back2newbelf
PostPosted: Thu Feb 03, 2011 5:11 pm Post subject: Adjusted rebounding Reply with quote
Using the same technique as in http://sonicscentral.com/apbrmetrics/vi ... php?t=2669 I computed adjusted rebounding for the last 1 1/2 years.
Home teams grab more available offensive rebounds (30.4% vs 29.7%).
Lambdas were 1500 for offensive rebounding and 3000 for defensive rebounding.
Offensive and Defensive numbers are per 100 "rebound opportunities"
If someone wants the raw data in bbv matchup format I can upload the file.
http://www.stats-for-the-nba.appspot.com/rebounding
_________________
http://stats-for-the-nba.appspot.com/
Crow
Thanks.
10 guys estimated at +2 better on total rebounds impact per 100 possessions at each end of the court. Less than 2%.
Only about 15% of the 538 players listed are estimated at +1 rebound or better on this and about 15% are -1 or worse. So it is estimated, by this regularized model, that about 70% of players are between +1 and -1. FWIW, it appears there were even fewer values beyond equivalent thresholds in Joe Sill's earlier estimates.
Food for thought about the relative importance of individual rebounding and maybe the degree of regularization.
Statman
10 guys estimated at +2 better on total rebounds impact per 100 possessions at each end of the court. Less than 2%.
Only about 15% of the 538 players listed are estimated at +1 rebound or better on this and about 15% are -1 or worse. So it is estimated, by this regularized model, that about 70% of players are between +1 and -1. FWIW, it appears there were even fewer values beyond equivalent thresholds in Joe Sill's earlier estimates.
Food for thought about the relative importance of individual rebounding and maybe the degree of regularization.
Which is why Berri's stuff often looks so absurd to us (well, at least me) - valuing player rebounds at such a high rate.
_________________
Dan
My current national college player rankings (and other stuff):
http://www.pointguardu.com/f136/statman ... post355594
back2newbelf
PostPosted: Fri Feb 04, 2011 7:45 am Post subject: Reply with quote
Crow wrote:
10 guys estimated at +2 better on total rebounds impact per 100 possessions at each end of the court.
I should have been clearer. It's not per possession. It's per "rebound opportunity". A "rebound opportunity" is every time a player of your team misses a shot that can be rebounded by your team (basically every shot except technical- and "first of two"-freethrows).
Compared to APM with points a missed shot (that can be rebounded) by Team A is equivalent to a possession by Team A. Team A getting the subsequent offensive rebound would be equivalent to Team A scoring one point.
I've also thought about doing this without all the available rebounds from free throws(when the last free throw was missed). I would imagine those kind of rebounds get very rarely rebounded by the offensive team and the guards of the offensive team have almost no influence on those. Thus, guards from teams that shoot (and miss) a lot of free throws get unfairly punished
_________________
http://stats-for-the-nba.appspot.com/
EvanZ
PostPosted: Fri Feb 04, 2011 8:29 am Post subject: Reply with quote
back2newbelf wrote:
I've also thought about doing this without all the available rebounds from free throws(when the last free throw was missed). I would imagine those kind of rebounds get very rarely rebounded by the offensive team and the guards of the offensive team have almost no influence on those. Thus, guards from teams that shoot (and miss) a lot of free throws get unfairly punished
FWIW, I don't count give credit for dreb off free throws for similar reasons. I do give credit for oreb, though.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Mike G
PostPosted: Fri Feb 04, 2011 10:24 am Post subject: Re: Adjusted rebounding Reply with quote
back2newbelf wrote:
...
Home teams grab more available offensive rebounds (30.4% vs 29.7%).
Hmm, how is it that league avg ORb% is .263 ?
Quote:
I've also thought about doing this without all the available rebounds from free throws(when the last free throw was missed). I would imagine those kind of rebounds get very rarely rebounded by the offensive team
Even then, off missed FG the league ORb% is around .280.
What is it I'm missing?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Ilardi
PostPosted: Fri Feb 04, 2011 10:38 am Post subject: Reply with quote
back2newbelf -
Nice work (as usual). Have you taken a look to see how your regularized adjusted rebound rates compare with simple ORR% and DRR%? I would guess the correlations must be pretty high, but would love to see the actual r's if you or anyone else have time to check.
back2newbelf
PostPosted: Fri Feb 04, 2011 10:50 am Post subject: Re: Adjusted rebounding Reply with quote
Mike G wrote:
Quote:
I've also thought about doing this without all the available rebounds from free throws(when the last free throw was missed). I would imagine those kind of rebounds get very rarely rebounded by the offensive team
Even then, off missed FG the league ORb% is around .280.
What is it I'm missing?
I didn't do this yet
Mike G wrote:
back2newbelf wrote:
...
Home teams grab more available offensive rebounds (30.4% vs 29.7%).
Hmm, how is it that league avg ORb% is .263 ?
Is it .263 of *all rebounds* or just of those that can actually be rebounded by the attacking team? I don't count it as rebounding opportunity if the first of two free throws is missed, but it's always listed as a team rebound for the defending team
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
mystic
PostPosted: Fri Feb 04, 2011 11:05 am Post subject: Reply with quote
Ilardi wrote:
back2newbelf -
Nice work (as usual). Have you taken a look to see how your regularized adjusted rebound rates compare with simple ORR% and DRR%? I would guess the correlations must be pretty high, but would love to see the actual r's if you or anyone else have time to check.
I just ran a linear regression with ORB% obtained by boxscores and those values for offensive rebounds:
Code:
Model Summary
Model R R Square Adjusted R Square Std. Error of the Estimate
1 ,363a ,132 ,130 ,7704233
a. Predictors: (Constant), ORB%
Same for DRB%
Code:
Model Summary
Model R R Square Adjusted R Square Std. Error of the Estimate
1 ,318a ,101 ,100 ,4245693
a. Predictors: (Constant), DRB%
And for the combination:
Code:
Model Summary
Model R R Square Adjusted R Square Std. Error of the Estimate
1 ,366a ,134 ,132 ,8530314
a. Predictors: (Constant), TRB%
Doesn't look like a high correlation to me. Which even gives more fuel to the "why Berri overrates rebounds" conversation.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
Ilardi
PostPosted: Fri Feb 04, 2011 11:42 am Post subject: Reply with quote
Thanks, mystic. Wow, r's down in the .32-.36 range. I would have guessed much higher.
