APBRmetrics Forum Index APBRmetrics
The statistical revolution will not be televised.
 
 FAQFAQ   SearchSearch   MemberlistMemberlist   UsergroupsUsergroups   RegisterRegister 
 ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

Technical DRtg Issues

 
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion
View previous topic :: View next topic  
Author Message
davis21wylie2121



Joined: 13 Oct 2005
Posts: 373
Location: Atlanta, GA

PostPosted: Sat Sep 15, 2007 12:29 pm    Post subject: Technical DRtg Issues Reply with quote

Hey guys,

I was playing around with my BoP spreadsheet and realized that my DRtg numbers are slightly off (by a fraction of a point) when compared to Justin's numbers at B-R...

Code:
Name            Season          Tm      Stops   Stop%   DRtg
Michael Jordan  1995-96         CHI     672.2   0.570   99.1
Scottie Pippen  1995-96         CHI     586.5   0.544   100.2
Toni Kukoc      1995-96         CHI     379.0   0.472   103.3
Dennis Rodman   1995-96         CHI     480.1   0.602   97.8
Steve Kerr      1995-96         CHI     303.6   0.415   105.7
Ron Harper      1995-96         CHI     374.9   0.521   101.2
Luc Longley     1995-96         CHI     317.4   0.507   101.8
Bill Wennington 1995-96         CHI     188.2   0.463   103.6
Jud Buechler    1995-96         CHI     141.7   0.502   102.0
Dickey Simpkins 1995-96         CHI     119.5   0.457   103.9
Randy Brown     1995-96         CHI     154.5   0.603   97.7
Jason Caffey    1995-96         CHI     97.4    0.468   103.4
James Edwards   1995-96         CHI     44.8    0.428   105.1
John Salley     1995-96         CHI     44.6    0.611   97.4
Jack Haley      1995-96         CHI     1.2     0.435   104.8


My Stops totals check out perfectly with B-R, so the problem is in one of the steps after Stops are calculated. Here are the equations I'm working with now:

Code:
Stop% = (Stops*TmOppMin)/(TmOppPoss*PlayerMin)

DScPoss = TmOppFG+(1-(1-(TmOppFT/TmOppFTA))^2)*TmOppFTA*0.4

DPtsPerScPoss = TmOppPts/DScPoss

DRtg = TmDRtg+0.2*((100*DPtsPerScPoss*(1-Stop%))-TmDRtg)


Does anyone who is familiar with BoP see where my problem is? Keep in mind that I made sure it isn't a data problem by inputting Justin's data precisely, so the formulas must be to blame. I'd appreciate any help!
Back to top
View user's profile Send private message
Eli W



Joined: 01 Feb 2005
Posts: 286

PostPosted: Sat Sep 15, 2007 1:44 pm    Post subject: Reply with quote

Using my own spreadsheet (which calculates things in a slightly different order), my results match yours exactly and fail to match those on B-R.com. I'll try to take a closer look at this.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Eli W



Joined: 01 Feb 2005
Posts: 286

PostPosted: Sat Sep 15, 2007 2:22 pm    Post subject: Reply with quote

I don't what raw data source Dean used in BOP, but his Table 17.19 (page 215) seems to match B-R in DRtg. And in comparing our results to BOP, it seems things go wrong as soon as the transition from Stops to Stop% (BOP has Pippen at 0.55, we have 0.544). We're getting systematically lower Stop%'s than BOP. But strangely, we're also getting systematically lower DRtg's than BOP (BOP has Pippen at 101, we have 100.2). This suggests there could be multiple problems (simply correcting the Stop% won't fix the DRtg, it will only make the discrepancies greater). But it could just be one problem - if we're overestimating team defensive possessions, decreasing them could both raise Stop% and raise DRtg. That's my preliminary look, but I have to go watch football now.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
jkubatko



Joined: 05 Jan 2005
Posts: 508
Location: Columbus, OH

PostPosted: Sat Sep 15, 2007 2:55 pm    Post subject: Reply with quote

Guys, I vaguely recall having this problem too (i.e., I couldn't get my figures to match Dean's), and I think the "trick" is to calculate team defensive ratings this way:

Code:

tmPoss = FGA - (ORB/(ORB+(oppTRB-oppORB)))*(FGA-FG)*1.07 + TO + 0.4*FTA

oppPoss = oppFGA - (oppORB/(oppORB+(TRB-ORB)))*(oppFGA-oppFG)*1.07 + oppTO + 0.4*oppFTA

gmPoss = (tmPoss + oppPoss) / 2

tmDRtg = 100*(oppPTS/gmPoss)


Let me know if that does it.

