APBRmetrics

[tomverve]

Minor terminological point here, but wouldn't a better name for "rebound rate" be "rebound percentage" or just "reb%" for short? "Rate" intuitively seems to imply something along the lines of per-minute or even per-possession rebound totals. Calling it reb% would emphasize that this is essentially a measure of the ratio of successes to opportunities. It would also fall more in line with conventional nomenclature such as FG% and FT%.

[Mike G]

Excellent point, if minor.

The '% of rebound opportunities', however, has been interpreted in a couple of ways. I tend to think rebounds collected by your teammates were never as 'available' as those gotten by the opponent. In other words, you Should be fighting your opponent for the board; and you Shouldn't be fighting your teammate for it.

Players tend to get more rebounds when they are with weaker-rebounding teams. Sharing rebounds with your teammates causes you to get fewer.

So, while it's technically true that a player's Reb% is just what it says it is -- % of total rebounds -- it isn't necessarily an adequate measure of relative ability to rebound. Suppose the Knicks outrebound the Suns, 60-40 : 100 Rebounds. Suppose Thomas (Phx) and Curry (NY) each have 10 Reb in 24 Min. Since they each played half the game, their Reb% is 10/50 = 20.0

But if we scale to opponent rebounds only, then Thomas got 10 of 60 (16.7), while Curry got 10 of 40 (25.0). And that is a huge difference.
_________________
`
36% of all statistics are wrong

[tomverve]

That's a good point-- I do think an even better measure of a player's rebounding ability would take into account the rebounding prowess of his teammates.

I don't know if your proposal would be the best way to do that though. Your example proposes reb% = player X rebounds/opp team rebounds, but that denominator counts rebounds that X could not have possibly grabbed (all those boards that were grabbed while X was on the bench). So that seems to lose the spirit of the "available" part of "% of available rebounds grabbed."

Another thing is that the boards a player grabs should also be counted in the denominator, if we're really talking about a successes/opportunities metric. Your metric seems to be not so much about successes/opportunities as it is a ratio of player totals to opposing team totals, which could be insightful and useful in its own right, but again seems to stray from the spirit of what should (IMO) be meant by reb%.

Also, while your metric avoids unfairly penalizing players for having good rebounding teammates, it ironically ends up penalizing players for having poor rebounding teammates (and padding the stats of players who have good rebounding teammates). While player rebounds will presumably vary somewhat depending on how good the player's teammates are at boarding, opponent rebounds will vary much more depending on that same variable. So if player X picks up 2 rpg after being traded to a very poor rebounding team, but his opponent team rpg jumps up by something like 10, X's rebounding numbers get unfairly penalized more than they get fairly compensated for. So while it's a good thought to try to reduce the 'skewiness' of the reb-rate numbers, ultimately I think your proposed solution might wind up skewing things even more.

Ultimately I think the best solution would just be to calculate reb-r (reb%) as usual, then come up with a more refined stat that can try to adjust reb-r in such a way as to take into account the reb-r of a player's teammates.

[Mike G]

Sorry, I just used a simplistic example. Of course, one has to factor the % of time the player is in the game. In my examples, he plays half the game and I halved the 'available' rebound total.

So rather than assume 'available' means (TmReb + OppReb), I said it could be = OppReb*2. This produces a sort of 'virtual availability', if you will.

And you're right, it penalizes too much the player on the poor-rebounding team. What I find works best is something in between. I'm scaling Reb/36 by the factor (TmRb/Tm+OppReb)^(2/3).
The exponent is 2/3. If it were zero, you'd have the old Reb%. If it's 1, you have the simplified example I gave.
_________________
`
36% of all statistics are wrong