So, if we were to fully trust these adjusted rebounding numbers, they would imply that simple ORR% and DRR% figures are misleading as often than not, i.e., that a player's rebounding performance is so heavily contextual - dependent on teammates and opponents - that simple unadjusted rebound rate is nearly uninterpretable on its own.
The implications are not trivial . . .
Back to top
View user's profile Send private message
EvanZ
PostPosted: Fri Feb 04, 2011 11:53 am Post subject: Reply with quote
I agree!
(Mystic, nice to see you here.)
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Ilardi
PostPosted: Fri Feb 04, 2011 11:55 am Post subject: Reply with quote
A few of the adjusted rebounding values are so surprising that I'm having trouble understanding how they could be valid. For example, Dwight Howard doesn't even place among the top 50, while he ranks #3 among NBA starters in overall rebounding percentage. Any thoughts?
Back to top
View user's profile Send private message
DSMok1
Location: Where the wind comes sweeping down the plains
PostPosted: Fri Feb 04, 2011 11:57 am Post subject: Reply with quote
Wow, those are low r-values!
I wonder if that's why, when I ran my ASPM regressions, rebounding was weighted so lightly?
_________________
GodismyJudgeOK.com/DStats
Twitter.com/DSMok1
Back to top
View user's profile Send private message Send e-mail Visit poster's website
EvanZ
PostPosted: Fri Feb 04, 2011 12:04 pm Post subject: Reply with quote
This is probably germaine.
When I take the REB100 from ezPM and correlate it with ORR (which is the offensive rebounding rate for a player relative to his counterpart), the R^2 is 0.43. When I do the same thing for DRR, the R^2 is 0.0001.
It's sort of blowing my mind. Either I'm giving not enough credit for defensive rebounds, or this has something to do with diminishing returns. I guess. Question
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
back2newbelf
Joined: 21 Jun 2005
Posts: 271
PostPosted: Fri Feb 04, 2011 12:06 pm Post subject: Reply with quote
Everyone, please don't jump to conclusions. There might be an error in my code as there seem to be discrepancies with this and average OReb%, it might still be a not-so-good approximation or maybe the methodology isn't perfect for that kind of problem.
All things considered, we can only know if this is useful data if we compare it to standard OReb/DReb% (averaged over all on-court-players) as an predictor for out of sample data, which isn't hard to do but I don't have much time
_________________
http://stats-for-the-nba.appspot.com/
recovered page 2
Author Message
Mike G
Joined: 14 Jan 2005
Posts: 3450
Location: Delphi, Indiana
PostPosted: Fri Feb 04, 2011 12:10 pm Post subject: Re: Adjusted rebounding Reply with quote
back2newbelf wrote:
Mike G wrote:
... league avg ORb% is .263 ?
Is it .263 of *all rebounds* or just of those that can actually be rebounded by the attacking team? I don't count it as rebounding opportunity if the first of two free throws is missed, but it's always listed as a team rebound for the defending team
See here:
http://www.basketball-reference.com/lea ... _2011.html
ORb% = ORb/(ORb + OppDRb)
It's the ratio of your team's OReb to all rebounds off missed shots by your team.
(Excluding missed FT, it's about .280 .)
So I don't see how it can be .304 at home and .297 on the road, to add up to .263 .
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
mystic
Joined: 18 Dec 2006
Posts: 5
PostPosted: Fri Feb 04, 2011 12:12 pm Post subject: Reply with quote
DSMok1 wrote:
Wow, those are low r-values!
I wonder if that's why, when I ran my ASPM regressions, rebounding was weighted so lightly?
I think that is the reason. Did you ever tested different datasets? Like from different seasons?
I got different values for ORB% ranging from having a clear negative impact to having a small positive impact.
And yes, it isn't trivial, but that is in agreement with examples of Marcus Camby or David Lee, where the team done worse in the rebounding department than average with those two on the court while being better without them. It looks like the impact of a rebounder isn't that huge overall. Probably he just takes a lot of rebounds away from his teammates. The offensive rebounds are also not that valuable in the end, especially when they are coming from own misses while missing the next layup again.
Looking at the current Celtics, we can see that they aren't really going for the offensive rebound, but rather going back on defense. Seems like a good trade off to me with the right team. Their scoring margin is the 2nd highest in the league. Maybe doing the right things on defense is more important to the team success than just trying to get every possible offensive rebound while in the end in most cases the defending team gets that board anyway.
Evan, I'm reading here for about 6 years now, just recently I got a new password to use that account, but I never really had the time to do something valuable for this community, most times I just played around with my own models and wrote some articles for a german basketball website.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
EvanZ
Joined: 22 Nov 2010
Posts: 173
PostPosted: Fri Feb 04, 2011 12:13 pm Post subject: Reply with quote
EvanZ wrote:
This is probably germaine.
When I take the REB100 from ezPM and correlate it with ORR (which is the offensive rebounding rate for a player relative to his counterpart), the R^2 is 0.43. When I do the same thing for DRR, the R^2 is 0.0001.
It's sort of blowing my mind. Either I'm giving not enough credit for defensive rebounds, or this has something to do with diminishing returns. I guess. Question
Just wanted to make sure people saw my comment on a new page.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
mystic
Joined: 18 Dec 2006
Posts: 5
PostPosted: Fri Feb 04, 2011 12:18 pm Post subject: Reply with quote
Evan, I saw your comment, the defensive part is really interesting. Now, as back2newbelf said, the code can be just wrong and we are comparing here wrong data anyway. Which means our tries to interpret that data is a bit premature.
Would be nice to have the data splitted up for teams the players were on and the time they were on the court to actual check whether the data can explain the overall team reb%, because in the end that should be the case.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
DSMok1
Joined: 05 Aug 2009
Posts: 535
Location: Where the wind comes sweeping down the plains
PostPosted: Fri Feb 04, 2011 12:57 pm Post subject: Reply with quote
mystic wrote:
Evan, I saw your comment, the defensive part is really interesting. Now, as back2newbelf said, the code can be just wrong and we are comparing here wrong data anyway. Which means our tries to interpret that data is a bit premature.
