Adding Counterpart Rebounds to ezPM

The current ezPM model estimates rebounds missed by a player by using the league average offensive and defensive rebounding rates at each position. This is an improvement over using box score stats alone, because the play-by-play (PBP) data allows you to know how many rebounds were actually missed while a player was on the floor. We can take the evaluation of rebounds one step further, though, by using counterpart data. Specifically, we can charge missed rebounds to a player when his position counterpart acquires a rebound (on offense or defense). Of course, this is still not “perfect”, because we are assuming that each time a counterpart gets a rebound, the player had a chance to get the same rebound. We also don’t know the rebounding “responsibilities” that a team has assigned to its players (but this would be true of other models, too, of course).

So, where to begin? Well, let’s first look at how different counterpart data are from the original estimates in the model (for players with above average number of possessions through 1/08/11):

Counterpart rebounds versus original ezPM model estimates.

First, I should say that ORR and DRR represent the offensive and defensive rebounding rates for a player. A 50% rate (on either side) means that a player acquired 50% of the rebounds between himself and his counterpart. What you can see is that, overall, there is a very high correlation between the two methods (nice sanity check), but there are certainly deviations between the two (and that’s why were doing this, after all).

Let’s look at the players whose REB100 (+/- attributed to rebounding per 100 possessions according to ezPM) increases the most (in other words, these players were underestimated by the original model):

Top 25 Underestimated Rebounders

(The last column represents the error by using estimates as opposed to counterpart data.)

RANK NAME TEAM ORR_CNPT DRR_CNPT REB100 ORR_EST DRR_EST REB100_EST REB_ERR
1 Wallace,  Gerald CHA 27.52% 81.87% 0.97 28.14% 57.11% -1.58 2.54
2 Wade,  Dwyane MIA 39.73% 89.30% 1.82 5.71% 82.27% -0.53 2.35
3 Varejao,  Anderson CLE 35.14% 68.27% 0.65 16.21% 67.50% -1.66 2.32
4 Williams,  Shelden DEN 43.88% 74.48% 2.01 23.25% 79.06% 0.14 1.88
5 Beasley,  Michael MIN 27.94% 74.35% 0.23 22.24% 62.59% -1.45 1.68
6 Mbah a Moute,  Luc MIL 32.96% 75.35% 1.02 29.64% 65.41% -0.40 1.42
7 Williams,  Mo CLE 20.90% 80.77% 0.28 18.74% 67.35% -1.07 1.35
8 Wright,  Dorell GSW 23.43% 77.55% 0.04 12.66% 69.92% -1.09 1.13
9 Telfair,  Sebastian MIN 15.91% 82.14% 0.07 8.27% 71.17% -0.93 0.99
10 Bonner,  Matt SAS 26.32% 65.49% -0.78 18.93% 63.13% -1.76 0.98
11 Westbrook,  Russell OKC 33.74% 82.73% 0.78 18.59% 80.59% -0.12 0.89
12 Smith,  Josh ATL 26.61% 77.81% 0.42 24.06% 73.40% -0.44 0.86
13 Dalembert,  Samuel SAC 47.13% 72.78% 2.56 42.69% 69.78% 1.73 0.82
14 Williams,  Deron UTA 22.32% 81.62% 0.15 15.51% 76.80% -0.64 0.79
15 Butler,  Rasual LAC 11.76% 81.43% -0.09 9.83% 74.35% -0.86 0.77
16 Bass,  Brandon ORL 38.46% 74.29% 1.36 35.85% 69.04% 0.61 0.75
17 Ilyasova,  Ersan MIL 36.71% 71.10% 1.16 30.20% 70.19% 0.48 0.68
18 Fernandez,  Rudy POR 17.14% 88.71% -0.02 15.34% 78.24% -0.66 0.64
19 Nowitzki,  Dirk DAL 13.48% 70.85% -1.28 10.86% 69.50% -1.88 0.60
20 Jackson,  Stephen CHA 19.20% 80.31% -0.03 16.33% 75.82% -0.60 0.56
21 Butler,  Caron DAL 18.49% 83.02% -0.01 17.57% 75.55% -0.57 0.56
22 Bell,  Raja UTA 17.05% 76.40% -0.55 13.74% 71.32% -1.11 0.56
23 Randolph,  Zach MEM 49.16% 77.00% 3.53 46.12% 74.52% 2.97 0.55
24 Love,  Kevin MIN 51.28% 80.18% 4.42 45.55% 79.38% 3.87 0.55
25 Howard,  Dwight ORL 46.98% 76.84% 3.01 38.18% 77.46% 2.46 0.55

