMPG            0.289        0.019      14.852 4.618E-49        0.251       0.328       0.251       0.328
Age                     0.041        0.016       2.588 9.677E-03        0.010       0.072       0.010       0.072
Team Mar                0.099        0.015       6.806 1.095E-11        0.071       0.128       0.071       0.128
ReMPG*Age              -0.002        0.001      -3.265 1.100E-03       -0.004      -0.001      -0.004      -0.001
ReMPG*Mar               0.005        0.001       8.056 9.352E-16        0.004       0.006       0.004       0.006

What is your interpretation of or conclusions from this? It would be nice to get a few sentences. Many will be confused or uncertain what to draw from such output, including me in this case. In the name of education, some English summary should be a standard inclusion when posting regression output, I think.