Would be nice to have the data splitted up for teams the players were on and the time they were on the court to actual check whether the data can explain the overall team reb%, because in the end that should be the case.
You mean like BasketballValue does on the expanded team pages?
_________________
GodismyJudgeOK.com/DStats
Back to top
View user's profile Send private message Visit poster's website
mystic
Joined: 18 Dec 2006
Posts: 5
PostPosted: Fri Feb 04, 2011 1:08 pm Post subject: Reply with quote
I know where I can find those kind of data for unadjusted rebounding values, but I would like to have that for the adjusted rebounding values back2newbelf presented here as well.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
Crow
Joined: 20 Jan 2009
Posts: 737
PostPosted: Fri Feb 04, 2011 1:21 pm Post subject: Reply with quote
Sorry for my improper assumption that the metric was based on possessions rather than rebounding opportunities. Per 100 own possessions / 200 total was the standard Joe Sill used and almost all other Adjusted +/- reports have used.
Given that the number are per 100 "rebound opportunities", then, if I am understanding it correctly, the players with the extreme impacts are having even less positive and negative impact on rebounding per game in this metric than I previously said / thought and these results are more inline with what Joe Sill reported.
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 173
PostPosted: Fri Feb 04, 2011 1:38 pm Post subject: Reply with quote
mystic wrote:
Evan, I saw your comment, the defensive part is really interesting. Now, as back2newbelf said, the code can be just wrong and we are comparing here wrong data anyway. Which means our tries to interpret that data is a bit premature.
Would be nice to have the data splitted up for teams the players were on and the time they were on the court to actual check whether the data can explain the overall team reb%, because in the end that should be the case.
Just wanted to be clear, that my results had nothing to do with b2b's data set. The correlations I'm referring to are performed on rebounding data from my PBP code.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Guy
Joined: 02 May 2007
Posts: 100
PostPosted: Fri Feb 04, 2011 2:02 pm Post subject: Reply with quote
EvanZ: I'm not quite following what correlation you are reporting. Is it between players and team, or between two different metrics for the same player?
Back to top
View user's profile Send private message
bbstats
Joined: 25 Apr 2010
Posts: 31
PostPosted: Fri Feb 04, 2011 2:08 pm Post subject: Reply with quote
Just so we're all on the same page, there are well-documented diminishing returns of rebounding: http://www.countthebasket.com/blog/2008 ... g-returns/
(at least on the single-team-level...not sure if Eli had tried measuring this accounting for both teams on the court)[/url]
_________________
http://thebasketballdistribution.blogspot.com
http://twitter.com/bbstats
Back to top
View user's profile Send private message Visit poster's website
back2newbelf
Joined: 21 Jun 2005
Posts: 229
PostPosted: Fri Feb 04, 2011 2:15 pm Post subject: Re: Adjusted rebounding Reply with quote
Mike G wrote:
So I don't see how it can be .304 at home and .297 on the road, to add up to .263 .
I tried to find where the difference comes from but right now I'm just confused. 82games has sites like this http://www.82games.com/1011/10ORL18.HTM with Howards' On/Off OReb% at 28.4% and 27.4%.
Then there's http://basketballvalue.com/teamplayers. ... l&team=ORL with Howards' On/Off OReb% at 35.4% and 30.1% and pretty much everybody in the league with 29% or more
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 737
PostPosted: Fri Feb 04, 2011 2:28 pm Post subject: Reply with quote
It might help to see a run of "Adjusted rebounding" without being regularized to see how much that affects the regularized results.
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 173
PostPosted: Fri Feb 04, 2011 2:30 pm Post subject: Reply with quote
Guy wrote:
EvanZ: I'm not quite following what correlation you are reporting. Is it between players and team, or between two different metrics for the same player?
The latter. Let me explain in case others are also wondering.
I give credit for defensive rebound (0.26) and offensive rebound (0.74). I debit for missed defensive rebound (-0.74) and missed offensive rebound (-0.26). A "missed rebound" is defined as one that your counterpart acquires. It is not a missed rebound if your teammate acquires it. Team rebounds are split amongst teammates.
REB100 would be the above sum for each player normalized to 100 possessions.
I keep track of an offensive rebounding rate (ORR), which is defined as follows:
ORR=OREB/(OREB+OREBCP)
Where OREBCP is the number of rebounds your counterpart gets. DRR is defined similarly. The underlying assumption is that a player's "true" rebounding opportunities are the ones he actually gets plus the ones his counterpart gets.
The correlation I referred to earlier was REB100 vs. ORR and REB100 vs. DRR.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 173
PostPosted: Fri Feb 04, 2011 2:35 pm Post subject: Reply with quote
bbstats wrote:
Just so we're all on the same page, there are well-documented diminishing returns of rebounding: http://www.countthebasket.com/blog/2008 ... g-returns/
(at least on the single-team-level...not sure if Eli had tried measuring this accounting for both teams on the court)[/url]
Jon Nichols also did a study, although strangely, the link appears to be dead now.
Cherokee_ACB did it here a few years ago.
So did Phil B. recently.
So did I a little while back.
I think pretty much everyone here is in agreement that there are diminishing returns, and probably substantial. In my mind, the only question is how best to account for it.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
mystic
Joined: 18 Dec 2006
Posts: 5
PostPosted: Fri Feb 04, 2011 6:00 pm Post subject: Reply with quote
EvanZ wrote:
Just wanted to be clear, that my results had nothing to do with b2b's data set. The correlations I'm referring to are performed on rebounding data from my PBP code.
Ok, thanks for clarifying this, because I thought you compared your rebounding rating to the adjusted rebounding values.
_________________
http://bbmetrics.wordpress.com
page 3
Author Message
back2newbelf
Joined: 21 Jun 2005
Posts: 276
PostPosted: Fri Feb 04, 2011 6:24 pm Post subject: Re: Adjusted rebounding Reply with quote
Mike G wrote:
So I don't see how it can be .304 at home and .297 on the road, to add up to .263 .
If you take this site:
http://www.basketball-reference.com/box ... 60LAL.html
there's an OReb% of 34.8% listed for Houston, 27.5% for LA.