Actually, this reminds me that I should probably change the way I calculate the team offensive and defensive ratings that are seen on the team and league pages.
_________________
Regards,
Justin Kubatko
Basketball Stats!
Back to top
View user's profile Send private message Send e-mail Visit poster's website
davis21wylie2121



Joined: 13 Oct 2005
Posts: 373
Location: Atlanta, GA

PostPosted: Sat Sep 15, 2007 3:22 pm    Post subject: Reply with quote

Okay, Justin, that trick worked -- provided I don't change TmOppPoss to gmPoss in the Stop% formula. But shouldn't that change be made as well? If we're considering (for all intents and purposes) gmPoss to be the number of possessions that the team faces on defense, then I think we should amend the Stop% formula to reflect this as well.

Anyway, thanks JQ and Justin, I'm glad we got to the bottom of this! Now it's time to watch football...
Back to top
View user's profile Send private message
davis21wylie2121



Joined: 13 Oct 2005
Posts: 373
Location: Atlanta, GA

PostPosted: Sun Sep 16, 2007 2:09 pm    Post subject: Reply with quote

Okay, I finally had the chance to update my database to reflect these changes, including using gmPoss instead of tmPoss/tmOppPoss in %Poss and Stop%. Everyone will probably be getting an e-mail about a new file on the APBRmetrics group, because I got rid of those (now-obsolete) BoP files and added an updated version that has complete stats since 1973-74. For pre-1978 seasons, I estimated turnovers by regressing 1978-2007 TO on height, age, and per-minute stat rates, and forced the team estimated TO totals to match the actual team totals. Which is a really long way of saying that I estimated ORtg/%Poss/DRtg for 1974-1977. Enjoy!
Back to top
View user's profile Send private message
HoopStudies



Joined: 30 Dec 2004
Posts: 533
Location: Near Philadelphia, PA

PostPosted: Sun Sep 16, 2007 4:14 pm    Post subject: Reply with quote

davis21wylie2121 wrote:
Okay, I finally had the chance to update my database to reflect these changes, including using gmPoss instead of tmPoss/tmOppPoss in %Poss and Stop%. Everyone will probably be getting an e-mail about a new file on the APBRmetrics group, because I got rid of those (now-obsolete) BoP files and added an updated version that has complete stats since 1973-74. For pre-1978 seasons, I estimated turnovers by regressing 1978-2007 TO on height, age, and per-minute stat rates, and forced the team estimated TO totals to match the actual team totals. Which is a really long way of saying that I estimated ORtg/%Poss/DRtg for 1974-1977. Enjoy!


This is good stuff. As I scroll through some of your estimates from pre-individual turnover times, I see the potential for so many stories about teams, players, and the league as a whole. So many guys I really don't know much about, never saw, but now start to get a feel for. There were some stories I put together from my historical survey in BoP, but these individual numbers could fill in many gaps and suggest some good new work.

It sounds like you took a reasonable approach to estimating individual turnovers, too. Can I ask what stat rates seemed significant in making those projections?
_________________
Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com
Back to top
View user's profile Send private message Visit poster's website
davis21wylie2121



Joined: 13 Oct 2005
Posts: 373
Location: Atlanta, GA

PostPosted: Sun Sep 16, 2007 5:27 pm    Post subject: Reply with quote

Here were the results of the regression I ran (all stats are per-minute rates except age and height):

