2013 RAPM/IPV/etc.

[lorak]

Parker's DRAPM seems to be the biggest bug here. He's better on defense than on offense, and more impactfull defender than Sanders, Asik or Deng?! No way.

[Doormatt]

how about westbrook being higher than durant offensively and overall?

[CBA]

Parker's DRAPM seems to be the biggest bug here. He's better on defense than on offense, and more impactfull defender than Sanders, Asik or Deng?! No way.

Yeah, I can't square that.

[Doctor MJ]

Severe lack of Amir Johnson in Top 3.

To reiterate:

People like me pushing back against hybrid models - mainly against Engelmann's because of how he did it - isn't based on me saying that on its own other +/- models will typically give better results, but rather based on me already knowing that I"m going to be using box score data in my analysis in addition to the +/- data. Factoring the box score data into the +/- stat and then looking at it again later for me will lead to redundancy, and realistically I think it will happen for anyone else who doesn't have a really, really deep understanding of the hybrid stat.

This is not to say I'll refuse to use the hybrid model for some situations any more than I refuse to use PER or Win Shares, but "Which is better?" debates here cannot be judged so simply.

[E-Balla]

Parker's DRAPM seems to be the biggest bug here. He's better on defense than on offense, and more impactfull defender than Sanders, Asik or Deng?! No way.

Yeah, I can't square that.

Since entering the league he's played with Duncan a lot. He's always taken some of Duncan's credit according to RAPM. Not so much this year though.

[SideshowBob]

IPV's been updated.

http://talkingpracticeblog.com/

They've got offensive/defensive splits now.

[nikomCH]

IPV's been updated.

http://talkingpracticeblog.com/

They've got offensive/defensive splits now.

These rankings seem a bit more accurate to me. Kobe gets rated as a very pedestrian offensive player by RAPM and Parker doesn't look like a defensive god using this.

[CKRT]

Why is Harden ranked at 12, if he had a 4.8 IPV (good for 4th) ?

I'm on my phone so maybe something isn't loading fully.

[SideshowBob]

Why is Harden ranked at 12, if he had a 4.8 IPV (good for 4th) ?

I'm on my phone so maybe something isn't loading fully.

+4.8 is his offensive number. He's a -1.0 on defense so he comes out as +3.8.

[CKRT]

Ohh. That explains it. The defensive column wasn't loading. Thanks!

[SideshowBob]

For those that don't know, IPV is just RAPM that takes an SPM prior for offense and a defensive RAPM prior (from previous seasons).

[SideshowBob]

xRAPM was basically taking a similarly based stat, and blending it with the normal prior informed RAPM.

[Krodis]

Taking RAPM as a defensive prior seems to give more accurate results (to my eye, at least), than using any statistical priors for defense. It at least eliminates the problem with xRAPM where it was really skewed defensively in favors of big rebounders. (to the point where it had Al Jefferson as a positive defender iirc). Box score numbers for offense tend to be more useful, so I think SPM priors for offense can be fairly useful and it doesn't seem to give weird results (in fact, it seems to give more intuitive results than normal RAPM, which doesn't necessarily make it correct admittedly).

It looks good, I think. Still nice to look at raw RAPM too, to avoid box score noise altogether.

[SideshowBob]

Taking RAPM as a defensive prior seems to give more accurate results (to my eye, at least), than using any statistical priors for defense. It at least eliminates the problem with xRAPM where it was really skewed defensively in favors of big rebounders. (to the point where it had Al Jefferson as a positive defender iirc). Box score numbers for offense tend to be more useful, so I think SPM priors for offense can be fairly useful and it doesn't seem to give weird results (in fact, it seems to give more intuitive results than normal RAPM, which doesn't necessarily make it correct admittedly).

It looks good, I think. Still nice to look at raw RAPM too, to avoid box score noise altogether.

I don't quite understand what they mean by:

On defense, the prior is based primarily on defensive plus/minus in previous seasons (even advanced boxscore statistics are pretty useless on defense, though we now do use a small boxscore-based component to the defensive prior).

I understand the basic idea, but how would a box-score component be incorporated into a defensive/plus minus? Is it simply a weighted mean of defensive RAPM and defensive SPM?