If you add up the Rockets' individual ORebs, add them to LAs' individual DRebs you get 46. Divide Houstons' individual ORebs by that and you get 16/46 = 34.8%. Same with the other team
If you look at the playbyplay though, both teams did get a whole lot of team rebounds (mostly through loose ball fouls)
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 307
PostPosted: Fri Feb 04, 2011 7:17 pm Post subject: Reply with quote
mystic wrote:
EvanZ wrote:
Just wanted to be clear, that my results had nothing to do with b2b's data set. The correlations I'm referring to are performed on rebounding data from my PBP code.
Ok, thanks for clarifying this, because I thought you compared your rebounding rating to the adjusted rebounding values.
I'm not sure what I did before, but now I'm not getting the same result. I need to work on this some more.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 825
PostPosted: Fri Feb 04, 2011 9:12 pm Post subject: Reply with quote
Ilardi wrote:
A few of the adjusted rebounding values are so surprising that I'm having trouble understanding how they could be valid. For example, Dwight Howard doesn't even place among the top 50, while he ranks #3 among NBA starters in overall rebounding percentage. Any thoughts?
Howard is tied for 15th best on this adjusted defensive rebounding impact but barely below neutral on offensive rebounding impact.
On Joe Sill's report for rebounding factors thru 2008-9 Howard was a strong 88th percentile on DR impact but also 91th percentile on offensive rebounding.
But in Joe's database there was no listing for Gortat and even though Gortat has moved he is in this database and has OR and DR impact ratings very close to Howard's. Between them they have 90% of the time at center for the season at 82games. Are they somehow canceling each other almost all the way out? Bass and Lewis at PF, are they affecting the measurements of the centers despite the attempt to isolate impacts?
This could be an example of a situation where player pair Adjusted factors might have additional value to try to help sort out the story. Actually you'd probably want to look at all the various individual season level rebounding ratings including the Adjusted factors, the Adjusted factor pair ratings, raw and maybe Adjusted lineup rebound ratings and raw individual data in lineup ratings and pairs in lineup ratings. If that and observation does not help solve it, then either try again, perhaps differrently somehow, or work on something you can solve and get benefit from.
Back to top
View user's profile Send private message
back2newbelf
Joined: 21 Jun 2005
Posts: 276
PostPosted: Sat Feb 05, 2011 9:20 am Post subject: Reply with quote
Ilardi wrote:
A few of the adjusted rebounding values are so surprising that I'm having trouble understanding how they could be valid. For example, Dwight Howard doesn't even place among the top 50, while he ranks #3 among NBA starters in overall rebounding percentage. Any thoughts?
I think one possible explanation would be that Orlando plays a bit of a different style when Howard is offcourt. Looking at the 5-man-units at basketballvalue there is just one Orlando unit in the top 40 (or so, by minutes) that doesn't involve Howard at Center and involves either Lewis or Turkoglu at PF. All other non-Howard units either have Bass/Anderson or Gortat/Bass at C/PF, with Lewis or Turkoglu at SF. The with-Howard units contain lots of Lewis/Turkoglu at PF.
Lewis and Turkoglu are not rated horrible enough (which they very well might be, at least at PF) by the algorithm because the rebounding is somewhat OK when they're oncourt, the fact that they switch positions from lineup to lineup gets lost.
Next up: different lambda values for different player positions
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
back2newbelf
Joined: 21 Jun 2005
Posts: 276
PostPosted: Sat Feb 05, 2011 1:04 pm Post subject: Reply with quote
Our cat just died.
Rest in peace.
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 825
PostPosted: Sat Feb 05, 2011 2:45 pm Post subject: Reply with quote
Condolences.
A short explanation of why different lambda values for different player positions might help and what it actually does would be appreciated, when you get to it.
Back to top
View user's profile Send private message
back2newbelf
Joined: 21 Jun 2005
Posts: 276
PostPosted: Wed Feb 09, 2011 5:45 am Post subject: Reply with quote
Crow wrote:
Condolences.
A short explanation of why different lambda values for different player positions might help and what it actually does would be appreciated, when you get to it.
Thank you.
I think it's plausible that low post players have bigger influence on rebounding than wing players/guards. If so, higher lambdas for smaller players and lower lambdas for bigger players should lead to better out of sample prediction.
BTW, OReb% for missed free throws is 12.4% for this year and 31.3% for non-free-throw-misses.
Test on normal approximated RAPM lead me to believe that 3 or 4 year analysis is optimal for predicting future events, thus I'll probably soon take down the 2 year analysis and replace it with a 4 year analysis,
one with Free Throw misses and one without.
Stay tuned
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
back2newbelf
Joined: 21 Jun 2005
Posts: 276
PostPosted: Thu Feb 10, 2011 5:15 am Post subject: Reply with quote
4 year analysis is up with Free Throw misses
http://stats-for-the-nba.appspot.com/rebounding
and without
http://stats-for-the-nba.appspot.com/rebounding_no_fts
Dwight Howard improves quite a bit by not including FTs, probably because he himself is missing a lot of FTs while not being in an optimal position to grab the rebound (standing at the FT line)
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 307
PostPosted: Thu Feb 10, 2011 9:40 am Post subject: Reply with quote
I would bet (someone else's money) that Love leads the league in the category of "tip-ins + 3-pointers". Smile
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
mystic
Joined: 18 Dec 2006
Posts: 21
PostPosted: Thu Feb 17, 2011 10:00 am Post subject: Reply with quote
A correlation analysis of the data for rebounding and overall APM numbers from your site:
0.03 correlation coefficient for offensive rebounding and offensive APM
0.42 correlation coefficient for defensive rebounding and defensive APM
0.30 correlation coefficient for overall rebounding impact and overall APM
I calculated the overall rebounding impact by taking 0.3 * offensive rebounding + 0.7 * defensive rebounding. League average ORB% is 30%, if I got your informations right. I also adjusted it to league average rebounds per 48 minutes. Overall Kevin Love' impact ends up with 0.84 rebounds more per 48 minutes than an average replacement. Offensive rebounding seems like not having a huge impact on the overall offensive efficiency at least for the recent years. We get a 0.17 correlation coefficient, if we compare ORB% and ORtg for teams from 1979/80 to 2009/10. Taking the data from b-r.com for this season, we get a -0.12 correlation coefficient for that. If we compare the ORB% of the teams in this season with the margin of victory, we are getting a -0.21 correlation coefficient.
