APBRmetrics

permaximum wrote:BTW I wonder, why is Jordan placed 36th in 15-year age adjusted RAPM while he was placed 268th in 14-year age adjusted RAPM?

The 14y RAPM was done in the following way:
1. adjust for age when doing the regression. This was done so e.g. a random player's estimate doesn't suffer from playing with 36-year-old instead of 29-year-old Shaq.
This is essentially equivalent to saying 'what would be everyone's performance if they were all the same age'

2. Because we were interested in everyone's performance during that timeframe, though, I factored everyone's age back in after the regression

This is more "fair" than leaving age out completely, as your teammates don't get dinged for playing with a 40y old you, only yourself get dinged for being 40


The 15y RAPM I posted above does just 1.

J.E., on the ESPN RPM site it has Curry at +8.95 and LeBron at +8.50. It says it was updated at June 23 but I remember those two having that same exact number before the finals. Does RPM get updated with playoff info?

J.E. wrote:

permaximum wrote:BTW I wonder, why is Jordan placed 36th in 15-year age adjusted RAPM while he was placed 268th in 14-year age adjusted RAPM?

The 14y RAPM was done in the following way:
1. adjust for age when doing the regression. This was done so e.g. a random player's estimate doesn't suffer from playing with 36-year-old instead of 29-year-old Shaq.
This is essentially equivalent to saying 'what would be everyone's performance if they were all the same age'

2. Because we were interested in everyone's performance during that timeframe, though, I factored everyone's age back in after the regression

This is more "fair" than leaving age out completely, as your teammates don't get dinged for playing with a 40y old you, only yourself get dinged for being 40


The 15y RAPM I posted above does just 1.

That's what I thought exactly. Thanks for the clearence. The two steps you mention seem crucial for something like this-long RAPM if you're analyzing the player performance in that timeframe. Any chance we can see the results for the 15-year one?

Thank you JE.

I don't have a request but a question. Do you ever see RAPM (in any of your variants) being hosted on a website where it would be indexed, organized and easy to sort? I often want to use RAPM for analysis, but I find it too annoying to actually use (things like - seeing the 2013 Raptors RAPM numbers or looking at the RAPM numbers each year in Kyle Lowry's career).

colts18 wrote:J.E., on the ESPN RPM site it has Curry at +8.95 and LeBron at +8.50. It says it was updated at June 23 but I remember those two having that same exact number before the finals. Does RPM get updated with playoff info?

RPM was being updated every few weeks during the postseason. The current set listed on ESPN is the pre-finals one, correct. It's been getting updated periodically all season so I expect a final 2015 update within the next week.

Hi J.E., on your vanilla multi year RAPM can you tell me which Chris Johnson is the one that played this year in (MIL/PHI/UTA), I believe you have the other Chris Johnson listed as well with one player at -.49 and the other at -1

Here's what I need:
-A 2014-2015 RAPM (two seasons) for SportVU and other recent data. However, I want it to include other seasons beyond those two, but weigh 2014 and 2015 more heavily. You already do similar vanilla RAPMs, so you can use your judgement here but at least cite which seasons are actually used. (Maybe 2011 to 2015 would work well? 2011 weighed the least.) Complete age adjustments would be useful needed too.

-A 2005-2015 RAPM would be interesting because of how the NBA changed. This would basically be the 3PT era.

My request:
PI (with only one previous season as "prior") and NPI RAPM for each season from 1997 (first year PbP is available) to 2015, no box score component, height or age adjustment, just +/- regression.

Single year NPI for 2001 onwards.

https://www.dropbox.com/sh/11181n4avq5w ... NNHEa?dl=0

Don't have data before that

Hope this works. I currently have all kinds of weird issues with DropBox

J.E. wrote:Single year NPI for 2001 onwards.

https://www.dropbox.com/sh/11181n4avq5w ... NNHEa?dl=0

Don't have data before that

Hope this works. I currently have all kinds of weird issues with DropBox

Thanks J.E., this is great data. Do you have multi-year vanilla RAPM for each of 02-15 (or through 14 I guess, since you kindly shared for 15 the other day)? Also, might be a silly exercise, but I wonder what 15-year (01-15) playoffs possessions only RAPM would look like (if you think it's worth investigating...with or without an age curve, whichever you feel is more informative).

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?

J.E. wrote:Single year NPI for 2001 onwards.

https://www.dropbox.com/sh/11181n4avq5w ... NNHEa?dl=0

Don't have data before that

Hope this works. I currently have all kinds of weird issues with DropBox

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.

J.E. wrote:Single year NPI for 2001 onwards.

https://www.dropbox.com/sh/11181n4avq5w ... NNHEa?dl=0

Don't have data before that

Hope this works. I currently have all kinds of weird issues with DropBox

Thx. Does it include playoffs?

Playoff possessions only, if possible, NPI, but with age curve, whatever ya got, whether it's 2001 onward or...

Trying to answer a couple of questions in this post

First, a clarification:
People have requested "prior informed" RAPM (where the prior stems from the RAPM of season X-1). The "true" way of doing this is to daisy chain the RAPMs. This is something I have done in the past.
The other way of doing this is by using multiple years of data in one regression, and weighing the older years less

I do prefer the latter and don't really see any benefit in doing former. The latter just "makes more sense" from a theoretical standpoint.
For this reason, whenever someone requests "prior informed RAPM", the 2nd version is actually what I'll do. I can assure you that the daisy-chained version offers no real benefit

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

Hi J.E., on your vanilla multi year RAPM can you tell me which Chris Johnson is the one that played this year in (MIL/PHI/UTA), I believe you have the other Chris Johnson
listed as well with one player at -.49 and the other at -1

The Chris Johnson that played last season is the one with the better rating

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

Thx. Does it include playoffs?

It does

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

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

First, a clarification:
People have requested "prior informed" RAPM (where the prior stems from the RAPM of season X-1). The "true" way of doing this is to daisy chain the RAPMs. This is something I have done in the past.
The other way of doing this is by using multiple years of data in one regression, and weighing the older years less

I do prefer the latter and don't really see any benefit in doing former. The latter just "makes more sense" from a theoretical standpoint.
For this reason, whenever someone requests "prior informed RAPM", the 2nd version is actually what I'll do. I can assure you that the daisy-chained version offers no real benefit

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

Thanks for the clarification RE: "prior informed"and "multi year" RAPM (as well as for the RPM note). Whenever you have the chance, it would be awesome if you could share your multi-year sets for each of 02,03,...,13,14 (comparable to this set: https://docs.google.com/spreadsheets/d/ ... edit#gid=0).

BTW I'm not 100% sure, but I think these are the versions colts18 mentioned:

https://sites.google.com/site/rapmstats/2001-npi-rapm
https://sites.google.com/site/rapmstats/2006-npi

Though if the old versions didn't include playoffs, what you said could explain the difference.