Top 25 Overrated Rebounders

RANK NAME TEAM ORR_CNPT DRR_CNPT REB100 ORR_EST DRR_EST REB100_EST REB_ERR
215 Williams,  Reggie GSW 18.33% 62.28% -1.54 37.52% 73.23% 1.34 -2.89
214 Villanueva,  Charlie DET 17.57% 68.89% -1.30 34.96% 84.98% 1.35 -2.65
213 Turkoglu,  Hedo PHX 15.65% 61.95% -1.93 18.83% 86.49% 0.70 -2.62
212 Young,  Nick WAS 10.92% 79.17% -0.92 36.84% 78.72% 1.50 -2.42
211 Warrick,  Hakim PHX 28.17% 52.38% -2.26 21.70% 76.05% -0.24 -2.02
210 Udrih,  Beno SAC 17.89% 77.48% -0.32 35.09% 74.28% 1.57 -1.89
209 Wall,  John WAS 5.41% 80.43% -0.71 37.85% 67.69% 1.12 -1.83
208 Gallinari,  Danilo NYK 15.57% 56.57% -2.59 16.99% 73.45% -0.95 -1.64
207 Marion,  Shawn DAL 32.28% 67.76% -0.00 36.27% 78.85% 1.47 -1.47
206 Terry,  Jason DAL 6.90% 68% -1.28 28.48% 71.37% -0.04 -1.23
205 Wallace,  Ben DET 34.17% 64.56% 0.23 27.41% 84.46% 1.19 -0.97
204 O’Neal,  Shaquille BOS 32.41% 63.70% -0.66 32.11% 70.31% 0.24 -0.90
203 Smith,  J.R. DEN 22.12% 72.73% -0.32 24.01% 83.41% 0.55 -0.87
202 Jefferson,  Al UTA 27.25% 62.65% -1.42 28.35% 67.58% -0.56 -0.86
201 Bargnani,  Andrea TOR 20% 57.14% -2.32 18.62% 65.29% -1.48 -0.84
200 Hill,  Grant PHX 30.20% 65.56% -0.39 30.65% 77.10% 0.42 -0.81
199 Hayes,  Chuck HOU 38.60% 61.99% 0.23 38.32% 67.92% 1.02 -0.79
198 Horford,  Al ATL 29.50% 70.77% -0.05 31.57% 74.35% 0.69 -0.74
197 McGee,  JaVale WAS 43.42% 57.87% 0.28 42.98% 61.60% 1.01 -0.73
196 Brand,  Elton PHI 34.86% 64.98% 0.46 37.90% 68.15% 1.18 -0.72
195 Ford,  T.J. IND 12.90% 78.79% -0.67 17.17% 82.82% 0.03 -0.71
194 Lee,  David GSW 32.45% 61.25% -0.59 35.87% 63.56% 0.10 -0.69
193 Dunleavy,  Mike IND 14.39% 74.03% -0.85 13.99% 81.53% -0.21 -0.64
192 Davis,  Baron LAC 13.33% 72.88% -0.42 15.98% 81.21% 0.22 -0.64
191 Dudley,  Jared PHX 26.47% 65.77% -0.63 28.53% 71.93% -0.00 -0.63

Both lists are interesting. One should not (necessarily) interpret being on either of the lists as being good or bad, but it’s certainly worth thinking about what the data mean. I need to do that some more. To be continued…

16 thoughts on “Adding Counterpart Rebounds to ezPM”

  1. The ORR chart above looks strange to me, in that the two compared rates seem to either match up very well or have a huge error, with not much in between. Am I imagining things? And if not, do you have any guess as to why that quirk is showing up?

    1. First, I should say that the data represents 216 players, so I’m not surprised a handful or so are way off. Dwyane Wade appears to be one of those. His estimated ORR was only 5%, but the counterpart data says his ORR is closer to 40%. Off the top of my head, though, I can’t think of why the estimate for him would be so low to start with. He’s listed as a “2” by Aaron’s matchup file (which is what I’m using). That seems right, doesn’t it?

    2. A lot of the major errors are probably caused my the matchup file listing an incorrect position for a player. For example, Reggie Williams mostly plays 2, but the matchup lists him as a 3. Dorell Wright (same team) plays 3, but is listed as a 2. In this case, I “corrected” those players, yet they still show up on these lists. So, although Aaron does a great job coming up with those matchups, it still represents a limiting factor, in terms of accuracy.

  2. Nice work and post.

    A few comments, for what they may be worth:

    Looks like all but a very few players are above 60% defensive rebounding vs counterpart. Probably 2/3rds to 3/4ths over 70%? Not sure if there is anything to be done with that knowledge but there might be.

    If you wanted another idea related to rebounding to look at in the play by play I’ll throw the possibility of tracking the impact of 2+ inch height differences to see what the data shows in general and by position and by player.

  3. I am not sure what the ratio of guys in the middle between 2 positions is in the current file but in my original classification it was about 30% of players. Averaging the position expectations for those guys would seem like a worthwhile enhancement.

  4. Comparing the top 50 and bottom 50 rebounder lists I think 25 teams have at least one on both lists. That might suggest variations in defensive rebounding roles from the norm on at least some teams as well as skill variation. A few teammates probably take some rebounds from other teammates. It might be an argument for some regression of credit towards the mean for players at both extreme tails of defensive rebounding. Not a lot, but maybe a little. Say 10-20% of the marginal rebounding value above or below position average? Just an idea offered for possible consideration.

  5. It looks like at team level only about 20% of teams would be more than 1 defensive rebound away (up or down) from league average if they all faced the same level of opportunities so it seems unlikely that the top guys on this are adding as much as the raw data suggests.

    I’d probably leave the offensive rebounding data as is. Any adjustment would be several times smaller than for defensive rebounding and thus probably not worth it.

  6. Thanks.

    Among the other columns I see opponent 3 point shot points. I don’t think I’ve seen that anywhere else by player. Useful to see.

    1. Those shot totals are not counterpart, btw. They are opponent totals while the player was on the floor. Counterpart defense is next on my list of major todo things.

  7. It is dated a bit and not complete for today but “if” you wanted my older position designation file that assigned players a position number in .2 increments I’d give it to you (via e-mail, if you indicated where to send by private message at APBRmetrics).
    If not of interest, that’s ok. I won’t mention it again.

    1. Thanks, Crow, but it probably won’t be too much help. The issue is that my code is built around parsing the matchup file from bball-value. At some point, I should probably implement my own matchup generation code, so I can input correct positions. But I think Aaron has it mostly right.

Leave a Reply