Going for the offensive rebound doesn't seem to really help a team in terms of improving their overall chance to win games.
The comparison between scoring margin and DRB% gives us a 0.36 correlation coefficient. Defensive rebounding seems way more important than offensive rebounding in terms of winning a game. The numbers from those correlation analyses are supporting that conclusion.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
back2newbelf
Joined: 21 Jun 2005
Posts: 276
PostPosted: Thu Feb 17, 2011 11:08 am Post subject: Reply with quote
mystic wrote:
I calculated the overall rebounding impact by taking 0.3 * offensive rebounding + 0.7 * defensive rebounding.
That's not correct. Overall rebounding impact is already listed as "per 200"
Quote:
I also adjusted it to league average rebounds per 48 minutes. Overall Kevin Love' impact ends up with 0.84 rebounds more per 48 minutes than an average replacement.
That's probably not correct either. Loves' impact is 5.3 more rebounds per 200 rebounding opportunities. There are, I'd say, 90(?) rebounding opportunities per 48 minutes. That would lead to 2.4 more rebounds per 48 minutes
_________________
http://stats-for-the-nba.appspot.com/
Last edited by back2newbelf on Thu Feb 17, 2011 1:43 pm; edited 1 time in total
Back to top
View user's profile Send private message
mystic
Joined: 18 Dec 2006
Posts: 21
PostPosted: Thu Feb 17, 2011 11:30 am Post subject: Reply with quote
A player adds a specific number of rebounds to the 30% league average on offensive rebounds, if I interpret your numbers correctly. Thus I need to adjust for that. That basically means, I adjust for the "missed rebounds" here.
Edit: Forget about it, you are right. I thought way too complicated about it. As you pointed out, you already accounted for it. Thus my numbers are wrong and I can use your overall numbers for the comparison. That reduces the correlation coefficient even further to around 0.2 for the overall rebounding impact and overall APM.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
Ilardi
Joined: 15 May 2008
Posts: 265
Location: Lawrence, KS
PostPosted: Thu Feb 17, 2011 3:07 pm Post subject: Reply with quote
B2NB,
So, presumably, we could now translate Love's adjusted rebounding impact into "expected points added per game".
I know the expected point value of a possession is close to 1.0, but I'd be willing to bet that it would be a bit greater for the possession immediately following an offensive rebound than one following a defensive rebound. Surely someone must have looked at this carefully and published the results.
Does anyone know the exact expected value (in points) for possessions immediately following offensive and defensive rebounds, respectively?
Back to top
View user's profile Send private message
DSMok1
Joined: 05 Aug 2009
Posts: 611
Location: Where the wind comes sweeping down the plains
PostPosted: Thu Feb 17, 2011 4:05 pm Post subject: Reply with quote
Ilardi wrote:
B2NB,
So, presumably, we could now translate Love's adjusted rebounding impact into "expected points added per game".
I know the expected point value of a possession is close to 1.0, but I'd be willing to bet that it would be a bit greater for the possession immediately following an offensive rebound than one following a defensive rebound. Surely someone must have looked at this carefully and published the results.
Does anyone know the exact expected value (in points) for possessions immediately following offensive and defensive rebounds, respectively?
This thread has some info on expected points by possession start. Kevin Pelton also linked to another thread of that sort.
http://sonicscentral.com/apbrmetrics/vi ... ut&start=0
I'd love to see that research updated--someday I'd like to create a "state-based" plus/minus system.
EDIT: after an offensive rebound, it's not a new "possession". It's a new "play". Don't want to mix up the terminologies, here!
_________________
GodismyJudgeOK.com/DStats
Twitter.com/DSMok1
Back to top
View user's profile Send private message Send e-mail Visit poster's website
mystic
Joined: 18 Dec 2006
Posts: 21
PostPosted: Thu Feb 17, 2011 5:13 pm Post subject: Reply with quote
Don't know how much stock we can put into the data by Synergy Sports, but they have the Minnesota Timberwolves with 1.03 points per possession in their category "offensive rebound". Not quite sure what that exactly means, but I think they mean in possessions with an offensive rebound. Or they just count possessions differently. Confused
Going by the numbers in the link DSMOK1 posted, we can expect 1.12 times higher efficiency than average after an offensive rebound off a jumper, and 1.17 of a non-jumper. I usually used 1.2 times for offensive rebounds (and also for steals).
cherokee_ACB wrote:
I've run my scripts on last season NBA play-by-plays. The results:
Code:
Possesion start Efficiency
Start of quarter 96.0
2p FG made 103.8
3p FG made 104.1
FT made 104.2
Non-jumper missed, DR 108.3
Jumper missed, DR 104.1
FT missed, DR 101.9
Non-jumper missed, OR 123.3
Jumper missed, OR 118.5
Steal 124.0
Non-steal turnover 101.7
Average 105.5
If we take that and the average amount of rebounds per 48 minutes, we get between 1.9 and 2.0 points added per 48 minutes due to Kevin Love's offensive rebounding impact.
_________________
http://bbmetrics.wordpress.com
page 4
Author Message
Mike G
Joined: 14 Jan 2005
Posts: 3583
Location: Hendersonville, NC
PostPosted: Thu Feb 17, 2011 6:01 pm Post subject: Reply with quote
Quote:
If we take that and the average amount of rebounds per 48 minutes, we get between 1.9 and 2.0 points added per 48 minutes due to Kevin Love's offensive rebounding impact.
Kevin Love averages 6.13 OReb per 48 min.
If an OReb creates 1.14 times as many points as an average possession, that's 1.14 x 1.049 = 1.2 pts (if Wolves are typical in this regard).
