Regularized Adjust Plus-Minus (RAPM) numbers for Rockets 2013-14 are out

This is absolute garbage

 

Please see my previous post. And I am a software engineer so I know exactly what black box means. That the underlying algorithms and mathematics have not been published. That in itself cannot be an indictment of the statistical and mathematical model.

 

Its not hidden or a black box (details of the methodology have been explained by its creator on a public forum), and the article you quoted isn't even by the guy who came up with xRAPM. Hoopdon is creating different stats based on xRAPM and calling it "adjusted xRAPM". Your criticism doesn't apply to xRAPM itself.

 

I found most stats(RAPM included) have trouble accounting for role/fit issue.

When Lin excelled in NY, he's the ONLY guard on their roster capable of running PNR based offense with his vision, attacking off screen and shooting off dribble.

Last season Asik is the ONLY defensive anchor we have to rebound and protect the rim.

Bev is the ONLY guard we have who can defend at high level and doesn't need ball in his hand on offense.

All three had very good RAPM numbers in the scenario's I mentioned. And we are seeing declining stats when Lin/Asik has to play with Harden/Howard together(Asik's RAPM haven't shown much drop yet, because I believe in limited sample size RAPM is blaming the twin tower failure on other starters with Asik's good defenseive metrics in the past).

There're numerous examples across the league like Kendall Marshall, Blatche, Beaseley, Gerald Green... where players excelled in the right system/role but struggled badly in other scenarios.

 

I don't think it makes sense to have stats that gauge a player's abilities outside any particular system. Its a fact that some players are "better" in certain roles than others. Its a mistake to try to separate out fit/role from the analysis, in my opinion.

 

one issue with xRAPM is that it factors in height on the defensive end.

 

Funny, I assume the NBA teams do the opposite. I suspect Morey utilizes stats a player does for one team/role and approximates the value they have for the Rockets system.

 

Sorry for my pic, here it 2012-2013 xRAPM numbers, seems to align a bit more with common thinking



-Dwight and Omer and the 2nd and 6th ranked defensive players
-James Harden is the 3rd ranked offensive player,
-Parsons is oddly poor in this metric
-Lin and Beverley seem to have inverse effectiveness offensively/defensively, doesn't make sense but these are just numbers

 

On Parsons, xRAPM takes in more than one year's data to be calculated. Parsons had about 1 and half of data that shows him to be a role player with about half a year has a decent to good starter offensively. Still, Parsons is probably a little overrated by the fans as he may look better than he is because his replacements really suck.

On Lin and Beverley, Beverley is the better defensive and offensive player, but there was less than a season of data on him casuing Lin to look better defensively. xRAPM will most likely show that Beverley is clearly the better offensive and defensive player after the season.

 

I doubt Morey pays nearly as much attention to noisy player ratings as fans do. And, I think it's really impossible to quantify what someone isn't doing for their team that you think they'll do for your team.

 

Certainly, but if he is only using stats that are independent of system/role, then he'll get a less accurate picture of how good that player will be on his particular team.

Its like relying on just overall FG% instead of shooting zone stats to project how likely a player is to hit a shot from a particular spot on the floor.

 

Fans have xRAPM. Morey has a staff of experts. I would imagine that those experts look at a player both independent of role and projected role/fit with the Rockets system and roster.

One would hope that with the Synergy and SportsVU data fans will eventually come up with a better statistical metric than xRAPM. It just has not happened yet.

 

The numbers are clearly.

 

There appears to be some confusion in this thread over what exactly xRAPM is and its relationship to RAPM. So here's my attempt to clarify this based on my understanding.

The framework for xRAPM and how it is computed based on RAPM is given in this thread:

http://www.apbr.org/metrics/viewtopic.php?f=2&t=8025&start=75

Here's the description of xRAPM's calculation, based on his explanations:

<pre>Final Season xRAPM = 0.65*RAPM + 0.35*BoxScoreRating</pre>
RAPM is regularized adjusted +/-, using the xRAPM for the player in the previous season as a prior. Some details on how RAPM is calculated were given in 2010 by Joe Sill in his MIT Sloan paper. As he says in his paper, the idea is to minimize the following function:



where p_i is the number of possessions in a game snippet, t_i is the difference in offensive efficiency between the home and road teams for the game snippet, x is a vector of 1s, -1s, and 0s where each position corresponds to a player who is either playing for the home team, playing for the road team, or is not on the floor, and w is the vector of estimated player ratings. The "regularization" essentially adds the second term in that formula. Sill's approach used a gaussian distribution (based on the lambda) centered at 0 as a prior -- the effect being to pull a player's rating towards 0, with the pull being "stronger" the further away from 0 the rating would otherwise be. Engelmann's RAPM uses a guassian centered at the computed xRAPM from the previous season as a prior instead.

BoxScoreRating is a "statistical +/-" formula that regresses various "box score" stats onto +/-. It is used to stabilize the final rating. Its just a linear weights formula, computed in the following way (in his words):

So, that's the gist of it.

 

Probably already mentioned, but Beverley only has roughly 1 season of data so his numbers should be taken with a larger grain of salt than more established players. His error of estimation support this.

Also, every season there are energy guys who lack stats but make the list. Generally lower MPG guys but very high intensity who work well with his teammates. Beverley is likely showing that he's working to become one of the better role players in the NBA. Which this team has definitely lacked since Parsons went from hustling 2nd round pick to sexy all around model.

 

Not better than Harden's and Howard's best-case estimate. But it does put him solidly at no worse than the team's 3rd most important player.

His raw On/Off differential this year is +9.6, and last year it was +7.8. That's nearly 2000 minutes of us playing really well with him on the floor on average. So its not a surprise that RAPM would estimate his rating very highly.

Maybe he's just a guy who's a very good fit on the team and who's energy/tenacity makes a big difference. RAPM doesn't care about the numbers he puts up; just how much better the team seems to play due to his presence.

 

With my current understanding that is a best guess, and was never tested on the historical data. But you do seem to have a better handle on the subject. But that does not matter. It seems to be the the best available metric, but one could make a valid argument for ASPM.

 

rapm stats are more about how well a player fits than how good he is, i know rapm isn't the best, but george hill inside indiana perked my interest.