Code:
        Coefficient     P-value
Age     -0.000523602    3.57168E-10
Height  0.000266024     4.51694E-12
FGA     0.081464898     5.42246E-49
3P      -0.007188371    0.860221746
3PA     -0.046094526    0.001733145
FT      0.002001249     0.910578245
FTA     0.109641656     3.97036E-25
ORB     0.044739409     0.000762564
TRB     -0.012419562    0.063348846
AST     0.164489547     1.0591E-161
STL     0.137513462     4.00804E-18
BLK     0.060551874     9.62937E-06
PF      0.148149673     5.7288E-123
PTS     -0.035157392    3.41241E-11


So the most significant predictors of turnovers per minute seem to be assists and personal fouls, followed by field goal attempts and free throw attempts. This makes sense, because turnovers usually result from people trying to make plays (assists, FGA, FTA) or big men setting illegal screens (PF). What made this a lot more accurate for 74-77 was the fact that the NBA tracked team TO during this period; all I had to do was sum team estimated TO, compare that to .97*actual team TO (because, traditionally, 97% of team turnovers are individual turnovers), and adjust indiv. est. TO to be higher or lower based on the discrepancy between the two totals.

I really want to run your stats this way for pre-1974 seasons as well, but the NBA's lack of team TO, ORb, Stl, & Blk tracking is going to make it very hard. I have estimated individual TO, ORb, Stl, & Blk for every player since 1952 using the same regression method, but the results won't hold water if the team totals don't make sense (for instance, w/o the team adjustment I was consistently getting low team TO totals because of individual regression to the mean). My initial thought is to see if league size had any bearing on the standard deviations of the team-totals in those missing categories, and if so, apply this predicted league standard deviation to the sum-totals of the estimated players for each team. If anyone has a better tactic, though, I'd really love to hear it.

I also should note that I forgot to add the ABA to the spreadsheet, since they tracked all individual stats through 1973-74 as well. I'll hopefully get around to that at some point in the near future.
Back to top
View user's profile Send private message
HoopStudies



Joined: 30 Dec 2004
Posts: 533
Location: Near Philadelphia, PA

PostPosted: Sun Sep 16, 2007 5:46 pm    Post subject: Reply with quote

davis21wylie2121 wrote:
...(for instance, w/o the team adjustment I was consistently getting low team TO totals because of individual regression to the mean)...


I was going to ask this, but I'd guess that you were low because turnovers have been trending down for so long. Players have been reducing their mistakes...

The lack of a good estimate for Wilt, Oscar, and Russell always bugs me. I did my best with Wilt and Russ for the one chapter, but I think it can be done better... Just not by me these days...
_________________
Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com
Back to top
View user's profile Send private message Visit poster's website
Mountain



Joined: 13 Mar 2007
Posts: 216

PostPosted: Mon Sep 24, 2007 12:29 pm    Post subject: Reply with quote

DW how do your numbers compare or fit with or do other things distinctive of the "starting point for basketball analysis" regression or the table for statistical +/-? I'd be interested in hearing you walk thru that.

I see 3pt made with a negative coefficient? Isnt that it contrast with assertions and the statiscal +/- regression results that 3pt shooting is a positive? Is it a eFG% treatment issue?
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 665
Location: Washington, DC

PostPosted: Mon Sep 24, 2007 1:25 pm    Post subject: Reply with quote

Are there any video archives available of some of these older games? One way of generating an estimate might be to actually sit down and track the stats for some of these older games and see how many blocks, steals, turnovers, etc. were actually taking place.
_________________
My blog
Back to top
View user's profile Send private message Visit poster's website AIM Address Yahoo Messenger
HoopStudies



Joined: 30 Dec 2004
Posts: 533
Location: Near Philadelphia, PA

PostPosted: Mon Sep 24, 2007 1:43 pm    Post subject: Reply with quote

kjb wrote:
Are there any video archives available of some of these older games? One way of generating an estimate might be to actually sit down and track the stats for some of these older games and see how many blocks, steals, turnovers, etc. were actually taking place.


It is a definite possibility in the not too distant future.
_________________
Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com
Back to top
View user's profile Send private message Visit poster's website
Display posts from previous:   
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion All times are GMT - 5 Hours
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © 2001, 2005 phpBB Group