If the alternative to the OReb is end of possession -- 0 points -- then why wouldn't 6.13 OReb be worth 6.13 * 1.2 = 7.3 Pts/48 ?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
mystic
Joined: 18 Dec 2006
Posts: 21
PostPosted: Thu Feb 17, 2011 6:14 pm Post subject: Reply with quote
We are talking about the numbers from b2nb's website. Love improves the offensive rebounding by 4.2 per 100 opportunities over an average replacement. There are 41.18 rebounds per 48 minutes in average, that means Love adds 1.73 offensive rebounds per 48 minutes over an average replacement. That makes 1.94 to 2.02 points per 48 minutes, if we take 1.12 and 1.17 as the boundaries. With 1.2 it becomes 2.08.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
back2newbelf
Joined: 21 Jun 2005
Posts: 265
PostPosted: Thu Feb 17, 2011 6:31 pm Post subject: Reply with quote
Mike G is probably not convinced that this is a better way to measure rebounding impact than simple individual rebounding numbers. And if that's the case I can't blame him. I'm not convinced either. Until we have an out-of-sample-data test of the two methods we can't really say what's better. And I do know one thing where individual numbers would kill the adjusted 'whatever' method in out-of-sample-prediction: Free Throw Shooting
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
mystic
Joined: 18 Dec 2006
Posts: 21
PostPosted: Thu Feb 17, 2011 6:47 pm Post subject: Reply with quote
Well, you are right, that your numbers aren't confirmed, but they are looking quite convincing to me. For example: While David Lee and Troy Murphy were on the court, their teams usually got worse in rebounding. Your data shows that for both players. They have impressive individual rebounding numbers, but they usually skipped the boxing out part and took on a position to get the board. But rebounding overall is about boxing out and team positioning, not just about grabbing the rebound.
The opposite example would be Jason Collins. The guy basically is just boxing out and is making room for his teammates to grab the rebound on the defensive end. His individual rebounding numbers aren't great, but he has a +1.7 in your defensive rebounding per 100 opportunities.
I think your idea is a really good one. The numbers are basically in good agreement with my own judgement.
That's obviously not enough as a confirmation of your data, but we also know that each team grabs offensive rebounds even without Love on their roster. That means the alternative without Love and an average replacement in is for sure not 0 points.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
back2newbelf
Joined: 21 Jun 2005
Posts: 265
PostPosted: Thu Feb 17, 2011 7:02 pm Post subject: Reply with quote
mystic wrote:
I think your idea is a really good one.
Thanks, it wasn't my idea though.
@Ilardi: I guess I could do "adjusted 2nd chance points" some time. That way, the type of offensive rebound and how much it was worth is somewhat built in. Players who have mostly tip dunks or tip ins should benefit from this. Some of this should already be reflected in standard RAPM though
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 272
PostPosted: Thu Feb 17, 2011 7:16 pm Post subject: Reply with quote
Top 10 rebounders according to ezpm (> 2000 possessions):
Code:
LAST REB100
Love 4.42
Randolph 3.97
Humphries 3.89
Camby 3.60
Howard 3.34
Griffin 2.72
Okafor 2.10
Wade 1.94
Blair 1.77
Bogut 1.66
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
BobboFitos
Joined: 21 Feb 2009
Posts: 195
Location: Cambridge, MA
PostPosted: Thu Feb 17, 2011 11:06 pm Post subject: Reply with quote
Mike G wrote:
Quote:
If we take that and the average amount of rebounds per 48 minutes, we get between 1.9 and 2.0 points added per 48 minutes due to Kevin Love's offensive rebounding impact.
Kevin Love averages 6.13 OReb per 48 min.
If an OReb creates 1.14 times as many points as an average possession, that's 1.14 x 1.049 = 1.2 pts (if Wolves are typical in this regard).
If the alternative to the OReb is end of possession -- 0 points -- then why wouldn't 6.13 OReb be worth 6.13 * 1.2 = 7.3 Pts/48 ?
Shouldn't we instead find his marginal offensive rebounds above replacement player, and that's how much his Oboarding is worth?
I don't know what the replacement value is on oboards per 48, but obv whatever it is is some number between 1.2 pts and 7.3 pts.
_________________
-Rob
Back to top
View user's profile Send private message AIM Address
Mogilny
Joined: 05 Aug 2010
Posts: 23
PostPosted: Fri Feb 18, 2011 10:29 am Post subject: Reply with quote
Isn't a problem when measuring offensive rebounds in points like this, that the player might be out of his defensive position when he doesn't get the rebound? Even if the offensive rebound is worth a set value, the attempt to get the rebound might be worth less because the opponents doesn't attack against a set D? If this would be the case, doesn't the utility of measuring grabbed offensive rebounds in points diminish?
Sorry if the question is fuzzy, I'm from Sweden so english isn't my natural language. Smile
Back to top
View user's profile Send private message
DSMok1
Joined: 05 Aug 2009
Posts: 608
Location: Where the wind comes sweeping down the plains
PostPosted: Fri Feb 18, 2011 10:35 am Post subject: Reply with quote
Mogilny wrote:
Isn't a problem when measuring offensive rebounds in points like this, that the player might be out of his defensive position when he doesn't get the rebound? Even if the offensive rebound is worth a set value, the attempt to get the rebound might be worth less because the opponents doesn't attack against a set D? If this would be the case, doesn't the utility of measuring grabbed offensive rebounds in points diminish?
Sorry if the question is fuzzy, I'm from Sweden so english isn't my natural language. Smile
It's clear, and I agree.
_________________
GodismyJudgeOK.com/DStats
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mike G
Joined: 14 Jan 2005
Posts: 3583
Location: Hendersonville, NC
PostPosted: Fri Feb 18, 2011 11:17 am Post subject: Reply with quote
Conversely, doesn't the threat of an offensive rebounder cause the opponent to delay in getting out on the break? Until that defensive rebound is secured, players shouldn't be leaving the scene.
An elite OReb machine like Love or Randolph may cause some teams to actually double-team him to keep him off the glass. This may open up OReb opps for his teammates.
It looks as though Darko's DReb are down this year, but his OReb% is up. Tolliver is up somewhat, as well. Kosta Koufos suddenly gets OReb. Even guards Brewer and Webster are up from last year. It's a domino effect (other than Beasley).
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
mystic
Joined: 18 Dec 2006
Posts: 21
PostPosted: Fri Feb 18, 2011 11:42 am Post subject: Reply with quote
But somehow it doesn't help Minnesota to win games. As I pointed out before the correlation coefficient between ORB% and MOV is -0.21 right now. We have teams like the Spurs, Celtics, Heat and Mavericks which have a lower than average ORB%, on the other side we have the Timberwolves, Kings and Clippers with the best ORB% in the league.
