APBRmetrics

[T.G. Randini]

Guy Wrote:

Quote:

But it simply isn't true. Garnett in his prime averaged +330 boards compared to average, but his team was just average (and Minn's rebounding has improved with his departure). Rodman in his prime was +540 reb/season, but his teams were only about +170 in those years. Berri constantly criticizes Curry for being a poor rebounder (among other things), but as a team the Knicks have been above-average on rebounds with Curry. These are just anecdotes, of course, but the pattern is quite clear: the benefit/cost to the team of a high/low individual reb total is only a small fraction of that player's total.

and...

Quote:

HRs are a teammate-independent stat: if I replace an average 3Bman with A-Rod, my team will hit more HRs. Rebounds are a different animal...

Excellent. I have made the similar observation more than once at WoW about the difference between continuous (basketball) and discrete (baseball) activities, but they don't get it over there. Guy, it's nice to see those individual vs. team rebounding stats. At WoW, they talk about the 'law of diminishing returns' but fail to realize it's giving a whacko rating that completely overvalues the guys grabbing the rebounds. And I had my final say yesterday there about rebounding. You can bring a horse to the water trough, but you...
_________________
T.G. Randini

[Guy]

I thought this post by Jason at WOW was interesting:

[Harold Almonte]

Guy. The only explanation for team (total rebounds) margins consistently smaller than its player's (grabbing margins) is that probably those rebounders that are +3 above average (grabbing rebounds), are also -3, or -2 under average (avoiding rebounds), remember, their opponents grab rebounds. If they are good defenders (boxing out) and are average or above (keeps his opponent at average +0), then his teammates are performing negative.

Once you set all positions average to 0 rating, you can forget about diminishing return and boxing out. That's not the problem, the problem is they say an average center is 14.4 rebs=14.4 rating points, while the true is 0 rating points (if his average opponent center also grabs 14.4 rebs.).

They can repair the metric and reset the rebounding scale making positions average=0, they can use the 0.7/0.3 shortage weighting (not my like), but I insist he had the best tool in the team defense adjust to include this allowance as he did with points allowed. They are very intelligent and the only reason I could imagine is that they just forgot it and lost precious time to repair it even with the book outside.

[Harold Almonte]

The mindset we need to change is that when opponents grabs a "defensive" rebound, the offensive players need to be punished (with the exception of the shooter probably). It's difficult to accept it, because we tend to look rebounding (from the offensive end) as a matter of gambling, but the statistical rules just obligue us.

[Guy]

[Harold Almonte]

First, he's not zeroing until rest the opponents rebounds.

[cherokee_ACB]

[Guy]

Thanks for the data.

[cherokee_ACB]

[Harold Almonte]

Guy. The problem with the Berri's average rebounding approach is he compares an average C=10.8 with the rebounding value of a different position, let's say PG=4.7. He's comparing absolutes, not relatives, wich is what a rating must do. He thinks a center produce more wins than the PG because he grabs 6.1 more rebounds, but the opponent average center grabs 6.1 more rebounds than the average opp. PG. The absolutes don't mean nothing because is a matter of floor advantage. Rebounding creation is not such an independent player skill than scoring creation (And I have my doubts with scoring too).

This thing has been explained from long time ago with all kind of points of view in this forum and in knickerblockers, but the most basic of all is that in a zero sum game, all averages must be set to zero.

And the rebounding absolutes can never produce more wins than FGMades absolutes, because FGMades has extra points-prize that rebounds don't have. That's why you see teams winning being negative in the rebounding margin.

[Mountain]

It is just one case but having looked at impact of Bynum in that new thread I wonder if offensive rebounding is largely independent while defensive rebounding is fairly dependent. Wonder if offensive rebounding changed for individual and team as much as defensive rebounding in the pairs cherokee listed. 82games player pair data in public does not reveal that info, though it could be complied fairly easily from basketballvalue.com at least for the team rebounding performances under the different pairs (assuming other things were near equal). The individual data would have to pulled from the play by play download I believe.


"Another simple exercise for Jason: take the starting C for every team, and calculate the overall Reb48 for the other 4 starters. Now run a correlation between C's Reb48 and Other 4's Reb48. If WOW is right, there should be only a small correlation. But it's not -- it's large, and negative. Cs who are "great rebounders" invariably have bad rebounders as teammates, and those who are "poor rebounders" are almost always blessed with talented rebounders as teammates. Isn't that an interesting coincidence? "

To confirm, you've done this? It "makes sense" to me but the empirical evidence is at issue so I want to understand where things sit right now. If you've done it, then Jason can do it too or we (and he if he wishes) can take your word for it. I'd take your word on it, in the discussion.

[Guy]

I suspect you're right on off/def Reb's. The real team benefit is probably something like .1 for each DReb and .5 for each OReb. You could probably ignore DRebs entirely in a valuation metric and not lose much.

[Mountain]

For Bynum is was near 1 for offensive rebs and near .5 for defensive rebs but of course a bigger sample would obviously be appropriate before making initial or settling on estimates.



Thanks for the clarification in last post.
Assuming the dominant rebounder / rest of team relationship is as posited, then your first test

["Just take all centers over the past couple of seasons whose Reb48 is, say, +2.0 compared to average (14.4, I think). Calculate the net rebounds they contributed above average: (Reb48-12.4)*MP/48. The average will probably be something like +3 Reb/G. Now look at their teams: are the teams close to +3 on average?" ... "What I think you'll find is that the team advantage is something like .20/.25 for each player rebound above average.* ']

would be influenced to a degree by the proportion of teams with strong rebounders with expected weak rest of casts vs those that buck that generalization? With cancelling out largely true but just a weak net effect on average?

And then we could look at the dominant rebounders who buck that general impact level and give team a bigger net rebounding bang and ask are they taller, quicker, stronger, move to ball better, read it better than average superior rebounder or what?

Or look at rest of cast and say that some of them were "more independent" / less deferential to the dominant rebounder than average?

Is it about players or is it coach / scheme at a very detailed level of design & instruction?

I am not planning on doing the ultimate rebounding study either but like you I willing to kick the can a little and say I'd like to see more if anyone has done more and wants to share it or wants to do it.

[Harold Almonte]

The best proof that defensive rebounding is not entirely a matter of skill is that def. rebounding leaders are not exactly the off. rebounders leaders. But we need to say in fairness that some rebounders are left to live in the paint by their coachs, while others are not benefitted by style of play of their teams (bigs in fast pace teams, bigs who can shoot from some distance)

[Mountain]

Harold is right that where you play matters on both sides of court. And where your teammates play matters too, especially the bigs. There was this study at 82 games
http://www.82games.com/rebounds.htm