RAPM last season for recent acquisitions of some teams

[Crow]

Mavs’ main 5 new guys break out as 3 positive, 2 negative last season. Monta Ellis was actually the best at +1.1.

Rockets got the highest rated player who moved in Howard.

Hawks were a very close second (0.1 difference) with Milsap at less than half the cost. Where were the other / higher bids of advanced analytics teams or any other teams?

GSW- 2 pos. / 2 neg.

Clips- 1 pos. / 3 neg.

Lakers- 3 negative. One of them quite bad (3) in Nick Young, one among the very worst (-5) in Wes Johnson. Sign of tank, sign of stupidity or sign of just not valuing RAPM?

Jazz- 4 negative.

Nets- 3 pos. / 2 neg.

I didn’t rate teams by RAPM of guys they let go. It would be worth doing but I’ll probably let it go for now with this much.

[Saints14]

Love what the Jazz did though. They weren't gonna use that cap space anyway, might as well pick up a few assets for it. And maybe improve draft slot along the way.

[nbacouchside]

Nets acquisition of Pierce (+3.8), Garnett (+4.1), Terry (-1.6) sending out Humphries (-1.0), Wallace (+1.8), Bogans (-3.9), Brooks (-4.1)

Pretty big win for Nets, obviously.

[colts18]

Where are you getting the 2013 RAPM data?

[Crow]

As I noted in another recent thread, Jerry has added back the 2012-3 data on his site.

http://stats-for-the-nba.appspot.com/ratings/2013.html

Or look up a particular team from the front page.
http://stats-for-the-nba.appspot.com/

[nanawacin]

As I noted in another recent thread, Jerry has added back the 2012-3 data on his site.

http://stats-for-the-nba.appspot.com/ratings/2013.html

Is that real RAPM (based on play by play data) or "fake" (based on box score)?

[Crow]

I believe it is "real" RAPM.

[Crow]

Bottom 25 on 2012-3 RAPM (estimated –4.4 to –6.4 per 100 possessions)

Hawks 4
Suns 3
Cavs 2
Hornets 2
Pacers 1
Bucks 1
Kings 1
Bulls 3
Lakers 2
Griz 1
Blazers 1
76ers 1
Knicks 1
Bobcats 1
Heat 1




Perimeter 19
Big men 6

East 15
West 10

“Analytic” teams 12
Not recognized as “analytic” 13

"black" player 25
"white" player 0

Black head coaches 14
White head coaches 11

Top basketball exec played NBA 8
Top basketball exec didn’t play NBA 17

Made playoffs 9
Lottery 16

Retained by last year’s team at least 10



Top 25 on RAPM

Heat 3
Thunder 2
Clips 2
Spurs 1
Griz 2
Lakers 1
Jazz 1
Knicks 1
Rockets 2
Raptors 1
Cavs 1
Pacers 1
Celtics 2
Bulls 1
Warriors 1
Nuggets 1
Mavs 1
Timberwolves 1




Perimeter 10
Big men 15

East 10
West 15

“Analytic” teams 18
Not recognized as “analytic” 7

"black" 19
"white" 6

Black head coaches 9
White head coaches 16

Top basketball exec played NBA 4
Top basketball exec didn’t play NBA 21

Made playoffs 21
Lottery 4

Retained by last year’s team 20

[nanawacin]

I believe it is "real" RAPM.

So for 90s it's also real RAPM... Why there's no separation between what is real RAPM and what fake RAPM?

[v-zero]

It is xRAPM I.e. RAPM blended with a box score metric as well as player height. It is superior to raw RAPM.

For the 90s it is the same, except the RAPM part of the blend is replaced by something akin to RAPM but using box scores. I don't know why the two aren't clearly separated, but I'm sure there is no mal intent.

[nanawacin]

It is xRAPM I.e. RAPM blended with a box score metric as well as player height. It is superior to raw RAPM.

Superior as tool for predictions, but worse in explaining past - xRAPM's individual defensive values are screwed (for example as all box score metrics it overvalues good defensive rebounders)

[v-zero]

Superior as tool for predictions, but worse in explaining past - xRAPM's individual defensive values are screwed (for example as all box score metrics it overvalues good defensive rebounders)

Do you have any evidence for that assertion?

[nanawacin]

So please, explain how xRAPM assigns defensive credit.

BTW, for example 2012 defensive xRAPM vs 2012 defensive real RAPM:

Sefolosha 2nd (real RAPM, non prior informed) defesively vs 32nd (xRAPM)
Deng 5th vs 35th
Al Jefferson 404th vs 116th
McGee 456th vs 50th

So typical for box score based metrics - overeating bigs (height, DRB, BLK)

[v-zero]

By my understanding xRAPM uses a blend of a box score model and RAPM to come to values with the best out of sample prediction ability the two can have combined. Whilst box score metrics are biased against perimeter defenders/in favour of interior defenders, RAPM itself is biased against players based on sample size. Whilst one is a systematic bias against a certain player type and the other is a bias against players who don't get much court time, they both introduce bias. This gets into the issue of the bias-variance trade off, which is a very useful thing to read about if you haven't.

However, to simplify the conversation I will just say this: better prediction accuracy implies a more accurate estimate of player ability, on average. Certainly some players will be more wrong in the xRAPM estimates than the RAPM estimates, but a greater number will be more wrong in the RAPM estimates than the xRAPM estimates.

So, they're both wrong, but overall xRAPM is less wrong than RAPM - so why choose (short term) RAPM over (short term) xRAPM?

I'm not advocating for either, I'm just trying to explain that things aren't black and white.

[nanawacin]

However, to simplify the conversation I will just say this: better prediction accuracy implies a more accurate estimate of player ability, on average.

Yes, but overall ability. However we don't know how accurate are offense-defense splits and that's the whole point of using real RAPM - it describes individual defense better (of course I can't prove it, that's just my feeling, however examples I showed above kind of confirm that and there's many other such examples).