Looking at the numbers by b2nb we will find a 0.07 correlation coefficient between offensive rebounding impact and overall APM. If offensive rebounding does help, it is a very small impact overall.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
EvanZ
Joined: 22 Nov 2010
Posts: 272
PostPosted: Fri Feb 18, 2011 12:08 pm Post subject: Reply with quote
mystic wrote:
But somehow it doesn't help Minnesota to win games. As I pointed out before the correlation coefficient between ORB% and MOV is -0.21 right now. We have teams like the Spurs, Celtics, Heat and Mavericks which have a lower than average ORB%, on the other side we have the Timberwolves, Kings and Clippers with the best ORB% in the league.
Just to add to that list, Denver is #1 in ORTG (112.4) and 29th in ORB% (23.2%). This goes back to the thread I made a few weeks ago showing a negative correlation between eFG and ORB.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
mystic
Joined: 18 Dec 2006
Posts: 21
PostPosted: Fri Feb 18, 2011 12:14 pm Post subject: Reply with quote
EvanZ wrote:
This goes back to the thread I made a few weeks ago showing a negative correlation between eFG and ORB.
Well, more missed shots are giving more offensive rebounding opportunities.
_________________
http://bbmetrics.wordpress.com
Back to top
View user's profile Send private message Visit poster's website
EvanZ
Joined: 22 Nov 2010
Posts: 272
PostPosted: Fri Feb 18, 2011 12:24 pm Post subject: Reply with quote
mystic wrote:
EvanZ wrote:
This goes back to the thread I made a few weeks ago showing a negative correlation between eFG and ORB.
Well, more missed shots are giving more offensive rebounding opportunities.
Sorry, I meant ORB%.
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3583
Location: Hendersonville, NC
PostPosted: Fri Feb 18, 2011 12:29 pm Post subject: Reply with quote
The top 100 ORb% teams of all time (> .345) have a median W% of .573
For the top 100 DRb%, it's .598
In this century, top 100 DRb% median W% is also .598
Top 100 ORb% median W% is just .512
Does this mean that metrics valuing OReb more than DReb have got it backwards?
OReb% is of course independent of missed shots. It's the percent of available misses gotten.
_________________
`
36% of all statistics are wrong
recovered page 5
Author Message
Jeff Fogle
Joined: 11 Jan 2011
Posts: 70
PostPosted: Fri Feb 18, 2011 12:45 pm Post subject: Reply with quote
Is there a connection Mike G between teams with lower ORB% and better defensive efficiencies (and, I guess allso between higher ORB% and worse defensive efficiencies)?
Some of ORB% looks to the naked eye to be coaching choices about having guys not crash the boards so they can get back on defense. Indiana's been emphasizing crashing the boards under Vogel, but then Wade had that TD pass alley-oop when they didn't get back (and then the Pacers had their worst offensive rebounding game under Vogel the rest of the way).
Interesting data on the correlation between orb% and drb% and winning. Are there correlations with defensive efficiency?
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 308
PostPosted: Fri Feb 18, 2011 12:56 pm Post subject: Reply with quote
Jeff, I posted some stuff on this a few weeks ago. It may be useful to you.
http://www.sonicscentral.com/apbrmetric ... php?t=2733
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Mogilny
Joined: 05 Aug 2010
Posts: 25
PostPosted: Fri Feb 18, 2011 1:01 pm Post subject: Reply with quote
EvanZ wrote:
Jeff, I posted some stuff on this a few weeks ago. It may be useful to you.
http://www.sonicscentral.com/apbrmetric ... php?t=2733
I just wrote a post which I didn't post about reading something like this but couldn't recall if it was on one of the blogs I have in my reader or if it was on this forum so I figured it would be better if I found the article/post first. When I hit back I see you have already posted it. Thanks for saving me the effort of looking it up. Laughing
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3626
Location: Hendersonville, NC
PostPosted: Fri Feb 18, 2011 1:05 pm Post subject: Reply with quote
This is all I'm doing:
http://bkref.com/tiny/WtsEt
The top 100 (about 28%) of teams since 2000 in ORb% (over 29.1) have a median DRtg of 105.5
The next 100 (27.2 - 29.1) median DRtg is 105.7
The 3rd 100 (25.2-27.1 ORb%) median DRtg is 105.5
The bottom 55 (15.5%), with ORb% from 20.9-25.1, median DRtg is 106.4
From this, there's no apparent correlation between a good ORb% and a good DRtg.
Intuitively, we can suppose that teams with poor shooters will send guys crashing the offensive boards. This doesn't make them a good team, but it makes them better. Better, that is, than just putting your best bad shooters out there without trying for OReb.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Mike G
Joined: 14 Jan 2005
Posts: 3626
Location: Hendersonville, NC
PostPosted: Fri Feb 18, 2011 1:07 pm Post subject: Reply with quote
Mogilny wrote:
EvanZ wrote:
Jeff, I posted some stuff on this a few weeks ago. It may be useful to you.
http://www.sonicscentral.com/apbrmetric ... php?t=2733
I just wrote a post which I didn't post about reading something like this but couldn't recall if it was on one of the blogs I have in my reader or if it was on this forum so I figured it would be better if I found the article/post first. When I hit back I see you have already posted it. Thanks for saving me the effort of looking it up.
If people would just post here, instead of all over the web, it would all be easier to find, for everyone, forever.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Crow
Joined: 20 Jan 2009
Posts: 825
PostPosted: Fri Feb 18, 2011 1:51 pm Post subject: Reply with quote
(material moved to own thread)
Last edited by Crow on Fri Feb 18, 2011 3:35 pm; edited 13 times in total
Back to top
View user's profile Send private message
Mogilny
Joined: 05 Aug 2010
Posts: 25
PostPosted: Fri Feb 18, 2011 1:52 pm Post subject: Reply with quote
Mike G wrote:
Mogilny wrote:
EvanZ wrote:
Jeff, I posted some stuff on this a few weeks ago. It may be useful to you.
http://www.sonicscentral.com/apbrmetric ... php?t=2733
I just wrote a post which I didn't post about reading something like this but couldn't recall if it was on one of the blogs I have in my reader or if it was on this forum so I figured it would be better if I found the article/post first. When I hit back I see you have already posted it. Thanks for saving me the effort of looking it up. Laughing
If people would just post here, instead of all over the web, it would all be easier to find, for everyone, forever.
I wouldn't mind it since this forum is the first thing I read in the mornings (even though I don't post much) and I could stop adding blogs to my reader. Smile
Back to top
View user's profile Send private message
Jeff Fogle
Joined: 11 Jan 2011
Posts: 70
PostPosted: Fri Feb 18, 2011 2:03 pm Post subject: Reply with quote
Thanks Evan, Mike, all;
Easy to cherry-pick some individual cases:
This Year:
Boston lowest ORB%, second best def eff
Last Year:
Boston 3rd lowest, 5th best
Orlando 6th lowest, 2nd best
Cleveland 10th lowest, 7th best (with LeBron)
08-09:
San Antonio lowest, 6th best
Orlando 3rd lowest, best
But, that's cherry-picking. As your data shows, there doesn't seem to be a general league-wide correlation in a way I was hoping. Maybe it's just a description of strategy for some teams. Wonder if there's a difference between playoff games and regular season games...meaning that part of the reason playoff scores go down is that teams crash the offensive boards less and send their perimeter guys back to protect against fast breaks once a shot goes up.
Back to top
View user's profile Send private message
back2newbelf
Joined: 21 Jun 2005
Posts: 278
PostPosted: Fri Mar 18, 2011 9:56 am Post subject: Reply with quote
Any ideas/comments on how to compare rebound prediction performance of various methods? I want to use RMSE between actual and predicted rebounding for every rebound opportunity but I`m not sure that this is the way to go. Because of the binary nature of rebounding, even the 'perfect' predictor will have a high RMSE*. Does that matter to us at all?
*Say, if you "knew" 30% would be the best prediction for offensive rebounding and there were 100 rebounding opportunities (30 of which would be rebounded by the offense) RMSE would be:
sqrt[{30*pow(0.3-1, 2)+70*pow(0.3-0, 2)}/100]=0.4583
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
back2newbelf
Joined: 21 Jun 2005
Posts: 278
PostPosted: Tue Apr 12, 2011 4:44 am Post subject: Reply with quote
I can't seem to find diminishing returns for rebounds, neither for adjusted rebounding nor when using individual rebounding numbers. Since everybody else seemed to find them I was hoping someone else could shed some light on the matter.
I split several games into training and test set, then I collected all individual rebounding numbers in the training set. Then, for each game in the test set add up the offensive rebounding %s of the attacking players and the defensive rebounding %s of the defending players. The numbers should be around 30% for offense, 70% for defense. Divide the offensive rebounding % by the sum of both %s to properly scale things. Use the resulting number to forecast offensive rebounding % of that 5on5 lineup. Compute forecast error for each 10man lineup using Brier Score
Now, doesn't the theory of diminishing returns say that once we find a 5on5 lineup that is forecasted, using above method, to grab X % of rebounds, with X being some number above "average reb %" it would actually be better to forecast a weighted sum of X and "average reb %", with the weights adding up to 1.0.
Say you have a lineup that is initially forecast to grab 40% of offensive rebounds, 30% would be average, it's better to forecast 35%
When I move intial forecasted defensive rebounding % closer to average my results get worse, even if I move them just a tiny bit closer to average. Lineups that were forecasted to grab between 40% and 42% of offensive rebounds (extremely high) did in fact grab 332 out of 796 (41.7%) and it's the same story in all intervals
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3626
Location: Hendersonville, NC
PostPosted: Tue Apr 12, 2011 6:34 am Post subject: Reply with quote
Interesting.
One possibility is that 'several games' is not a big enough (or representative) sample.
Intuitively, it seems possible that you could load up on offensive rebounders and keep getting more rebounds, to a point at least. If you have just one great OReb guy out there, defenses can concentrate on keeping that one guy off the boards. Like having multiple scorers will help everyone's shooting%.
Maybe the falloff is above 50%. DReb are almost always over 50%, and maybe that's where diminishing returns are found.
On second thought, maybe it's in the definition. You're dividing by total rebounds, and adding a .25 DReb% player to another .25 DReb% player, that 2nd rebounder won't get .25 of all DReb opps, but .25 of what's left?
.25 of .75 is just .19 . Is that a diminishing return?
Sorry, first cup of coffee. Sometimes the obvious thing is easiest to overlook, just saying.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
back2newbelf
Joined: 21 Jun 2005
Posts: 278
PostPosted: Tue Apr 12, 2011 6:55 am Post subject: Reply with quote
Mike G wrote:
Interesting.
One possibility is that 'several games' is not a big enough (or representative) sample.
It's a 1000 game sample. Does 'several' usually stand for 'not that many'? English isn't my first language. Online translation both gives words equivalent to "a huge amount" and "a tiny amount" in my first language
Quote:
Maybe the falloff is above 50%. DReb are almost always over 50%, and maybe that's where diminishing returns are found.
>50% almost never happens. Even the absolute worst defensive rebounding unit is usually predicted to rebound north of 50% on defense
Quote:
On second thought, maybe it's in the definition. You're dividing by total rebounds, and adding a .25 DReb% player to another .25 DReb% player, that 2nd rebounder won't get .25 of all DReb opps, but .25 of what's left?
.25 of .75 is just .19 . Is that a diminishing return?
Yeah it now believe that is the case here. If all attackers add up to 35% and defenders add up to 70%, the attackers are obviously not predicted to rebound 35% but 35/(35+70)=33.3% and thus 1.7% get lost. How much gets lost is dependent on the defensive lineup though
_________________
http://stats-for-the-nba.appspot.com/
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 308
PostPosted: Tue Apr 12, 2011 8:41 am Post subject: Reply with quote
back2newbelf wrote:
It's a 1000 game sample. Does 'several' usually stand for 'not that many'? English isn't my first language. Online translation both gives words equivalent to "a huge amount" and "a tiny amount" in my first language
"Several" usually means more than 3 but, say, less than 10. 1000 is quite a few more than that. Laughing
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir