APBRmetrics

page 1

Author Message
Jon Nichols



Joined: 18 Aug 2005
Posts: 368


PostPosted: Wed Jul 18, 2007 1:02 pm Post subject: Defensive Composite Score Reply with quote
I just "created" a defensive stat called Defensive Composite Score, and you can find the link here: http://www.82games.com/nichols1.htm I know that you guys here are the best of the best in terms of basketball statistics, so I was looking for input. My methodology is explained somewhat in the article, but I can explain it further if anybody is interested. How can I make this stat better? This stat has only had a little bit of tweaking involved, and I figure you guys would have some ideas on how to make it even more realistic.

Thanks,
Jon Nichols
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 855
Location: Washington, DC

PostPosted: Wed Jul 18, 2007 1:29 pm Post subject: Reply with quote
By using box score stats and DeanO's drtg, aren't you sorta double-counting the box score stats? Dean's system uses the same stats to estimate the rating. Which box score stats are you using? I'm sure blocks, steals and fouls. Defensive rebounds? On a per minute basis? Any thought of scaling DCS to 1000 instead of 100 to allow for greater differentiation between players? Do you adjust box score stats for position? In other words, guards are more likely to produce steals, centers more likely to produce blocks. Comparing the two isn't exactly an apples to apples thing.

Overall, I like the effort, and despite the questions, the results "feel" about right. Although, looking at my favorite team (the Wizards), Stevenson rates well below Jamison, which isn't right. Jamison is a truly poor defender. Stevenson isn't a shut-down defender, but he at least executes the team's defensive scheme (such as it is) and gives an effort. I'd describe him as "adequate" defensively. Jamison is "inadequate."
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 368


PostPosted: Wed Jul 18, 2007 2:31 pm Post subject: Reply with quote
Yes, it is a a bit of double counting. I was thinking of weighting either the box score stats lower or the defensive rating lower (probably that one because it seems to favor all players on good teams too much; I don't like Robert Horry being that high).

I'm only using blocks, steals, and (blocks + steals)/foul. Everything is on a per minute basis.

Scaling to 1000 would actually be a pretty good idea.

I didn't adjust for position, although I'm not exactly sure how I would do that. The problem is that all the players are ranked against every player in the league, not just at their position. I feel like since I'm doing that, I shouldn't adjust for position.

Thanks,
Jon
Back to top
View user's profile Send private message AIM Address
Neil Paine



Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA

PostPosted: Wed Jul 18, 2007 2:42 pm Post subject: Reply with quote
No use of 82games' counterpart PER in your system? When I did a system like this last year, I used Counterpart PER (25%), OnCourt DRtg (25%), On/Off DRtg Differential (25%), and Dean's boxscore DRtg (25%), which gave reasonably decent results. By ignoring counterpart production and double-counting boxscore stats, you're overvaluing help defenders (guys who rack up steals and blocks by leaving their man) and undervaluing man defenders.
Back to top
View user's profile Send private message Visit poster's website
Jon Nichols



Joined: 18 Aug 2005
Posts: 368


PostPosted: Wed Jul 18, 2007 2:51 pm Post subject: Reply with quote
Interesting. Does 82games have those numbers for each player as one number or do they only have it by position?
Back to top
View user's profile Send private message AIM Address
Charles



Joined: 16 May 2005
Posts: 134


PostPosted: Wed Jul 18, 2007 2:53 pm Post subject: Reply with quote
I often feel that people developing player ratings worry too much about passing the "smell test." I suppose this is done in an effort to make the numbers commercially acceptable, but too often the ratings seem to simply re-cycle the official data (often re-enforcing misconceptions.)

Therefore, I appreciate that you have included the OnCourt/OffCourt data, rather than relying solely on the league numbers. It takes considerable gumption to rank the reigning DPOY as the fourth best defensive big on his own team. But, I think you are probably on to something.

Dan Rosenbaum's superb study ranked Nene as Denver's top defender - 6th in the league over-all - while Camby didn't even make the top ten at his own position. And it doesn't appear that anything has changed since that study was done. Denver still consistently posts better defensive numbers with Nene on the court than Camby.


Code:

DENVER'S DEFENSE PER 100 POSSESSIONS

with Camby with Nene
2006-07 107.8 104.9 allowed 2.9 less with Nene on the court
2004-05 104.7 102.7 allowed 2.0 less with Nene on the court
2003-04 105.8 102.2 allowed 3.6 less with Nene on the court
2002-03* 98.4 91.0 allowed 7.4 less with Nene on the court

* per 48 minutes in 2002-03



Code:

DENVER'S DEFENSIVE REBOUNDING PERCENTAGE

Camby Nene
2006-07 67.7% 70.1%
2004-05 68.8% 71.7%
2003-04 67.3% 68.1%
2002-03 72.0% 74.0%


Perhaps, Camby has played against slightly stronger competition over-all, but Nene has started more than two-thirds of Denver's games and is on the court beside the other starters almost as much (Camby is on the court with Anthony 75% of his minutes, Nene 68% of his minutes.)

I think Camby's selection was one of the worst Award blunders ever. Yet, unlike the furor over Nash's MVPs, this is the first mention I have seen of it on this board.

Any way, back on topic, I think you have done a very nice job with a difficult task here. Looking forward to more.
Back to top
View user's profile Send private message
Neil Paine



Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA

PostPosted: Wed Jul 18, 2007 2:57 pm Post subject: Reply with quote
Under each team's main page (for instance, my unfortunate favorite team), it lists "opponent production", which is just counterpart PER across all positions. And Roland, if you're lurking out there, I want to say thanks for adding that feature a few years ago -- back in the day, you'd have to compile the data position-by-position for each player, which was a real chore.
Back to top
View user's profile Send private message Visit poster's website
Jon Nichols



Joined: 18 Aug 2005
Posts: 368


PostPosted: Wed Jul 18, 2007 3:09 pm Post subject: Reply with quote
One question about that. Correct me if I'm wrong (and I probably am), but aren't opponent PER and DRtg calculated in a somewhat similar way? Don't they just calculate how the opponents team's players at a certain position performed? In that case, wouldn't the only difference between counterpart PER and defensive rating be the difference in the way PER and offensive rating determine efficiency?
Back to top
View user's profile Send private message AIM Address
Neil Paine



Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA

PostPosted: Wed Jul 18, 2007 3:20 pm Post subject: Reply with quote
Putting aside the fact that PER and points per possession "ratings" are two completely different metrics measuring two completely different things (in fact, I always say PER technically measures nothing at all -- but that's a different debate), Dean's DRtg is based entirely on boxscore stats (STL, BLK, PF, DRB, etc.) with adjustments made using team defensive stats (forced FG misses, etc. are all estimated from team stats). In other words, it's a rough estimate, to be used mainly in the absence of 82games-style advanced stats. 82games counterpart PER, on the other hand, is determined from NBA.com game logs by tracking who is theoretically guarding whom for every 5-on-5 matchup in every minute of every game. The result is a metric that tracks how each player's man performed, which is usually a very good indicator of man-to-man ability. Sometimes cross-matching fouls up the system, but it's generally a valuable stat to include in any defensive meta-rating.
Back to top
View user's profile Send private message Visit poster's website
Jon Nichols



Joined: 18 Aug 2005
Posts: 368


PostPosted: Wed Jul 18, 2007 3:34 pm Post subject: Reply with quote
Well, it sounds like something that could be incorporated. Although, it does look like it punishes guys that play inside (Okafor has a higher counterpart PER than Walter Hermann, Raymond Felton, and Matt Carroll).
Back to top
View user's profile Send private message AIM Address
Chronz1



Joined: 22 May 2006
Posts: 198


PostPosted: Wed Jul 18, 2007 11:01 pm Post subject: Reply with quote
JNichols42887 wrote:
Well, it sounds like something that could be incorporated. Although, it does look like it punishes guys that play inside (Okafor has a higher counterpart PER than Walter Hermann, Raymond Felton, and Matt Carroll).

I dont think comparing counterpart PER of players who play at different positions is right.
Back to top
View user's profile Send private message
benji



Joined: 31 Dec 2004
Posts: 32


PostPosted: Thu Jul 19, 2007 1:22 am Post subject: Reply with quote
Suggestion:
The counterpart PER includes all their rebounds, blocks, etc. though. What I do in my own defensive statistical work, is just use the counterpart data for the rest of the stuff. Most of the stuff you would need (points, FGA, FTA, eFG, TO, AST, etc.) is already on there.
Back to top
View user's profile Send private message Send e-mail
kjb



Joined: 03 Jan 2005
Posts: 855
Location: Washington, DC

PostPosted: Thu Jul 19, 2007 8:45 am Post subject: Reply with quote
I'm personally not a fan of the counterpart data because it's based on an automated matchup system. It doesn't really say how good a man defender a player actually is. It tells you what the player designated to be his counterpart produced. But it may have nothing to do with that player -- he may have been matched against someone else. Plus, there's no accounting for switches, zones, transition or help D.

For example, back in 04-05, I tracked half a season of the Wizards defense. The counterpart data said that Hughes faced 18.8 FGA per 48 minutes. My tracking put it closer to 12 FGA per 48 minutes. Counterpart data said Haywood faced 11 shots per 48 minutes -- my tracking put him at 21+. These are HUGE differences, and they're very important, in my opinion.

Now, the Wizards are/were probably more extreme than most teams in this sort of thing. I've done limited tracking of other teams, and I think counterpart data may be a little better for some of those teams. But not enough better to be used as a major part of a defensive rating system.

Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Truthiness



Joined: 30 May 2007
Posts: 8


PostPosted: Thu Jul 19, 2007 10:12 am Post subject: Reply with quote
I enjoy the new numbers but the first thing that jumps out at me is the lack of guards being being rated higher. Its almost as if PF/C are rewarded for being big and being able to hover around the basket. The prime example for this is Tayshaun Prince being the second worst defender on the Pistons. He consistently gets the hardest matchup on the team and is probably one of the better defenders in the league.


I was also going to say that I think you are counting box score data twice but that has been pretty well covered. I think that this has potential to be a pretty good rating system but at the moment there are some early issues that I am sure you can figure out by playing with the numbers and weighting things differesntly.


Until Project Defensive Scoresheet is implimented (from Oliver's book), I think it will be really hard to have an extremely accurate rating for defense.
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 368


PostPosted: Thu Jul 19, 2007 10:35 am Post subject: Reply with quote
Big men in general do have better ratings with DCS, but I don't really mind that. I remember reading in Basketball on Paper when Oliver pointed out that so many teams acquire big men that are good defenders and poor on offense because of the importance of having good defense inside. It would be great if it's not skewed at all, but if it had to be, I would want big men to be rewarded. And it is still possible for a big guy to have a bad DCS -- see Eddy Curry.


page 2

Author Message
gabefarkas



Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC

PostPosted: Thu Jul 19, 2007 11:42 am Post subject: Reply with quote
kjb wrote:

For example, back in 04-05, I tracked half a season of the Wizards defense. The counterpart data said that Hughes faced 18.8 FGA per 48 minutes. My tracking put it closer to 12 FGA per 48 minutes. Counterpart data said Haywood faced 11 shots per 48 minutes -- my tracking put him at 21+. These are HUGE differences, and they're very important, in my opinion.

T-Mac is being guarded by Hughes out near the perimeter. Hughes, as he is known to do, gambles and goes for the steal. McGrady drives by him into the paint where he faces Haywood and attempts a lay up.

Your tracking (I'm assuming) would assign that FGA to Haywood. Counterpart data would assign it to Hughes. In this case, I'm inclined to agree with the Counterpart Data. Why? It was Hughes' defensive assignment and he blew it.
Back to top
View user's profile Send private message Send e-mail AIM Address
NickS



Joined: 30 Dec 2004
Posts: 384


PostPosted: Thu Jul 19, 2007 11:53 am Post subject: Reply with quote
I just want to echo what other people have said about their being overlap between the box score stats and DeanO's drtg.

That isn't necessarily a problem. If you wanted to you could have a metric which was just 2 parts drtg and 1 part +/- (which could make sense if you thought that +/- was less reliable than drtg).

Then the question becomes how do you weight box score stats differently from drtg? Put another way, what would the "stat-based" component of this rating look like without including the +/-? Who improves or suffers compared to drtg by itself.

The other question I have about +/- is whether you feel like it's "noisier" for low minute players than box score stats are. My anecdotal impressions is that for low minutes players +/- is less stable from year to year than box score stats, but I haven't looked at that systematically.
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 861
Location: Washington, DC

PostPosted: Thu Jul 19, 2007 12:07 pm Post subject: Reply with quote
gabefarkas wrote:
kjb wrote:

For example, back in 04-05, I tracked half a season of the Wizards defense. The counterpart data said that Hughes faced 18.8 FGA per 48 minutes. My tracking put it closer to 12 FGA per 48 minutes. Counterpart data said Haywood faced 11 shots per 48 minutes -- my tracking put him at 21+. These are HUGE differences, and they're very important, in my opinion.

T-Mac is being guarded by Hughes out near the perimeter. Hughes, as he is known to do, gambles and goes for the steal. McGrady drives by him into the paint where he faces Haywood and attempts a lay up.

Your tracking (I'm assuming) would assign that FGA to Haywood. Counterpart data would assign it to Hughes. In this case, I'm inclined to agree with the Counterpart Data. Why? It was Hughes' defensive assignment and he blew it.


Actually, it would split credit/blame between Haywood and Hughes. Which (in my opinion) is more accurate than the counterpart data because it accounts for Hughes as the primary defender and Haywood's help responsibilities.

By the way, your example is interesting because if the counterpart system matched McGrady to Hughes, it would have been wrong most of the time. The way the Wizards matchup, they likely would have put Jeffries on McGrady.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Ben F.



Joined: 07 Mar 2005
Posts: 391


PostPosted: Thu Jul 19, 2007 12:18 pm Post subject: Reply with quote
gabefarkas wrote:
T-Mac is being guarded by Hughes out near the perimeter. Hughes, as he is known to do, gambles and goes for the steal. McGrady drives by him into the paint where he faces Haywood and attempts a lay up.

Your tracking (I'm assuming) would assign that FGA to Haywood. Counterpart data would assign it to Hughes. In this case, I'm inclined to agree with the Counterpart Data. Why? It was Hughes' defensive assignment and he blew it.

Your argument makes sense with this example, but what about something like this: Hughes blows an assignment, Haywood comes to help and forces an awkward layup miss. Now Hughes gets credit for Haywood's effective help defense. kjb's tracking indicates that there is help defense on a large number of plays, and therefore that counterpart data is largely affected by who the help defender is on the floor.

Or to take your example one step farther, Hughes blows the assignment, Haywood comes to help and because of that TMac dishes to an open Yao for the dunk. With counterpart data Haywood gets penalized for that dunk, but really the breakdown is Hughes' fault.

There's a lot of "cloudiness" in this manner with counterpart data. I don't know whether it is more or less cloudy than any of the other measures we have out there, but it's not nearly as cut and dried as you make it sound.
Back to top
View user's profile Send private message
bchaikin



Joined: 27 Jan 2005
Posts: 681
Location: cleveland, ohio

PostPosted: Thu Jul 19, 2007 1:32 pm Post subject: Reply with quote
I just "created" a defensive stat called Defensive Composite Score

good job.... definitely a nice effort, plus the complete listing you've presented for the most part looks very good.....

I'm personally not a fan of the counterpart data because it's based on an automated matchup system. It doesn't really say how good a man defender a player actually is.

although i'm guessing alot of us (myself included) have quoted this data for just this purpose, as you stated the data simply shows that when PG A was on the floor this is what his team's opposing PGs did while he played, same for the SG, etc, and that in and of itself is incredibly valuable information, because what was once just a single team defensive FG% (or eFG%) is now broken down even further...

yes it doesn't take into consideration defensive switches, zones, or shots blocked by players other than the counterpart, but considering this information is gleaned simply from play-by-play data and not from actual video, which few of us have access to nor the time to view it if we had the access, its very useful and why i am a big fan of it...

for example if you go to:

http://www.82games.com/0607/06NJN7C.HTM

it shows opposing nj nets SFs shooting an eFG% of just 41.5% with a very low PER of 11.0 when vince carter was at SF, insinuating carter is a lock down defender. but if you watch new jersey you know jason kidd often guards the opposing SG or SF, giving carter the easier assignment. but what it does show is that someone (or a combination of someones) is playing awfully good defense...

same for bruce bowen in san antonio. his SF counterpart stats:

http://www.82games.com/0607/06SAS8C.HTM

infer he is a good defender at the SF position, but look at brent barry's SG counterpart stats:

http://www.82games.com/0607/06SAS5C.HTM

these numbers show a counterpart eFG% of just 40.5% and a low PER of just 11.3. but again if you watch the spurs you know bowen (or ginobili) routinely guard the tougher SF or SG when barry's in the game, so someone was playing great D...

after viewing this data for 3-4 years now, i find it extremely valuable for rating players defensively, and when used in conjunction with +/- or adjusted +/-, patterns do emerge helping to show which players are better or worse defenders...

Denver still consistently posts better defensive numbers with Nene on the court than Camby.

the numbers you showed aren't exclusively either/or - for example in 06-07 camby played 2369 minutes, nene just 1715, and from this page:

http://www.82games.com/0607/0607DEN2.HTM

it looks like camby and nene were on the floor together for at least 657 minutes, which is over 1/3 of nene's total playing time. both were good defenders, but if one is going to say nene is a better defender than camby in particular, then you are talking about just 1715 - 657 = 1058 total minutes of nene without camby. how significant is it to say a player playing those few minutes is a much better defender than one who played more and that in the eyes of some was one of the best defenders in the league?...

I think Camby's selection was one of the worst Award blunders ever. Yet, unlike the furor over Nash's MVPs, this is the first mention I have seen of it on this board.

the nuggets were just 11th best in the league in least points allowed per defensive team possession in 06-07, 12th best in 05-06, and actually gave up more points per team defensive possession in 06-07 compared to 05-06. the bulls went from 7th best in the league in 05-06 in least points allowed per team defensive possession, to best in the league in least points allowed per team defensive possession in 06-07, giving up 0.030 less points per defensive team possession. they improved offensively just 0.011 points per offensive team possession (23rd best in 05-06 to 19th best in 06-07)...

four bulls players - hinrich, deng, gordon, and duhon - played just over 1/2 of the team's total minutes played each season. ben wallace played about the same number of minutes as tim duncan did in 06-07 and the team improved from 41-41 to 49-33, but based solely on per game point differential, from an estimated 42-43 wins to 54 wins. but wallace got all of one first place vote for DPOY, and wasn't even named to the 1st or 2nd all-D teams...

the spurs finished 3rd in least points allowed per team defensive possession in 06-07, and bowen/duncan finished 2nd and 3rd in the DPOY voting and both were named to the all-D 1st team. go figure...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
gabefarkas



Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC

PostPosted: Thu Jul 19, 2007 3:12 pm Post subject: Reply with quote
kjb wrote:
Actually, it would split credit/blame between Haywood and Hughes. Which (in my opinion) is more accurate than the counterpart data because it accounts for Hughes as the primary defender and Haywood's help responsibilities.

Hmmm, I'm not sure I necessarily agree with that either. If Haywood blocks or successfully defends the shot, then I would say he deserves all the credit, not just half of it. If Haywood is caught off guard and can only half-heartedly defend the FGA, then I might say Hughes deserves most of the blame.

kjb wrote:
By the way, your example is interesting because if the counterpart system matched McGrady to Hughes, it would have been wrong most of the time. The way the Wizards matchup, they likely would have put Jeffries on McGrady.

Eh, just a hypothetical with the first guy that came to mind...
Back to top
View user's profile Send private message Send e-mail AIM Address
kjb



Joined: 03 Jan 2005
Posts: 861
Location: Washington, DC

PostPosted: Thu Jul 19, 2007 9:17 pm Post subject: Reply with quote
BenF's post was pretty interesting, and actually did a better job making my point than I did. Smile

Re: the splitting of credit/blame between Haywood and Hughes, the point Dean drummed into me is to avoid making judgements in the tracking about who I think deserves the credit/blame. Just record what happened. In the example given, Hughes allowed penetration, and Haywood helped. Whether Hughes' man scores or gets his shot blocked, the result ends up split between the two defenders. The relative values of different defensive activities shows up in analysis of the numbers.

Bob: I understand your point about how you use the counterpart data. I still think manual tracking is still the best way to understand individual contributions on the defensive end, but I concur that it is time-consuming. While it's true that counterpart data shows that someone played good defense, it often doesn't say who that someone was.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 20, 2007 3:12 pm Post subject: Reply with quote
How to fix the double counting: I've tinkered with adjusting DCS by either removing the box score stats I used completely, and I've also tried weighing the box score stats and defensive rating each 0.5. The latter seems to work the best. What does everyone think?

I also added in counterpart PER adjusted for position, and that was a great addition. It really helped guys like Tayshaun Prince, Bruce Bowen, and Raja Bell who were rated too low. However, it has also vaulted Manu Ginobili to the top of the rankings...

-Jon
Back to top
View user's profile Send private message AIM Address
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 20, 2007 3:21 pm Post subject: Reply with quote
Just for fun, this is how the new modified version of my stat ranks the top 10 vote getters for DPOY:

1. Tim Duncan - 994 (DPOY Rank:3)
2. Jermaine O'Neal - 982 (DPOY Rank:10)
3. Shane Battier - 963 (DPOY Rank:5)
4. Bruce Bowen - 933 (DPOY Rank:2)
5. Ron Artest - 823 (DPOY Rank:8)
6. Ben Wallace - 811 (DPOY Rank:6)
7. Marcus Camby - 771 (DPOY Rank:1)
8. Shawn Marion - 759 (DPOY Rank:4)
9. Tyson Chandler - 744 (DPOY Rank:9)
10. Gerald Wallace - 649 (DPOY Rank:7)


Okafor, who ranked 11th in voting, would have been 4th on the above list.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Fri Jul 20, 2007 4:57 pm Post subject: Reply with quote
That rank order for the top 10 DPOY votegetters seems ok. Using your new formula who are the top 10 among all players in the league? How well does it sort guys into top, middle and bottom?
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 20, 2007 5:22 pm Post subject: Reply with quote
The new top 10 is:

Ginobili - 997
Duncan - 994
Tyrus Thomas - 991
Garnett - 988
Jason Maxiell - 985
Jermaine O'Neal - 982
Balkman - 979
Wells - 976
Garbajosa - 973
Przybilla - 970

6 of those players weren't starters (7 if you count Ginobili). Players are getting punished for matching up against good players. If I can figure out some way to adjust for that, that would help.

It's pretty evenly distributed in terms of position throughout, with a slight skew favoring big men.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Fri Jul 20, 2007 5:30 pm Post subject: Reply with quote
Any big names, who you consider surprises to be near bottom, using new formula?
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 20, 2007 6:32 pm Post subject: Reply with quote
DeShawn Stevenson at 280 and Udonis Haslem at 357 are very low.
Back to top
View user's profile Send private message AIM Address
Chronz1



Joined: 22 May 2006
Posts: 200


PostPosted: Sat Jul 21, 2007 4:31 am Post subject: Reply with quote
So is the updated stat on the site now?
Back to top
View user's profile Send private message
supersub15



Joined: 21 Sep 2006
Posts: 273


PostPosted: Sat Jul 21, 2007 7:45 am Post subject: Reply with quote
Something that I have not seen mentioned yet is where defensive systems are set up in such a way that they allow the offensive player to get past the defender, but only in a certain direction on the court (i.e. funneling). A lot of times, fans think that the defender got beat, where in fact the defender is merely funneling the opponent into the help defense. It doesn't matter if the opponent ends up making the field goal with a hand in his face, the defender has done his job.

I'm not sure any metric can capture that.

page 3

Author Message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Sat Jul 21, 2007 3:00 pm Post subject: Reply with quote
Chronz1 wrote:
So is the updated stat on the site now?


No, I'm not going to try to get anything on the site until I make all the revisions that I want.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Sat Jul 21, 2007 6:00 pm Post subject: Reply with quote
Funneling on purpose and prepared for it has some logic with a superior inside defender (like Duncan) and team awareness/footspeed for rotations to cover his man. But perhaps even with an average one, playing the percentages that drives will yield less efficient on average midrange shots and well contested shots at the rim (instead of true surprises or a lot of damaging 3s) and last second passing decisions? How many teams can be said to use this strategy heavily and how good are their results?
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC

PostPosted: Mon Jul 23, 2007 8:54 am Post subject: Reply with quote
JNichols42887 wrote:
The new top 10 is:

Ginobili - 997
Duncan - 994
Tyrus Thomas - 991
Garnett - 988
Jason Maxiell - 985
Jermaine O'Neal - 982
Balkman - 979
Wells - 976
Garbajosa - 973
Przybilla - 970

6 of those players weren't starters (7 if you count Ginobili). Players are getting punished for matching up against good players. If I can figure out some way to adjust for that, that would help.

It's pretty evenly distributed in terms of position throughout, with a slight skew favoring big men.


You might try MikeG's system for adjusting stats based on time spent playing against starters.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
kjb



Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC

PostPosted: Mon Jul 23, 2007 9:09 am Post subject: Reply with quote
supersub15 wrote:
Something that I have not seen mentioned yet is where defensive systems are set up in such a way that they allow the offensive player to get past the defender, but only in a certain direction on the court (i.e. funneling). A lot of times, fans think that the defender got beat, where in fact the defender is merely funneling the opponent into the help defense. It doesn't matter if the opponent ends up making the field goal with a hand in his face, the defender has done his job.

I'm not sure any metric can capture that.


Tracking captures this, but I don't see any automated stat system getting at it.

Mountain: I think funneling makes sense even with an average defensive big because it makes things more predictable for the defense. The system makes things easier on that average big because he knows where the ball will be and therefore knows where he's supposed to be. He may not be as effective as Duncan would be, but he's likely to be more effective than he would be otherwise.

Also, it helps the perimeter defenders because they don't have to take everything away -- their jobs are more precisely defined. Force to the sideline, force to the baseline, don't let him shoot an uncontested jumper.

Finally, size matters when it comes to forcing misses. For example, in my tracking of the Wizards, the two best at forcing misses in 04-05 were Brendan Haywood and Kwame Brown. Now Haywood is actually a pretty good defender. Kwame was indifferent on the defensive end that season (which showed up in some ways in the tracking), but guys still missed when he was involved in defending them. Trailing behind those two were guys like Etan Thomas, Ruffin, Jeffries -- all tall guys.

The Wiz were NOT running a funneling system, and they weren't good defensively. But the point here is that tall guys do a better job of forcing misses than shorter guys. To the extent that a coach can get opponents shooting a greater proportion of their shots against tall defenders, he's likely to lower opponent shooting percentage and have a better all-around defense.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Tue Jul 24, 2007 12:04 am Post subject: Reply with quote
kjb wrote:
JNichols42887 wrote:
The new top 10 is:

Ginobili - 997
Duncan - 994
Tyrus Thomas - 991
Garnett - 988
Jason Maxiell - 985
Jermaine O'Neal - 982
Balkman - 979
Wells - 976
Garbajosa - 973
Przybilla - 970

6 of those players weren't starters (7 if you count Ginobili). Players are getting punished for matching up against good players. If I can figure out some way to adjust for that, that would help.

It's pretty evenly distributed in terms of position throughout, with a slight skew favoring big men.


You might try MikeG's system for adjusting stats based on time spent playing against starters.


Where can I find that? I'd be very interested.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Tue Jul 24, 2007 12:25 am Post subject: Reply with quote
Good points kjb. Following on them, I'd say confidence in the funneling system which leads to contested shots over tall interior defenders may also benefit by reducing panic fouls. Spurs funnel and have very low rate of fouls, lowest in league last season. The previous year Pistons were lowest, with B Wallace. I don't know if they funneled the same way but both Wallaces covered. Nuggets with Camby 2nd lowest in 06/07 in fouls committed, Hornets with Chandler 4th. They probably can funnel, but do they? Really all the 10 lowest teams on foul rate have a notable big man who may contribute to that, though Phoenix might need a longer look at Amare and/or Marion tape.

supersub15, when you introduced this subject were you thinking of the Spurs or another team?

If you know that you are facing a funneling team, what is in your strategy? Exercise greater caution about driving? Or make sure you draw contact? Or have a big man you can give a dump pass to and let him punish them? Set up more 3s or dish back more often for them? Rapid passing and backdoor options? Spurs take away the 3pt line as much as possible because it is a main threat and it tends to set off the funneling process too. I guess running is also an option for those who can rebound and run well. Looking at teams who did well on offense against Spurs some of these elements seem prominent.

Good defense is an advanced science and art.
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC

PostPosted: Tue Jul 24, 2007 5:48 am Post subject: Reply with quote
JNichols42887 wrote:
kjb wrote:
JNichols42887 wrote:
The new top 10 is:

Ginobili - 997
Duncan - 994
Tyrus Thomas - 991
Garnett - 988
Jason Maxiell - 985
Jermaine O'Neal - 982
Balkman - 979
Wells - 976
Garbajosa - 973
Przybilla - 970

6 of those players weren't starters (7 if you count Ginobili). Players are getting punished for matching up against good players. If I can figure out some way to adjust for that, that would help.

It's pretty evenly distributed in terms of position throughout, with a slight skew favoring big men.


You might try MikeG's system for adjusting stats based on time spent playing against starters.


Where can I find that? I'd be very interested.


Mike has posted how to do it on the boards. You can do a search or just send him a private message.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
supersub15



Joined: 21 Sep 2006
Posts: 273


PostPosted: Tue Jul 24, 2007 6:07 am Post subject: Reply with quote
Mountain wrote:
supersub15, when you introduced this subject were you thinking of the Spurs or another team?


That's the first team that came to my mind, actually. Very Happy

I'm pretty sure that all teams do it against certain players too, e.g. 2 seasons ago, I specifically watched how the Raptors were handling Paul Pierce in one game, and Morris Peterson "sat" on Pierce's right hand all game long and forced him to go left into the help defender or give up the ball. I love those game-in-a-game scenarios!
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC

PostPosted: Tue Jul 24, 2007 6:39 am Post subject: Reply with quote
[MikeG's system for adjusting stats based on time spent playing against starters.]

[Where can I find that? I'd be very interested.]

See the Balkman thread recently. Cherokee acb offers an alternative/improved version. The question is what to do with St% (the percent of a player's opponents which were starters).

I've inserted 'my' version of St% in front of the players listed.

.66 Ginobili - 997
.77 Duncan - 994
.45 Tyrus Thomas - 991
.73 Garnett - 988
.49 Jason Maxiell - 985
.75 Jermaine O'Neal - 982
.45 Balkman - 979
.49 Wells - 976
.78 Garbajosa - 973
.89 Przybilla - 970

Przy offers one of the strangest stat lines I've seen lately. He starts every game but plays just 1/3 of it. He blocks shots like crazy, but fouls twice as often. He offers no offense whatsoever, but somehow he gets turnovers.

Some of these guys are known 'defensive specialists off the bench'. Bonzi? I'd thought he just had a terrible year.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Tue Jul 24, 2007 10:17 am Post subject: Reply with quote
Thanks guys. Hopefully I can figure out some way to implement it, and I'll be sure to give credit if necessary.

-Jon Nichols
Back to top
View user's profile Send private message AIM Address
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Tue Jul 24, 2007 1:27 pm Post subject: Reply with quote
Here is the new top 10, adjusted for percentage of minutes against starters (thanks guys):

1. Joel Przybilla - 996.95 (wow)
2. Tim Duncan - 993.90
3. Jorge Garbajosa - 990.85
4. Jermaine O'Neal - 987.80
5. Kevin Garnett - 984.76
6. Yao Ming - 981.71
7. Shane Battier - 978.66
8. Manu Ginobili - 975.61
9. Zydrunas Ilgauskas - 972.56
10. Bruce Bowen - 969.51

If you're curious about another player, let me know. I think I like these new scores, although Pryzbilla is a bit concerning. But I suppose he is just an outlier.
Back to top
View user's profile Send private message AIM Address
Mike G



Joined: 14 Jan 2005
Posts: 3552
Location: Hendersonville, NC

PostPosted: Tue Jul 24, 2007 1:59 pm Post subject: Reply with quote
Why are all these players separated from one another by exactly 3.05 units on your scale? (I guess earlier each place was separated by 3, and I just hadn't noticed.)

If you're just converting 'rankings' (R) to '1000 - 3.05*R', why not just list the figures that gave you your rankings?
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Tue Jul 24, 2007 2:57 pm Post subject: Reply with quote
Mike G wrote:
Why are all these players separated from one another by exactly 3.05 units on your scale? (I guess earlier each place was separated by 3, and I just hadn't noticed.)

If you're just converting 'rankings' (R) to '1000 - 3.05*R', why not just list the figures that gave you your rankings?


I always thought that having everything relative and on a straight scale of 0-1000 would make things easier for people to grasp.
Back to top
View user's profile Send private message AIM Address
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Tue Jul 24, 2007 5:32 pm Post subject: Reply with quote
I also tried doing it without the box score stats altogether. The system now basically is a combination of defensive rating, +/-, and counterpart PER adjusted for position. The whole thing is then adjusted for playing time against starters. The top 10 with that system:

Quote:
Player Team DCS Final
przybilla,joel por 996.95
bowen,bruce san 993.90
duncan,tim san 990.85
ming,yao hou 987.80
o'neal,jermaine ind 984.76
ilgauskas,zydrun cle 981.71
garnett,kevin min 978.66
battier,shane hou 975.61
ginobili,manu san 972.56
brown,p.j. chi 969.51
Back to top
View user's profile Send private message AIM Address
NickS



Joined: 30 Dec 2004
Posts: 384


PostPosted: Tue Jul 24, 2007 7:25 pm Post subject: Reply with quote
JNichols42887 wrote:
Mike G wrote:
Why are all these players separated from one another by exactly 3.05 units on your scale? (I guess earlier each place was separated by 3, and I just hadn't noticed.)

If you're just converting 'rankings' (R) to '1000 - 3.05*R', why not just list the figures that gave you your rankings?


I always thought that having everything relative and on a straight scale of 0-1000 would make things easier for people to grasp.


The problem is that it hides the amount of difference between the players. What if the top 4 players are really close together, and there's a big drop-off at number 5? That's useful information, and by just giving rankings, you remove that information.

page 5

Author Message
Jon Nichols



Joined: 18 Aug 2005
Posts: 369


PostPosted: Wed Jul 25, 2007 3:04 pm Post subject: Reply with quote
kjb wrote:
JNichols42887 wrote:
I should add that I don't think someone like Caron Butler is a below average defender. It just so happens that he's on the Wizards, and that makes his numbers look bad. I would like to give a boost to good defenders on bad defenses and vice versa, but I feel like making a cutoff would be too arbitrary.


From tracking Caron, I'd say he's no better than average (at best). He can be good at times, but he's not real quick, and he's often shorter than the guys he faces. He's not terrible, but he's also not anything more than average.

One thought -- might the starter adjustment be doing too much work now? I wonder because Przybilla jumps ahead of Duncan, and the prevalence of starters at the top of the Wizards rankings. I'm looking at things like Jamison ahead of Blatche, which doesn't seem right. (My tracking of Blatche is pretty limited, but Jamison rates as pretty poor in tracking.)


You're right. It's really hard to decide how much "work" the starter adjustment should be doing. The decision ends up coming down to what my opinion is, which is exactly what I don't want the stat to be about. By the way, with the newest system, Przybilla is not ahead of Duncan anymore. Top 10 overall:

Quote:
duncan,tim 72.99027996
przybilla,joel 61.45264714
bowen,bruce 58.3495357
ming,yao 56.95056614
ginobili,manu 56.264105
garnett,kevin 54.49087084
ilgauskas,zydrun 49.93854064
thomas,tyrus 48.46593216
o'neal,jermaine 47.98615045
harris,devin 47.63975297
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 860
Location: Washington, DC

PostPosted: Wed Jul 25, 2007 3:17 pm Post subject: Reply with quote
Is this still 10x the z-score?

Now, for one more quibble. Smile I notice 3 guys from San Antonio in the top 10. How much of that is a result of those 3 playing together a significant amount of the time?

Would it be possible to run this system on previous seasons?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
NickS



Joined: 30 Dec 2004
Posts: 384


PostPosted: Wed Jul 25, 2007 3:51 pm Post subject: Reply with quote
Hmmm, I don't think these 3 values are independent of each other.

Look at the top players in the league. If they're 2+ SD above average on 3 separate measures that would mean that they're a [(1/40)^3] 1/64,000 player.

That makes me think that the 3 Z scores are not independent measures Smile As I said previously, I don't mind that, it's perfectly fine that some things are weighted more heavily, but it makes it harder to interpret the rating.

I don't know that you should change anything, but just noting.
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 369


PostPosted: Wed Jul 25, 2007 3:55 pm Post subject: Reply with quote
As soon as I perfect it for this year, I will do it on previous years.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Wed Jul 25, 2007 5:22 pm Post subject: Reply with quote
Short of having the adjusted defensive +/- on/off for a player (which hopefully will arrive, next season?) , you can compare difference of the team on/off for player and his principal sub and the difference in counterpart points allowed (or a measure of scoring efficiency is you prefer) for them and see how much of the on/off difference is coming from that position vs. elsewhere according to the recordkeeping system. If the later is high then the team on/off there is a good chance (but not a certainty) it is giving useful information on each player's defensive performance relative to each other. If the later difference is small then the team on/off difference could be coming from the quality of their help defense or largely be about the performance and quality of the other players. Or imperfections of the recordkeeping could be affecting the stats and the impression they make.

Your rankings could be different if you broke team on/off into these two components and then choose a set of weights for each part different than the roughly 1part for the counterpart defense difference-4parts for action recorded at the rest of positions as it is now in team on/off. Weighting the parts evenly or 2:1 in favor of the counterpart difference over action at the other positions appeals to me. But bring on the adjusted defensive +/- stats please anyone who can.

Last edited by Mountain on Wed Jul 25, 2007 5:55 pm; edited 3 times in total
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 860
Location: Washington, DC

PostPosted: Wed Jul 25, 2007 5:41 pm Post subject: Reply with quote
In my tracking of the Wizards defense, Haywood has shown up in 3 consecutive seasons as the team's best defender. This latest iteration is showing him as being among the top 8-10 defenders in the league. That seems high to me -- though I do think he's a good defender. I don't think I have any real suggestions except to maybe take a look at other ways to handle that starter adjustment.

Forgive me if you've addressed this already, but are you still using rankings in each category, or have you shifted to using z-scores there?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 369


PostPosted: Wed Jul 25, 2007 7:45 pm Post subject: Reply with quote
kjb wrote:
In my tracking of the Wizards defense, Haywood has shown up in 3 consecutive seasons as the team's best defender. This latest iteration is showing him as being among the top 8-10 defenders in the league. That seems high to me -- though I do think he's a good defender. I don't think I have any real suggestions except to maybe take a look at other ways to handle that starter adjustment.

Forgive me if you've addressed this already, but are you still using rankings in each category, or have you shifted to using z-scores there?


I believe I'm going to stick with just the z-scores.

Edit: Looking at the top 50, 11 are bench players. 28 in the top 100. I'm not sure if that's too low. What does everyone else think?
Back to top
View user's profile Send private message AIM Address
gabefarkas



Joined: 31 Dec 2004
Posts: 1311
Location: Durham, NC

PostPosted: Wed Jul 25, 2007 8:27 pm Post subject: Reply with quote
If your z-scores aren't normalized then I would wonder why not?
Back to top
View user's profile Send private message Send e-mail AIM Address
kjb



Joined: 03 Jan 2005
Posts: 860
Location: Washington, DC

PostPosted: Thu Jul 26, 2007 3:14 pm Post subject: Reply with quote
JNichols42887 wrote:
kjb wrote:
In my tracking of the Wizards defense, Haywood has shown up in 3 consecutive seasons as the team's best defender. This latest iteration is showing him as being among the top 8-10 defenders in the league. That seems high to me -- though I do think he's a good defender. I don't think I have any real suggestions except to maybe take a look at other ways to handle that starter adjustment.

Forgive me if you've addressed this already, but are you still using rankings in each category, or have you shifted to using z-scores there?


I believe I'm going to stick with just the z-scores.

Edit: Looking at the top 50, 11 are bench players. 28 in the top 100. I'm not sure if that's too low. What does everyone else think?


My opinion is that those numbers sound about right. Can you post numbers for the entire league (similar to what you did at 82games)?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 369


PostPosted: Thu Jul 26, 2007 4:22 pm Post subject: Reply with quote
Sorry for the long post...

Quote:
claxton,speedy atl 24.23
pachulia,zaza atl 23.36
smith,josh atl 20.61
childress,josh atl 8.27
johnson,joe atl 0.07
williams,marvin atl -3.55
williams,shelden atl -8.15
wright,lorenzen atl -17.32
ivey,royal atl -29.73
johnson,anthony atl -29.81
stoudamire,salim atl -33.31
jones,solomon atl -41.13
lue,tyronn atl -50.79

rondo,rajon bos 33.92
perkins,kendrick bos 20.47
jefferson,al bos 18.18
powe,leon bos 16.12
allen,tony bos 9.62
pierce,paul bos 3.72
west,delonte bos -7.32
scalabrine,brian bos -10.62
ray,allan bos -11.10
gomes,ryan bos -19.21
green,gerald bos -22.60
szczerbiak,wally bos -25.00
telfair,sebastia bos -28.49

okafor,emeka cha 39.14
felton,raymond cha 14.50
wallace,gerald cha 10.63
herrmann,walter cha 4.21
anderson,derek cha 4.00
knight,brevin cha 1.41
may,sean cha -4.86
carroll,matt cha -13.78
brezec,primoz cha -28.33
morrison,adam cha -28.60
mcinnis,jeff cha -32.12
voskuhl,jake cha -40.66

thomas,tyrus chi 48.47
deng,luol chi 42.12
wallace,ben chi 41.14
brown,p.j. chi 39.56
duhon,chris chi 29.42
hinrich,kirk chi 24.13
gordon,ben chi 20.30
sefolosha,thabo chi 15.29
nocioni,andres chi 10.09
griffin,adrian chi 2.60
allen,malik chi -0.47

ilgauskas,zydrun cle 49.94
james,lebron cle 37.08
pavlovic,sasha cle 35.11
hughes,larry cle 27.83
varejao,anderson cle 22.77
gooden,drew cle 16.75
snow,eric cle 16.09
marshall,donyell cle 10.11
gibson,daniel cle -1.07
jones,damon cle -17.06

harris,devin dal 47.64
dampier,erick dal 39.30
nowitzki,dirk dal 34.58
howard,josh dal 30.75
terry,jason dal 13.80
diop,desagana dal 8.15
george,devean dal 3.50
stackhouse,jerry dal -4.91
croshere,austin dal -10.48
buckner,greg dal -21.12

camby,marcus den 34.98
hilario,nene den 23.82
evans,reggie den 18.36
iverson,allen den 12.22
diawara,yakhouba den -5.48
kleiza,linas den -5.84
smith,j.r. den -12.60
najera,eduardo den -12.93
blake,steve den -14.44
anthony,carmelo den -20.63

maxiell,jason det 26.08
wallace,rasheed det 20.37
prince,tayshaun det 17.71
mcdyess,antonio det 15.38
hamilton,richard det 10.72
delfino,carlos det 8.23
billups,chauncey det 6.67
webber,chris det 2.89
mohammed,nazr det -8.98
murray,ronald det -10.02
hunter,lindsey det -17.82

biedrins,andris gsw 25.34
davis,baron gsw 24.53
jackson,stephen gsw 12.19
harrington,al gsw 10.77
richardson,jason gsw 5.28
ellis,monta gsw 0.22
pietrus,mickael gsw -13.47
barnes,matt gsw -14.87
jasikevicius,sar gsw -20.16
azubuike,kelenna gsw -44.07

ming,yao hou 56.95
hayes,chuck hou 46.11
wells,bonzi hou 45.72
battier,shane hou 41.33
mcgrady,tracy hou 40.39
alston,rafer hou 34.69
mutombo,dikembe hou 33.68
snyder,kirk hou -7.94
howard,juwan hou -9.37
head,luther hou -17.12

o'neal,jermaine ind 47.99
foster,jeff ind 17.54
diogu,ike ind 12.56
dunleavy,mike ind 8.58
tinsley,jamaal ind 7.49
granger,danny ind -1.51
daniels,marquis ind -9.80
murphy,troy ind -16.08
armstrong,darrel ind -19.03
mcleod,keith ind -25.82
williams,shawne ind -67.66

brand,elton lac 24.74
ewing,daniel lac 19.40
thomas,tim lac 13.60
kaman,chris lac 10.10
maggette,corey lac 7.94
mobley,cuttino lac 5.71
cassell,sam lac 2.37
ross,quinton lac -3.16
livingston,shaun lac -19.30
hart,jason lac -24.90

odom,lamar lal 15.56
bryant,kobe lal 10.99
bynum,andrew lal 3.75
farmar,jordan lal 3.20
brown,kwame lal 2.17
walton,luke lal 1.23
evans,maurice lal -17.48
parker,smush lal -17.59
turiaf,ronny lal -20.34
cook,brian lal -21.01
radmanovic,vladi lal -25.21
vujacic,sasha lal -39.32

gay,rudy mem -2.24
gasol,pau mem -12.24
stoudamire,damon mem -12.39
miller,mike mem -13.51
harrington,junio mem -16.58
swift,stromile mem -22.53
roberts,lawrence mem -22.62
johnson,alexande mem -25.27
atkins,chucky mem -30.15
warrick,hakim mem -30.37
kinsey,tarance mem -40.48
jones,dahntay mem -48.13

o'neal,shaquille mia 31.51
jones,eddie mia 28.14
mourning,alonzo mia 22.98
wade,dwyane mia 22.53
williams,jason mia 18.70
haslem,udonis mia 12.07
walker,antoine mia 2.61
posey,james mia -0.69
wright,dorell mia -1.68
doleac,michael mia -13.52
kapono,jason mia -15.49
payton,gary mia -27.24

bogut,andrew mil 13.22
skinner,brian mil 5.89
redd,michael mil 3.96
boykins,earl mil -2.33
williams,mo mil -11.43
patterson,ruben mil -11.87
villanueva,charl mil -18.67
gadzuric,dan mil -25.07
bell,charlie mil -27.06
ilyasova,ersan mil -45.47
noel,david mil -66.69

garnett,kevin min 54.49
jaric,marko min -4.26
davis,ricky min -4.30
blount,mark min -6.54
mccants,rashad min -8.07
smith,craig min -8.24
hassell,trenton min -16.13
hudson,troy min -20.07
james,mike min -21.37
foye,randy min -23.73

collins,jason njn 30.61
krstic,nenad njn 20.35
kidd,jason njn 15.51
carter,vince njn 13.48
boone,josh njn 11.34
jefferson,richar njn 6.01
moore,mikki njn -1.27
wright,antoine njn -4.63
nachbar,bostjan njn -5.16
williams,marcus njn -21.21
house,eddie njn -31.12
robinson,cliffor njn -43.49

chandler,tyson nor 23.46
armstrong,hilton nor 21.05
pargo,jannero nor 6.79
butler,rasual nor 6.67
jackson,bobby nor 3.88
west,david nor -0.87
brown,devin nor -4.83
paul,chris nor -6.63
jackson,marc nor -8.81
johnson,linton nor -10.24
mason,desmond nor -11.58
simmons,cedric nor -18.62

balkman,renaldo nyk 27.62
lee,david nyk 14.75
frye,channing nyk 4.96
crawford,jamal nyk -4.08
collins,mardy nyk -5.37
richardson,quent nyk -8.05
curry,eddy nyk -10.44
francis,steve nyk -10.62
jeffries,jared nyk -17.82
robinson,nate nyk -19.79
marbury,stephon nyk -20.28
rose,malik nyk -39.82

howard,dwight orl 44.09
battie,tony orl 34.90
ariza,trevor orl 17.12
milicic,darko orl 15.10
hill,grant orl 10.73
turkoglu,hedo orl 9.77
arroyo,carlos orl 4.31
dooling,keyon orl -4.32
nelson,jameer orl -5.90
bogans,keith orl -11.96
redick,j.j. orl -32.51

dalembert,samuel phi 28.45
williams,louis phi 16.00
smith,joe phi 8.33
ollie,kevin phi 8.14
hunter,steven phi 6.21
miller,andre phi 4.20
iguodala,andre phi -0.18
carney,rodney phi -13.96
green,willie phi -22.32
korver,kyle phi -32.95

marion,shawn pho 22.58
thomas,kurt pho 15.61
stoudemire,amare pho 11.13
bell,raja pho 9.81
nash,steve pho -0.36
diaw,boris pho -0.85
jones,james pho -1.15
barbosa,leandro pho -2.42
banks,marcus pho -64.61

przybilla,joel por 61.45
randolph,zach por 8.97
magloire,jamaal por 0.78
aldridge,lamarcu por -2.89
udoka,ime por -6.42
jack,jarrett por -14.09
outlaw,travis por -15.41
rodriguez,sergio por -25.15
roy,brandon por -26.65
jones,fred por -27.90
webster,martell por -29.22

miller,brad sac 24.51
artest,ron sac 17.23
thomas,kenny sac 4.52
williamson,corli sac -3.53
bibby,mike sac -5.82
abdur-rahim,shar sac -12.24
garcia,francisco sac -16.01
price,ronnie sac -16.16
martin,kevin sac -17.09
salmons,john sac -17.54

duncan,tim san 72.99
bowen,bruce san 58.35
ginobili,manu san 56.26
parker,tony san 40.03
horry,robert san 25.18
elson,francisco san 17.24
barry,brent san 7.54
oberto,fabricio san 7.17
bonner,matt san 1.88
vaughn,jacque san -3.55
udrih,beno san -12.75
finley,michael san -16.35

lewis,rashard sea -1.49
collison,nick sea -3.40
allen,ray sea -5.34
wilcox,chris sea -12.88
ridnour,luke sea -13.22
watson,earl sea -14.07
wilks,mike sea -21.47
gelabale,mickael sea -21.86
wilkins,damien sea -26.34
petro,johan sea -32.43

garbajosa,jorge tor 41.85
nesterovic,rasho tor 41.07
parker,anthony tor 33.52
bosh,chris tor 27.06
ford,t.j. tor 12.97
bargnani,andrea tor -9.57
calderon,jose tor -16.36
graham,joey tor -25.15
peterson,morris tor -25.88
dixon,juan tor -31.71
humphries,kris tor -35.94

kirilenko,andrei uta 35.98
boozer,carlos uta 16.82
millsap,paul uta 12.58
giricek,gordan uta 7.96
collins,jarron uta 4.33
brewer,ronnie uta 0.02
fisher,derek uta -3.41
harpring,matt uta -12.82
williams,deron uta -13.44
okur,mehmet uta -25.68

haywood,brendan was 21.55
butler,caron was -3.66
stevenson,deshaw was -7.89
arenas,gilbert was -13.32
thomas,etan was -13.33
jamison,antawn was -16.57
songaila,darius was -16.90
daniels,antonio was -20.66
blatche,andray was -25.68
hayes,jarvis was -58.01
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 860
Location: Washington, DC

PostPosted: Thu Jul 26, 2007 4:51 pm Post subject: Reply with quote
Your results confuse me a bit. My understanding is that with a normal distribution, a z-score over 2.0 (i.e. 2+ standard deviations from the mean) would represent a score in the top 2% (roughly speaking). You have 64 players (out of 329 listed) showing up with z-scores over 2.0 (I divided your defender score by 10). That's about 19% of the players -- not 2%.

Maybe I'm missing something -- I'm not the strongest math guy here, so that's always a possibility. What am I not understanding?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 369


PostPosted: Thu Jul 26, 2007 5:23 pm Post subject: Reply with quote
It's a combination of three z-scores. So (I'm also not a math major) I assume that to be in the top 2% a player would have to have a z-score of 6 (3*2). Then again, maybe I'm breaking all the rules of mathematics by doing that...

If you look at it, only Duncan has a rating above 60. But the ratings aren't simply z-scores (they have other modifications), so you can't really judge it by that. The z-scores for each individual category do fit in with what you expect.
Back to top
View user's profile Send private message AIM Address
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Thu Jul 26, 2007 6:03 pm Post subject: Reply with quote
JNichols:
Quote:
Big men in general do have better ratings with DCS, but I don't really mind that. I remember reading in Basketball on Paper when Oliver pointed out that so many teams acquire big men that are good defenders and poor on offense because of the importance of having good defense inside. It would be great if it's not skewed at all, but if it had to be, I would want big men to be rewarded. And it is still possible for a big guy to have a bad DCS -- see Eddy Curry.


Yeah. We know that a good inside defender big man provoke to shoot more from distance, and like funneling, perimeter defenders get (opp FG%) credits through that when counterparting and even in plus-minus. And that has some correlation with the big man's ability to block, steal, deflect, or just to force bad inside shots and then rebound, regarding the perimeter defender's ability to stay in front his man. And here is the same problem with offensive usage: should a defender who faces more attempts have some usage rewarding?

I like counterparting, not PER counterparting that crosses everything and double counts, but a correlated stats counterparting, and weighted by 4 factors; and all kind of adjusts, even time quality (clutch) adjusts that nobody has talked about. I would prefer adjusts than composite. It's difficult to weight different methods, but I think DanR did something that related plus-minus with boxscore stats.
Back to top
View user's profile Send private message
Ben F.



Joined: 07 Mar 2005
Posts: 391


PostPosted: Thu Jul 26, 2007 6:12 pm Post subject: Reply with quote
Putting aside this study for now (although I think it's good work), I want to focus on why exactly we are trying to lump everyone into one rating. It doesn't make sense to me to say: "Bruce Bowen is a better defender than Tim Duncan." They fill very different roles, and nobody who's evaluating a defense is going to wonder what would happen if you exchanged the two. So why compare them?

I wonder if we can't separate defenders into different categories and evaluate them in different ways based on their category (my proposal would be as simple as "man" and "help"). My guess is that blocks, opponent on/off eFG% and stats like that are much more the work of the help defenders (the Duncans of the world) and that counterpart stats are probably better put to use regarding the Bowens of the world. I wouldn't put all that much stock in, for example, Camby's counterpart numbers because he's forced to help a lot putting him out of position on his man. I would focus on his blocks and his effect on opponent shooting (which are tied together obviously, since a block is a forced miss, but I think a mix helps to take away some teammate responsibility while still seeing whether the player's non-block help is effective).

This is nothing concrete yet, and I'd invite all of your input, but I think that trying to evaluate Duncan vs. Haywood vs. Zaza Pachulia is important, but Earl Watson vs. Kwame Brown is not.
Back to top
View user's profile Send private message
NickS



Joined: 30 Dec 2004
Posts: 384


PostPosted: Thu Jul 26, 2007 6:21 pm Post subject: Reply with quote
Interesting to see the work done by the Starters/sub adjustment in the Seattle rankings. Rashard Lewis shows up as (slightly) better than Nick Collison, despite Nick Collison having better +/- and slightly better dRtg.

What percentage of time do you have Nick Collison playing against starters?

page 6

Author Message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Jul 26, 2007 6:22 pm Post subject: Reply with quote
Ben F. wrote:
Putting aside this study for now (although I think it's good work), I want to focus on why exactly we are trying to lump everyone into one rating. It doesn't make sense to me to say: "Bruce Bowen is a better defender than Tim Duncan." They fill very different roles, and nobody who's evaluating a defense is going to wonder what would happen if you exchanged the two. So why compare them?

I wonder if we can't separate defenders into different categories and evaluate them in different ways based on their category (my proposal would be as simple as "man" and "help"). My guess is that blocks, opponent on/off eFG% and stats like that are much more the work of the help defenders (the Duncans of the world) and that counterpart stats are probably better put to use regarding the Bowens of the world. I wouldn't put all that much stock in, for example, Camby's counterpart numbers because he's forced to help a lot putting him out of position on his man. I would focus on his blocks and his effect on opponent shooting (which are tied together obviously, since a block is a forced miss, but I think a mix helps to take away some teammate responsibility while still seeing whether the player's non-block help is effective).

This is nothing concrete yet, and I'd invite all of your input, but I think that trying to evaluate Duncan vs. Haywood vs. Zaza Pachulia is important, but Earl Watson vs. Kwame Brown is not.


I agree. I don't think it's possible to create one stat that does all the work. And actually, despite what I used to think, there are quite a few defensive statistics out there that can tell us a lot of things. In the future I personally wouldn't have a problem researching the kinds of things you suggest.
Back to top
View user's profile Send private message AIM Address
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Jul 26, 2007 6:24 pm Post subject: Reply with quote
NickS wrote:
Interesting to see the work done by the Starters/sub adjustment in the Seattle rankings. Rashard Lewis shows up as (slightly) better than Nick Collison, despite Nick Collison having better +/- and slightly better dRtg.

What percentage of time do you have Nick Collison playing against starters?


70.7%. Lewis at 70.9%.
Back to top
View user's profile Send private message AIM Address
NickS



Joined: 30 Dec 2004
Posts: 384


PostPosted: Thu Jul 26, 2007 6:54 pm Post subject: Reply with quote
JNichols42887 wrote:
70.7%. Lewis at 70.9%.


Ah, right, I presume it is counterpart ratings that hurt collison's rankings. If it isn't that, I'd be curious why Collison ranks lower than Lewis.
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Jul 26, 2007 8:03 pm Post subject: Reply with quote
NickS wrote:
JNichols42887 wrote:
70.7%. Lewis at 70.9%.


Ah, right, I presume it is counterpart ratings that hurt collison's rankings. If it isn't that, I'd be curious why Collison ranks lower than Lewis.


You are correct. The counterpart ratings killed Collison.
Back to top
View user's profile Send private message AIM Address
supersub15



Joined: 21 Sep 2006
Posts: 273


PostPosted: Fri Jul 27, 2007 7:17 am Post subject: Reply with quote
2 comments:

1. Why are the slow-paced teams' players all at the top? Houston and San Antonio have 9 players in the top 25. Tracy McGrady?
Shouldn't this metric be pace-adjusted, somehow?

2. Wouldn't it be equally beneficial to have an OCS (offensive composite score)? This way, it is possible to figure out who are the better all-around players, check out differential leaders (i.e. Bowen would probably be in the bottom 25 in OCS).
Back to top
View user's profile Send private message
gabefarkas



Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC

PostPosted: Fri Jul 27, 2007 7:56 am Post subject: Reply with quote
JNichols42887 wrote:
It's a combination of three z-scores. So (I'm also not a math major) I assume that to be in the top 2% a player would have to have a z-score of 6 (3*2). Then again, maybe I'm breaking all the rules of mathematics by doing that...


Sort of, yeah. But then again it's not a bad way to draw equal comparisons.

My suggestion would be to first make a combined metric, then take the z-score of the combined metric, rather than combining z-scores.
Back to top
View user's profile Send private message Send e-mail AIM Address
Statman



Joined: 20 Feb 2005
Posts: 241
Location: Arlington, Texas

PostPosted: Fri Jul 27, 2007 8:28 am Post subject: Reply with quote
Dirk Nowitski has a better score than Josh Howard? Interesting....
_________________
Dan

My current national college player rankings (and other stuff):
http://www.pointguardu.com/f136/statman ... post355594
Back to top
View user's profile Send private message Send e-mail
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 27, 2007 10:54 am Post subject: Reply with quote
supersub15 wrote:
2 comments:

1. Why are the slow-paced teams' players all at the top? Houston and San Antonio have 9 players in the top 25. Tracy McGrady?
Shouldn't this metric be pace-adjusted, somehow?

2. Wouldn't it be equally beneficial to have an OCS (offensive composite score)? This way, it is possible to figure out who are the better all-around players, check out differential leaders (i.e. Bowen would probably be in the bottom 25 in OCS).


It should be pace-adjusted, considering that every stat that was inputted was pace-adjusted (+/-, defensive rating, counterpart PER, percentage of playing time). I assume it's just because Houston and San Antonio happen to be great defensive teams.

I've thought about doing offensive composite score, but I'm not sure how much I could improve on the offensive stuff that's already out there.
Back to top
View user's profile Send private message AIM Address
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 27, 2007 10:56 am Post subject: Reply with quote
Statman wrote:
Dirk Nowitski has a better score than Josh Howard? Interesting....


I figure that's because Howard draws tougher defensive assignments every night.
Back to top
View user's profile Send private message AIM Address
Statman



Joined: 20 Feb 2005
Posts: 241
Location: Arlington, Texas

PostPosted: Fri Jul 27, 2007 11:47 am Post subject: Reply with quote
JNichols42887 wrote:
Statman wrote:
Dirk Nowitski has a better score than Josh Howard? Interesting....


I figure that's because Howard draws tougher defensive assignments every night.


Agreed.

Too bad the counterpart PER doesn't include the average PER of the counterparts.....

Maybe a RATIO. (Actual counterpart PER)/(Average PER of counterparts normally).

As it is - if a player is always given the toughest defensive assignment, then of course his counterpart PER will be much higher than others - since his counterparts have a much higher PER on average.

I'd like to know how much the player seems to effect the opponent's usual PER - I think that would be a much better possible indicator of defensive impact.
_________________
Dan

My current national college player rankings (and other stuff):
http://www.pointguardu.com/f136/statman ... post355594
Back to top
View user's profile Send private message Send e-mail
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 27, 2007 12:08 pm Post subject: Reply with quote
Statman wrote:
JNichols42887 wrote:
Statman wrote:
Dirk Nowitski has a better score than Josh Howard? Interesting....


I figure that's because Howard draws tougher defensive assignments every night.


Agreed.

Too bad the counterpart PER doesn't include the average PER of the counterparts.....

Maybe a RATIO. (Actual counterpart PER)/(Average PER of counterparts normally).

As it is - if a player is always given the toughest defensive assignment, then of course his counterpart PER will be much higher than others - since his counterparts have a much higher PER on average.

I'd like to know how much the player seems to effect the opponent's usual PER - I think that would be a much better possible indicator of defensive impact.


That would be really nice, although I'm not sure how you would do that...
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC

PostPosted: Fri Jul 27, 2007 2:52 pm Post subject: Reply with quote
JNichols42887 wrote:
supersub15 wrote:
2 comments:

1. Why are the slow-paced teams' players all at the top? Houston and San Antonio have 9 players in the top 25. Tracy McGrady?
Shouldn't this metric be pace-adjusted, somehow?

2. Wouldn't it be equally beneficial to have an OCS (offensive composite score)? This way, it is possible to figure out who are the better all-around players, check out differential leaders (i.e. Bowen would probably be in the bottom 25 in OCS).


It should be pace-adjusted, considering that every stat that was inputted was pace-adjusted (+/-, defensive rating, counterpart PER, percentage of playing time). I assume it's just because Houston and San Antonio happen to be great defensive teams.

I've thought about doing offensive composite score, but I'm not sure how much I could improve on the offensive stuff that's already out there.


If you have the time, I'd love to see an OCS. That might be interesting.

I took your numbers and did what Gabe suggested -- calculating a single z-score for the final result. Order doesn't change, but now the numbers are something I can make sense of. I also calculated percentiles, and those are very helpful for contextualizing a player's DCS.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Jul 27, 2007 4:43 pm Post subject: Reply with quote
kjb wrote:
JNichols42887 wrote:
supersub15 wrote:
2 comments:

1. Why are the slow-paced teams' players all at the top? Houston and San Antonio have 9 players in the top 25. Tracy McGrady?
Shouldn't this metric be pace-adjusted, somehow?

2. Wouldn't it be equally beneficial to have an OCS (offensive composite score)? This way, it is possible to figure out who are the better all-around players, check out differential leaders (i.e. Bowen would probably be in the bottom 25 in OCS).


It should be pace-adjusted, considering that every stat that was inputted was pace-adjusted (+/-, defensive rating, counterpart PER, percentage of playing time). I assume it's just because Houston and San Antonio happen to be great defensive teams.

I've thought about doing offensive composite score, but I'm not sure how much I could improve on the offensive stuff that's already out there.


If you have the time, I'd love to see an OCS. That might be interesting.

I took your numbers and did what Gabe suggested -- calculating a single z-score for the final result. Order doesn't change, but now the numbers are something I can make sense of. I also calculated percentiles, and those are very helpful for contextualizing a player's DCS.


Interesting. Could you e-mail those to me?

jn42887@bellsouth.net

Thanks.
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC

PostPosted: Fri Jul 27, 2007 5:50 pm Post subject: Reply with quote
You've got mail.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Sat Jul 28, 2007 10:39 am Post subject: Reply with quote
JNichols42887
Quote:
That would be really nice, although I'm not sure how you would do that...


I'm not sure either, but I think if you know the average PER of counterparts, and then know how these counterparts perform against (Dirk & Josh); if Josh guards tougher opponents (higher PER), and his opponent's PER differential (ratio) is higher than Dirk's, he's a better defender, and maybe that can reduce the noise produced by his offensive inside his own PER. But maybe this is more an assignment for 82games than you.

page 7

Author Message
HoopStudies



Joined: 30 Dec 2004
Posts: 705
Location: Near Philadelphia, PA

PostPosted: Sat Jul 28, 2007 11:37 am Post subject: Reply with quote
Harold Almonte wrote:
JNichols42887
Quote:
That would be really nice, although I'm not sure how you would do that...


I'm not sure either, but I think if you know the average PER of counterparts, and then know how these counterparts perform against (Dirk & Josh); if Josh guards tougher opponents (higher PER), and his opponent's PER differential (ratio) is higher than Dirk's, he's a better defender, and maybe that can reduce the noise produced by his offensive inside his own PER. But maybe this is more an assignment for 82games than you.


Note that Doug Steele at dougstats.com has his Defensive Tendex rating which does the analysis you're talking about -- comparing matched up player's tendex (just offensive parts) against their average and giving credit to the defender for the difference. He's been doing it for many years, so you can see Top 10 (or 20 sometimes) results through time...
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.
Back to top
View user's profile Send private message Visit poster's website
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Sat Jul 28, 2007 11:50 am Post subject: Reply with quote
Thanks for remembering that kind of counterparting exists. BTW Nowitzky is still ahead than Howard here.
Back to top
View user's profile Send private message
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Sat Jul 28, 2007 2:59 pm Post subject: Reply with quote
Anybody have a league average PER for each position for last season? Ideally it would be weighted by minutes played and adjusted for tweeners (perhaps according to 82games position time record). I get impression that PG and PF are highest, wings low. Center I see mixed signals and don't know if it is 3rd to 5th. Might be 3rd or 4th for starters but probably 5th for all minutes.

Hoopsstats.com matchup logs might be a way to identify a starter's matchups and see the game results and then compile counterpart PER from another source (HS has game and season "nba efficiency" scores for all, if you were willing to settle for that). Unless Doug Steele has all that for everybody already and was willing to share it. Defensive PER lists in public have always been short and unclear if it was computed for everybody or just likely suspects for the top.
Back to top
View user's profile Send private message
gabefarkas



Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC

PostPosted: Sat Jul 28, 2007 11:51 pm Post subject: Reply with quote
Mountain wrote:
Anybody have a league average PER for each position for last season?


It's 15 for the whole league. I would imagine you're not seeing more than a 1 unit swing in either direction.
Back to top
View user's profile Send private message Send e-mail AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Sun Jul 29, 2007 1:49 am Post subject: Reply with quote
Using the team by position pages at 82 games here is what I computed for average league own PER (all minutes of all teams last season according to their position assignment):

PG 16.57
SG 15.52
SF 16.26
PF 16.45
C 17.28

These numbers all seem a little high but I know there is some variation in PER numbers around the net.
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Wed Aug 01, 2007 12:16 pm Post subject: Reply with quote
Revised article is up:

http://www.82games.com/nichols2.htm
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Wed Aug 01, 2007 12:41 pm Post subject: Reply with quote
You have run the revision process well. As things stand right now are the 3 criteria equally weighted or if not what are the weights? Have you or will you test different weights? I believe counterpart might deserve a heavier weight than the others.
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Wed Aug 01, 2007 1:07 pm Post subject: Reply with quote
They're all weighted equally at the moment.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Wed Aug 01, 2007 6:52 pm Post subject: Reply with quote
Looking at the revised article's top 10 on DCS, only 2 make the top 25 on counterpart PER- Yao and Ginobili. A good number of the DCS 10 fall just outside that top 25 on counterpart PER. Maybe that says something about the balance between covering your man and being a team defender. And maybe the % starter is a factor too. Maybe it also fuels the weighting issue.
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Wed Aug 01, 2007 7:33 pm Post subject: Reply with quote
I think this old composite http://82games.com/comm26.htm had solved the weighting between (drtg. alone) and (plus-minus/on-off drtg.), and had a better solution for counterpart, given that counterpart'sPER includes some counterpart's defense.
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Wed Aug 01, 2007 9:19 pm Post subject: Reply with quote
Harold Almonte wrote:
I think this old composite http://82games.com/comm26.htm had solved the weighting between (drtg. alone) and (plus-minus/on-off drtg.), and had a better solution for counterpart, given that counterpart'sPER includes some counterpart's defense.


Wow, I had never seen that article before. I'm a little confused as to how they came up with the ratings numbers. Also, did they ever follow up on that?
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Wed Aug 01, 2007 10:27 pm Post subject: Reply with quote
Looking further at the data I see 2 extreme cases of weak counterpart PER numbers but good DCS I'll briefly mention. Chuck Hayes in Houston system has a DCS at almost 97th percentile but is in weakest 50 out of 327 on counterpart PER. I am not entirely comfortable calling him a great defender. Francisco Elson with Spurs at almost 79th percentile DCS but in bottom 20 on counterpart.

Jason Collins is about 40th weakest on counterpart PER but at 89th percentile on DCS and other studies vouch for him, so it can be argued. I am not sure about Hayes and Elson though.

Hayes team on/off is huge and how much it is because he is good and how much because Juwan Howard isn't I can't say. Elson has no team on/off impact but I guess on the Spurs that is saying something positive. But he is in a good situation.

I guess you could substitute team defensive ratings and see size of impact of it. By proxy, Chris Wilcox is pretty close to Elson on other two measures but drastically different team defensive rating puts him at 31st percentile. There probably is some difference in help value which impacts team results but that alone isn't involved.

Renaldo Balkman is better than Chuck Hayes on the other 2 measures but because of moderately different team defensive rating he is 10 percentile lower on DCS. Straight up who is the better defender?

I guess good defenders on the worst teams are hurt most and vice versa for not really special defenders on the best defensive teams and that is known and hard to remove entirely without losing team defense perspective. Changing component weights some to try to reduce that effect is an available judgment call.
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Thu Aug 02, 2007 8:02 am Post subject: Reply with quote
Perhaps on/off metrics must be factorized by the defensive strenght of the whole team, and replacement ( bad teammates hurt, also good replacements), and receive no more weight than an assist or shared stat (1/3), and give (2/3) to the counterpart. But, those adjusted plus/minus metrics aren't easily available, are they?. An adjusted metric might then deserves more weight.
Back to top
View user's profile Send private message
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Thu Aug 02, 2007 11:51 am Post subject: Reply with quote
I agree Harold that adusted defensive team +/- on/off would be welcome.

Adjusted defensive team +/- covers team and counterpart impact adjusted for players on the court (which is basically what DCS is going after) and in a natural balance. Really if you had that then DCS based of these 3 components wouldnt be as necessary.

David Lewin, or others, is there a plan for and an estimated time of arrival for 06-07 adjusted stats and is an offensive/defensive split possible or likely?

Last edited by Mountain on Thu Aug 02, 2007 6:44 pm; edited 2 times in total
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Aug 02, 2007 3:51 pm Post subject: Reply with quote
I stupidly forgot to do something: adjust defensive rating for position. After doing that, Pryzbilla drops to 4.


page 8

Author Message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Thu Aug 02, 2007 4:39 pm Post subject: Reply with quote
It was what Statman suggested some post ago. How did you do? Comparing counterpart's PER against the average PER of the position? Can this +/-PER compares oranges (perimeter) with apples (insiders)? Some split among starters, second unit, and deeper bench position PER?

Or you just positioned DRTG?
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Thu Aug 02, 2007 5:30 pm Post subject: Reply with quote
DanR weighted his defensive composite http://www.82games.com/rosenbaum3.htm 60% to +/-,on/off, and 40% +/-stats. He said he used to give more weight to stats before.
RRating weights his overall composite something near to 70% to counterpart and 30% to on/off. He used 50-50 before.
Back to top
View user's profile Send private message
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Aug 02, 2007 6:26 pm Post subject: Reply with quote
Harold Almonte wrote:
It was what Statman suggested some post ago. How did you do? Comparing counterpart's PER against the average PER of the position? Can this +/-PER compares oranges (perimeter) with apples (insiders)? Some split among starters, second unit, and deeper bench position PER?

Or you just positioned DRTG?


DRtg and counterpart PER.
Back to top
View user's profile Send private message AIM Address
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Aug 03, 2007 1:41 pm Post subject: Reply with quote
I just calculated DCS scores for the 2006 season and they (thankfully) are pretty similar to 2007. For example, with Memphis in 2006, Battier had a DCS of 44.74. With Houston in 2007, his score was 44.64. Obviously that was one of the closer ones, and there were some players with pretty big jumps or declines (usually those players changed teams and did not get a ton of minutes).
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 865
Location: Washington, DC

PostPosted: Fri Aug 03, 2007 3:47 pm Post subject: Reply with quote
I'd be interested in seeing the 2006 results, but even moreso, the 2005 (if you're up to it). That's because I tracked the Wizards pretty extensively in 2005 and I'm interested to see how DCS compares.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Aug 10, 2007 2:06 pm Post subject: Reply with quote
kjb wrote:
I'd be interested in seeing the 2006 results, but even moreso, the 2005 (if you're up to it). That's because I tracked the Wizards pretty extensively in 2005 and I'm interested to see how DCS compares.


2005 will take a little longer because I think I'll have to calculate the counterpart PER myself. It appears that it's only listed at 82games by position.

Quote:
Player Team DCS Adjusted
smith,josh atl 3.67
pachulia,zaza atl -6.96
ivey,royal atl -10.57
harrington,al atl -22.39
johnson,joe atl -23.99
williams,marvin atl -24.02
stoudamire,salim atl -25.35
childress,josh atl -35.84
lue,tyronn atl -36.60
olowokandi,micha bos 13.87
perkins,kendrick bos 9.92
allen,tony bos 9.75
west,delonte bos 8.46
scalabrine,brian bos 0.53
gomes,ryan bos -0.93
greene,orien bos -1.64
jefferson,al bos -6.80
pierce,paul bos -10.97
lafrentz,raef bos -11.98
szczerbiak,wally bos -14.27
okafor,emeka cha 36.21
wallace,gerald cha 25.82
felton,raymond cha 8.28
knight,brevin cha 0.32
carroll,matt cha -0.79
ely,melvin cha -0.86
rush,kareem cha -3.11
robinson,bernard cha -4.93
anderson,alan cha -8.43
voskuhl,jake cha -10.86
jones,jumaine cha -22.32
brezec,primoz cha -31.00
deng,luol chi 33.33
chandler,tyson chi 22.87
duhon,chris chi 19.76
allen,malik chi 19.62
hinrich,kirk chi 18.14
harrington,othel chi 7.79
sweetney,mike chi 7.26
gordon,ben chi 6.67
pargo,jannero chi 2.21
nocioni,andres chi 1.91
songaila,darius chi -3.29
varejao,anderson cle 32.88
ilgauskas,zydrun cle 31.18
snow,eric cle 18.37
gooden,drew cle 16.58
james,lebron cle 16.05
pavlovic,sasha cle 2.49
murray,ronald cle 1.24
hughes,larry cle -14.94
marshall,donyell cle -17.04
henderson,alan cle -25.08
jones,damon cle -28.76
griffin,adrian dal 29.05
howard,josh dal 24.38
diop,desagana dal 22.96
daniels,marquis dal 15.96
terry,jason dal 11.50
nowitzki,dirk dal 6.69
dampier,erick dal 6.41
harris,devin dal 3.50
stackhouse,jerry dal -1.09
vanhorn,keith dal -26.51
armstrong,darrel dal -33.01
camby,marcus den 49.99
elson,francisco den 26.28
martin,kenyon den 16.42
anthony,carmelo den 6.28
miller,andre den 4.83
johnson,dermarr den -0.26
buckner,greg den -3.15
najera,eduardo den -11.38
patterson,ruben den -17.44
evans,reggie den -23.63
boykins,earl den -24.23
kleiza,linas den -25.49
wallace,ben det 67.42
prince,tayshaun det 47.02
wallace,rasheed det 43.67
billups,chauncey det 32.36
hamilton,richard det 13.25
mcdyess,antonio det -23.44
evans,maurice det -24.28
delfino,carlos det -48.99
foyle,adonal gsw 45.51
fisher,derek gsw 12.17
dunleavy,mike gsw 12.06
richardson,jason gsw 10.98
davis,baron gsw 3.06
murphy,troy gsw -12.18
pietrus,mickael gsw -12.67
ellis,monta gsw -14.11
biedrins,andris gsw -19.87
diogu,ike gsw -21.78
cabarkapa,zarko gsw -48.97
ming,yao hou 45.17
hayes,chuck hou 35.20
mcgrady,tracy hou 22.60
mutombo,dikembe hou 12.99
alston,rafer hou 11.72
bogans,keith hou 10.91
head,luther hou 9.73
swift,stromile hou 4.82
bowen,ryan hou 4.66
howard,juwan hou 3.92
wesley,david hou -2.15
o'neal,jermaine ind 36.25
tinsley,jamaal ind 20.16
johnson,anthony ind 17.24
pollard,scot ind 14.30
foster,jeff ind 12.23
jackson,stephen ind 10.96
jasikevicius,sar ind 10.34
stojakovic,peja ind 10.29
croshere,austin ind 9.58
jones,fred ind 6.49
granger,danny ind 5.89
harrison,david ind 0.29
brand,elton lac 34.19
mobley,cuttino lac 25.30
ross,quinton lac 21.97
kaman,chris lac 17.95
cassell,sam lac 10.98
maggette,corey lac 4.41
livingston,shaun lac 3.69
ewing,daniel lac -18.82
radmanovic,vladi lac -19.26
singleton,james lac -23.14
odom,lamar lal 28.37
jackson,jim lal 8.66
brown,kwame lal 7.61
george,devean lal 4.36
vujacic,sasha lal 3.25
mihm,chris lal 1.82
walton,luke lal 0.56
parker,smush lal -0.10
bryant,kobe lal -1.69
cook,brian lal -5.33
battier,shane mem 44.74
wright,lorenzen mem 33.76
stoudamire,damon mem 32.37
jones,eddie mem 32.05
gasol,pau mem 27.76
miller,mike mem 12.06
tsakalidis,jake mem 11.85
jackson,bobby mem 10.56
burks,antonio mem 5.77
jones,dahntay mem -4.42
atkins,chucky mem -18.27
warrick,hakim mem -40.50
mourning,alonzo mia 53.76
o'neal,shaquille mia 27.87
wade,dwyane mia 25.75
williams,jason mia 20.18
haslem,udonis mia 19.84
kapono,jason mia 15.95
payton,gary mia 2.07
walker,antoine mia 1.39
posey,james mia 0.12
anderson,derek mia -24.38
anderson,shandon mia -24.93
bogut,andrew mil 21.98
magloire,jamaal mil 19.09
redd,michael mil 14.31
simmons,bobby mil 10.07
ford,t.j. mil 9.62
bell,charlie mil -11.59
williams,maurice mil -20.12
welsch,jiri mil -25.41
smith,joe mil -33.14
gadzuric,dan mil -35.39
kukoc,toni mil -36.48
garnett,kevin min 40.71
jaric,marko min 37.35
madsen,mark min 15.97
griffin,eddie min 7.84
blount,mark min -0.43
hassell,trenton min -4.76
hudson,troy min -6.81
mccants,rashad min -8.60
davis,ricky min -12.99
banks,marcus min -17.61
reed,justin min -17.93
carter,anthony min -22.01
carter,vince njn 46.22
collins,jason njn 45.27
kidd,jason njn 29.10
jefferson,richar njn 28.58
krstic,nenad njn 24.61
planinic,zoran njn 7.38
nachbar,bostjan njn 2.04
murray,lamond njn -0.21
robinson,cliffor njn -1.96
vaughn,jacque njn -8.90
padgett,scott njn -23.53
paul,chris nor 22.14
andersen,chris nor 20.56
snyder,kirk nor 8.58
claxton,speedy nor 6.68
brown,p.j. nor 5.81
mason,desmond nor 0.83
smith,j.r. nor -2.67
west,david nor -7.24
butler,rasual nor -14.09
williams,aaron nor -22.11
johnson,linton nor -45.11
jackson,marc nor -55.80
robinson,nate nyk -4.36
woods,qyntel nyk -4.81
frye,channing nyk -5.10
marbury,stephon nyk -12.69
curry,eddy nyk -13.95
butler,jackie nyk -14.64
richardson,quent nyk -20.55
crawford,jamal nyk -20.57
rose,malik nyk -21.36
lee,david nyk -22.07
francis,steve nyk -22.86
taylor,maurice nyk -31.84
rose,jalen nyk -36.63
milicic,darko orl 34.15
arroyo,carlos orl 29.38
ariza,trevor orl 15.60
howard,dwight orl 10.11
hill,grant orl 2.22
battie,tony orl -3.09
nelson,jameer orl -3.65
dooling,keyon orl -4.23
turkoglu,hedo orl -5.19
stevenson,deshaw orl -20.18
garrity,pat orl -30.75
dalembert,samuel phi 17.99
iguodala,andre phi 15.59
iverson,allen phi 2.18
hunter,steven phi -5.08
webber,chris phi -16.01
korver,kyle phi -17.02
ollie,kevin phi -22.81
salmons,john phi -25.24
barnes,matt phi -30.94
thomas,kurt pho 40.44
diaw,boris pho 27.67
marion,shawn pho 23.95
bell,raja pho 12.98
nash,steve pho 12.35
house,eddie pho 0.85
jones,james pho -18.72
barbosa,leandro pho -38.90
thomas,tim pho -49.47
przybilla,joel por 15.19
ratliff,theo por 2.86
miles,darius por -2.06
skinner,brian por -9.53
randolph,zach por -13.62
telfair,sebastia por -19.77
jack,jarrett por -29.25
blake,steve por -37.64
outlaw,travis por -37.70
dixon,juan por -39.27
webster,martell por -43.46
khryapa,viktor por -45.62
artest,ron sac 29.26
garcia,francisco sac 23.15
thomas,kenny sac 13.31
wells,bonzi sac 12.41
martin,kevin sac 7.61
abdur-rahim,shar sac 4.82
miller,brad sac 3.61
hart,jason sac 1.27
bibby,mike sac -15.67
duncan,tim san 61.46
ginobili,manu san 44.92
parker,tony san 41.99
nesterovic,rasho san 41.65
barry,brent san 41.20
bowen,bruce san 33.77
horry,robert san 24.17
finley,michael san 20.09
vanexel,nick san -9.79
mohammed,nazr san -26.09
udrih,beno san -30.23
watson,earl sea 10.67
allen,ray sea -15.53
collison,nick sea -20.81
swift,robert sea -23.21
wilkins,damien sea -27.14
petro,johan sea -35.09
wilcox,chris sea -35.36
lewis,rashard sea -43.89
ridnour,luke sea -46.66
moore,mikki sea -57.75
bosh,chris tor 7.53
araujo,rafael tor -7.75
sow,pape tor -19.92
james,mike tor -21.26
peterson,morris tor -21.50
davis,antonio tor -24.92
villanueva,charl tor -25.52
calderon,jose tor -30.58
graham,joey tor -36.74
bonner,matt tor -43.66
kirilenko,andrei uta 47.73
collins,jarron uta 27.41
giricek,gordan uta 26.80
palacio,milt uta 4.74
okur,mehmet uta 3.22
ostertag,greg uta -3.26
mcleod,keith uta -3.45
humphries,kris uta -10.32
harpring,matt uta -12.45
brown,devin uta -16.92
williams,deron uta -24.26
boozer,carlos uta -39.63
haywood,brendan was 29.37
jeffries,jared was 20.81
jamison,antawn was 8.74
butler,caron was 2.14
hayes,jarvis was -7.57
thomas,etan was -11.63
daniels,antonio was -13.51
arenas,gilbert was -27.15
ruffin,michael was -31.83
Back to top
View user's profile Send private message AIM Address
Carlos



Joined: 21 Jan 2005
Posts: 64
Location: Montevideo, Uruguay

PostPosted: Fri Aug 10, 2007 3:12 pm Post subject: Reply with quote
Very interesting. All Memphis players's ratings took a dive in 06-07, of course. Gasol went from 27.76 in 06 to -12.2 in 07, for example. It seems to me that there is a strong "system effect" at play. That is, being a member of a very good (or very bad) defensive team makes you look better (or worse) than you really are. I mean, Memphis's defense went from good in 06 to terrible in 07 and obviously SOMEBODY defended worse in 07, but the sudden change in a player from "very good defender" to "below average" suggests that there is something more here than just variation in individual ability.
Back to top
View user's profile Send private message
bchaikin



Joined: 27 Jan 2005
Posts: 690
Location: cleveland, ohio

PostPosted: Fri Aug 10, 2007 6:09 pm Post subject: Reply with quote
It seems to me that there is a strong "system effect" at play.

memphis went from the 2nd best/lowest points allowed per defensive team possession in 05-06 to the highest/worst in 06-07 among the 30 teams in the league. however in dropping from 49 wins in 05-06 to just 22 wins in 06-07, their points scored per offensive team position actually got better...

3 of the 5 players to play the most minutes for the grizzlies in 05-06 were shane battier, an excellent defender, and eddie jones and bobby jackson, two very good defenders. none of the three played for the team in 06-07. when you lose not one or two but three of your best defenders among your top five minutes played players, your defense will most likely get worse, regardless of the coach or system...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Aug 10, 2007 7:15 pm Post subject: Reply with quote
I agree. Right now I think there are two flaws with the system: It is highly influenced by what team you're on (as evidenced by Ginobili looking so good every year), and it doesn't account for matchups. I may be able to fix the first problem though. I'm just not sure if I want to punish bad players on good teams and vice versa. For example, some of Gasol's decline surely was a result of the Grizzlies losing some good defenders and becoming terrible defensively. But I also think that individually he probably played worse too. If everyone around you is bad on defense and doesn't care, you're probably not going to put forth your best effort either.
Back to top
View user's profile Send private message AIM Address
Carlos



Joined: 21 Jan 2005
Posts: 64
Location: Montevideo, Uruguay

PostPosted: Sat Aug 11, 2007 10:34 am Post subject: Reply with quote
Or maybe playing in a very good defensive team in 06 made him look better than he really was.. The thing is, I'm not sure how to (or if it's really possible to) separate individual defensive ability from team system and teammates's ability. Playing with good defenders in a good system means your guy receives worse entry passes, you don't have to help as much, defensive help arrives faster, there are fewer targets for the opponent to pass to, etc. These things don't really show in counterpart production and I'm unsure of how to account for them.
Back to top
View user's profile Send private message
KeeneKaufmanWheeler



Joined: 02 Aug 2006
Posts: 72


PostPosted: Wed Aug 15, 2007 9:57 pm Post subject: Reply with quote
I've found this research to be very interesting, and I've tinkered with the data a little to try and make the data easier to compare across teams.

I used my own ranking system which I like to extrapolate across whole teams (to create a "team offense ranking") and opponents (to give a "Team Defense" ranking). This gives me a relative measure to compare how much better one team is than another.

I then applied these percentages to DCS to come up with a Team DCS. This gave me a starting point to try and factor in team performance into individual ranking.

First, I decided to use the original -- version one -- DCS rankings. I felt that there were several players that jumped out at me as being way off in version two (notably Earl Boykins being considered an average defender).

I then filtered the data twice: once to factor out team-to-team differences and another to factor the individual’s influence on team performance.

For the team weighting, I added/subtracted half of the percentage difference that each team deviated from the norm (half of the difference because I theorized that since defense is part ability and part coaching style, I wanted one filter to be coaching related and one player ability related). For the individual weighting I calculated the percentage of minutes each player had played and factored in the value of the player vs. that of an average replacement-level player (a 50 DCS player).

I feel that the results look pretty accurate. They are definitely best when comparing players to their contemporaries at the same position – it seems like point guards in particular get knocked down significantly. Maybe they tend to have more offensive responsibilities and tougher nightly defensive matchups than other players?

The results (top 5 at each position):


OVERALL RANK ADJUSTED DCS
2 104.241 okafor,emeka cha C
3 104.058 duncan,tim sa C
4 102.805 biedrins,andris gs C
5 102.676 o'neal,jermaine ind C
24 96.242 nesterovic,rasho tor C
1 105.754 garnett,kevin min PF
6 101.591 marion,shawn pho PF
11 99.824 varejao,anderson cle PF
12 99.798 thomas,tyrus chi PF
13 99.631 millsap,paul uta PF
7 101.490 balkman,renaldo ny SF
15 98.921 battier,shane hou SF
18 98.171 james,lebron cle SF
23 96.592 ariza,trevor orl SF
25 95.292 artest,ron sac SF
9 101.263 ginobili,manu sa SG
29 94.598 wells,bonzi hou SG
37 91.200 sefolosha,thabo chi SG
41 90.734 wade,dwyane mia SG
43 90.026 mcgrady,tracy hou SG
8 101.339 davis,baron gs PG
10 100.938 rondo,rajon bos PG
22 96.740 alston,rafer hou PG
33 92.901 harris,devin dal PG
46 88.764 ewing,daniel lac PG

And the bottom at each position:

236 26.334 voskuhl,jake cha C
274 13.188 blount,mark min C
297 6.771 brezec,primoz cha C
320 -0.850 okur,mehmet uta C
326 -5.105 curry,eddy ny C
288 9.470 randolph,zach por PF
293 7.330 diaw,boris pho PF
294 7.286 robinson,cliffor nj PF
315 -0.071 warrick,hakim mem PF
319 -0.670 gomes,ryan bos PF
298 6.496 kapono,jason mia SF
309 1.480 williams,shawne ind SF
317 -0.393 hayes,jarvis was SF
318 -0.585 szczerbiak,wally bos SF
323 -3.116 morrison,adam cha SF
307 3.268 jones,dahntay mem SG
310 1.257 carroll,matt cha SG
312 0.692 green,willie phi SG
313 0.376 hassell,trenton min SG
316 -0.183 crawford,jamal ny SG
321 -0.966 atkins,chucky mem PG
322 -1.232 mcinnis,jeff cha PG
324 -3.749 lue,tyronn atl PG
325 -4.944 williams,deron uta PG
327 -5.987 marbury,stephon ny PG

The full data is posted at http://briefcase.yahoo.com/ipmhoops

Oh, and one difficulty with even attempting this analysis: the fact that charges taken is not a stat.

I think the results do a pretty good job of sorting the players relative to one another. It's tough to filter out the team efffect when a big part of defense is coaching style and a big part of coaching style is player composition. So it doesn't really surprise me that the best defenders seem somewhat concentrated on the best defending teams.

Maybe Trenton Hassell just wasn’t all that great a defender this year – players like Bruce Bowen and Artest who also take on the best opponent every night don’t get similarly hammered.

JNichols, thanks for your efforts on this. I think it’s very illuminating.
Back to top
View user's profile Send private message Visit poster's website
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Aug 16, 2007 1:29 am Post subject: Reply with quote
That's very interesting stuff. How did you calculate team DCS? It seems the hardest decision is how much of a team factor do you associate to individual player ability. You chose to do half, which is probably more realistic than the way I did it (fully determined by player ability).

Also, you chose the first DCS system released in the first 82games article. I personally like the second one a lot better. Sure there are some players that are a bit off, but it helped so many guys like Tayshaun Prince that I think it did more good than harm. But hey, that decision in the end just comes down to our personal opinions and biases, so I guess there isn't a right way to do it.

I'm glad you're taking my data and running with it. I take that as a huge compliment.

-Jon
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 865
Location: Washington, DC

PostPosted: Thu Aug 16, 2007 10:55 am Post subject: Reply with quote
Jon: Finally got around to looking at your 05-06 results. Interesting stuff. Out of curiosity, what do you make of Jamison's "better than average" rating?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Aug 16, 2007 11:31 am Post subject: Reply with quote
kjb wrote:
Jon: Finally got around to looking at your 05-06 results. Interesting stuff. Out of curiosity, what do you make of Jamison's "better than average" rating?


To be honest, a combination of him having an exaggerated reputation of being a bad defender and the fact that maybe this stat rewards starters too much at the moment (I'm currently working to fix that).
Back to top
View user's profile Send private message AIM Address
kjb



Joined: 03 Jan 2005
Posts: 865
Location: Washington, DC

PostPosted: Thu Aug 16, 2007 12:47 pm Post subject: Reply with quote
Another factor that might be getting picked up in your numbers is that Jamison had the best rebounding season since his rookie year in 05-06. It was his career high in defensive rebounds per 40 minutes. And that was for a bad defensive rebounding team, so those boards were fairly important.

page 9 of 9

Author Message
KeeneKaufmanWheeler



Joined: 02 Aug 2006
Posts: 72


PostPosted: Thu Aug 16, 2007 7:35 pm Post subject: Reply with quote
JNichols42887 wrote:
How did you calculate team DCS?


My ranking system (which I call IPM) is similar to PER. I have a system to predict W/L records by using teams' IPM scores for both their own performance and opponent performance (the calcluation is similar to exponential winning percentage, with IPM taking the place of points scored and opponent IPM being points allowed).

The Opponent IPM is not truly a defense stat (because it includes defensive components like blocked shots and steals that the opposition racks up) but it gives a good idea of a teams' defensive performance. Compared between teams, I get relative IPM scores. I just assigned the average IPM score a DCS value of 50 and calculated each team's percentage difference from that.

A spreadsheet laying out Team IPM performance is also available at the link I posted above.

My next goal is to convert player DCS scores into IPM ... because theoretically that would indicate who's offense makes up for his defense, and vice versa.

How much better (percentage-wise) is the best defender than the worst, do you think? It can't be 100%, because that would mean that anyone off the street could defend better than Stephon Marbury.
Back to top
View user's profile Send private message Visit poster's website
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Aug 16, 2007 11:30 pm Post subject: Reply with quote
KeeneKaufmanWheeler wrote:
JNichols42887 wrote:
How did you calculate team DCS?


My ranking system (which I call IPM) is similar to PER. I have a system to predict W/L records by using teams' IPM scores for both their own performance and opponent performance (the calcluation is similar to exponential winning percentage, with IPM taking the place of points scored and opponent IPM being points allowed).

The Opponent IPM is not truly a defense stat (because it includes defensive components like blocked shots and steals that the opposition racks up) but it gives a good idea of a teams' defensive performance. Compared between teams, I get relative IPM scores. I just assigned the average IPM score a DCS value of 50 and calculated each team's percentage difference from that.

A spreadsheet laying out Team IPM performance is also available at the link I posted above.

My next goal is to convert player DCS scores into IPM ... because theoretically that would indicate who's offense makes up for his defense, and vice versa.

How much better (percentage-wise) is the best defender than the worst, do you think? It can't be 100%, because that would mean that anyone off the street could defend better than Stephon Marbury.


Tough question. I don't think I'd be able to put a number to that.
Back to top
View user's profile Send private message AIM Address
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Fri Aug 17, 2007 12:27 am Post subject: Reply with quote
Kevin,

Washington's 2005 DCS numbers:

Quote:
Player Team DCS Adjusted
haywood,brendan was 40.01385867
jeffries,jared was 21.64344459
hughes,larry was 8.876909722
jamison,antawn was 0.605211682
arenas,gilbert was -2.387186221
brown,kwame was -2.457737288
hayes,jarvis was -17.07807681
thomas,etan was -17.24999351
blake,steve was -28.68580766
ruffin,michael was -29.63585088
dixon,juan was -34.19932447
peeler,anthony was -61.53311776
Back to top
View user's profile Send private message AIM Address
thref23



Joined: 13 Aug 2007
Posts: 90


PostPosted: Wed Aug 22, 2007 5:09 am Post subject: Reply with quote
Hey guys,

I came across this thread not too long ago. Anyways, shortly after Jon posted his first article on 82games.com, I took his data and played with it myself, and made some additions to it in order to come up with my own defensive composite scores. This wouldn't have been possible without Jon's initial work, so Jon, you get credit there. [warning: long post]

I actually e-mailed him on the rankings a few weeks back, but have made some tweaks since then. I figured I'd share my results here along with a formulated explanation as to what I did and didn't do. If I can't come up with a definitive DCS formula myself, maybe I can provide input to others' formulas, or others could provide input which could be used to tweak mine.

What I did and didn't do initially:

1.) +/-. I changed slightly the way +/- rank was calculated. On each team, each player was assigned a number - the player ranked best on the team was given a 10, second best 9, etc, all the way down to one for those at the bottom of each team. I then multiplied this number by a number between 1.7 and 2 depending on the number of points a player's team allowed per game. I then additionally multiplied by a number between 1-1.3 depending on how many points a player's team scored per game (more points scored a certain ratio of points is likely allowed as a result). The highest ranking player scored 23.somethin points

The reason: a guy like Chuck Hayes was at the top of +/- rank previously because he was on the floor whenever Juwan Howard wasn't. Pryzbilla was ranked around the top too because he was a respectable defensive player in isolated low minute roles on a team with a bunch of poor defenders. My system doesn't solve the problem completely, but at least Chuck Hayes is ranked 6th, and Pryzbilla I think 35th, as random examples.

There's also the issue of a guy like Ben Gordon suffering in this category because the Bulls score a lot more when he is on the court. I didn't think that was worth trying to fix though - as a result, he also has a very impressive counterpart PER - because the guy he's matched up against has to exert so much effort defensively he's not able to do as much offensively.

This category could ideally be tweaked further.

2.) DRTG - simple, I used the data provided by Jon, and based on rank gave each player a score between 0-25. I couldn't say necessarily how this category could be tweaked because I can't find a great explanation as to exactly how its computed. Though, it is one established method of measuring defensive fortitude, and defensive rebounding does have an impact on D.

3.) Box Score Stats - I took the box score stats that Jon initially computed, and again, gave each player a score between 0-25. I disagree that they should have been removed from the formula, and feel they are important for three reasons: a.) Blocks and steals - especially when combined with foul efficiency, specifically do help measure a player's ability to provide effective help defense (counterpart PER helps to measure man on man, so this is fitting IMO), b.) I agreed with Jon's initial assertion that this category helped to indicate athleticism and defensive potential (this doesn't necessarily have to be a player's future potential, some players are capable of stepping up their defensive game when they feel it matters the most), c.) DRTG, while not being completely unreliable, does obviously favor big men because of its emphasis at times on defensive rebounding. The box score category inherently favors wings and guards for two reasons - a good guard in this category typically comes up with more steals than a good big comes up with blocks, and because big men are more likely to find themselves in foul trouble.

At the very least, I feel this stat needs to be weighed equally with DRTG, regardless of how much weight the two categories combined make up. IMO, Jon's initial formula didn't suffer because it included this category in addition to DRTG, but because the two categories were weighed too heavily without the addition of more criteria and/or another category.

4.) Counterpart PER - Based on rank I gave each player a score between 0-20. This was then multiplied by a number between 1 and 1.30 to account for how many points a player's team scored per game (on a team that scored more points per game, the opponent generally received more possessions per minute, positively affecting opponent PER). The top ranked player in this category scored around 24.5 points.

I have read the argument that certain players are often forced to matchup with the other team's top offensive players - and hence that affects this category. Though, other teams are also more likely to attempt to run offense through players not being covered by a team's best defender - and +/- should help to offset the affect as well (as in the case of Bruce Bowen).

5.) Minutes/Games Played - This category was initially added for fairly obvious reasons. I used the % of minutes played stat from 82games.com as I was able to copy and paste it in the same breath as copying and pasting counterpart PER data. I assigned a number between 1 and 1.2 for each player, and multiplied it by their existing combined score. The top 10% of players all received a 1.2, and the bottom 10% all received a 1, with varying scores in between.

This ended up accounting for about 16% of the final score. Ideally I would have broken it down to about 10% minutes per game and 6%
games played - but that would have been a lot of work, games played does help measure health, which does affect a player's ability to be consistent, and the results looked good enough anyway.

Jon, in his revised rankings, used a stat meant to measure how high a percentage of a player's minutes came against first string opponents. I thought this was a great idea. However, upon scrutiny, I disagree that it helps so much. Bench players play alongside other bench players which at times can negatively affect one or two categories, and also play against bench players. IMO it evens itself out.

What is/was needed, IMO, is something punishing a player for not playing many minutes - its easy for any player to come off the bench in specific types of roles based on specific matchups and be a statistically good defender. Minutes played awards starting caliber players for their ability to maintain their defensive presence consistently in versatile situations. v Plus minutes played helps indicate a coach's confidence level in a player's ability - a truly good defender will always find his place in a rotation.

6.) I then simply multiplied each player's score by a constant variable to make the top ranked player score about 100 points.

-----

The results looked pretty accurate to me. For the most part the formula eventually passed my own scrutiny in terms of its logic. I did feel there were minor tweaks to be made.

Recently, I decided to do three things to it:

1.) I calculated an estimation of each team's effectiveness/reliance when it came to help defense - I did this by adding steals, blocks, and a percentage of turnovers forced per game. I adjusted the counterpart PER ranking slightly, by around 20% or so, on the notion that the ranking held less weight for players on teams that relied so much on effective help defense, and more weight on teams that relied on everyone to cover their man. Also, this way players didn't benefit so much in this category by being on the floor at the same time as other good defenders.

2.) In virutally the same manner, I calculated team defense scores by adding a percentage of fouls committed and points allowed. The logic: foul efficiency is harder to come by for a player when he is on the floor with other poor defenders, and help defense is also more influential in that situation as well.

3.) I made sure that the top rated player in each category scored an even 25 points in the category.

The formula then made more sense to me logically, but I was unsure of the results (although only minor changes resulted).

--

The next attempt, I did everything the same, but based on criticism that there is too much overlap between DRTG and the boxscore category, I counted those categories as worth only 20 points each, while counting the other two categories at 30 points each. The results weren't perfect, but I liked them as much as my initial calculations.

---

Then, as a point of reference, I took the rankings used in the initial formula (un-adjusted PER/Box Score stats), and ran them with the same lowered emphasis on DRTG and Box Score stats.

---

After doing four separate calculations, no version of the formula was a clearcut winner. So, I decided to take the average of all four scores, and in affect incorporated each version of the formula into one new formula.

Though I still noticed two imperfections.

1.) PGs on bad defensive teams seemed to score rather low, whereas guys like Rafer Alston seemed a little higher than I would have expected. Plus, to begin with, I had thought about this, because a PG's opponent's PER is affected in large part by assists, the PG is affected most of all by his teamates' defensive abilities. For PGs only, I slightly awarded PGs on poor defensive teams, while slightly punishing PGs on good defensive teams. This was not done perfectly - i.e. a PG on a good defensive team with a lower counterpart PER was punished slightly more than one with a high opponent counterpart PER. But regardless, the end results seemed right, so I didn't try to tweak it farther (it would have been complicated).

2.) There were still several low minute players that I thought seemed to be overrated. After examining the list, I determined that players scoring below a 1.09 in the minutes/games played category seemed to be overrated. Accordingly, players below this mark were punished by a small and proportionate percentage of their overall score. After I did this, in certain respects, the results looked much more accurate to me.

In the end:

Players posted an average score of around 48. Cs averaged higher scores than PGs, but: a.) Cs are taller more physically capable players than Cs, b.) there were 30 more PGs (i.e. more 3rd stringers) on the list than Cs, and c.) Just as C is often the most important position defensively and PG is most important offensively. Tyrus Thomas still may seem to be overrated, but despite his inexperience and bench role, being the top rated player in the DRTG category puts him in the same class with a number of historically very solid players - so he may just be developing into a truly premier defender. Only truly noticeable downfall to me: guys that were traded midseason were negatively affected by weighting games played for their team.

TOP 25:

duncan,tim 99.73
garnett,kevin 97.69
ginobili,manu 94.93
o'neal,jermaine 91.04
deng,luol 88.28
battier,shane 88
marion,shawn 87.82
howard,josh 87.78
thomas,tyrus 85.89
hilario,nene 85.81
james,lebron 84.84
mcgrady,tracy 83.95
bowen,bruce 83.92
artest,ron 83.71
harris,devin 81.72
kirilenko,andrei 81.46
rondo,rajon 81.4
camby,marcus 80.94
okafor,emeka 80.91
balkman,renaldo 80.66
garbajosa,jorge 80.24
alston,rafer 80.13
pavlovic,sasha 79.31
davis,baron 79.28
ming,yao 79.13

BOTTOM 25:

roy,brandon 22.16
redick,j.j. 22.11
okur,mehmet 21.72
kapono,jason 21.61
korver,kyle 21.13
ivey,royal 20.64
hassell,trenton 20.43
morrison,adam 20.3
radmanovic,vladi 19.69
robinson,cliffor 19.27
payton,gary 19.23
mcinnis,jeff 18.61
telfair,sebastia 18.04
szczerbiak,wally 16.93
noel,david 15.7
james,mike 14.44
johnson,anthony 14.23
jones,dahntay 13.73
ilyasova,ersan 13.56
jackson,marc 13.47
hayes,jarvis 13.03
williams,shawne 11.98
brezec,primoz 10.93
warrick,hakim 9.96
lue,tyronn 5.32

RANDOM NOTABLES:

Jason Maxiell: 73.59
Monta Ellis: 62.17 (would be impressive for a 20 year old guard)
Andres Nocioni: 31.31 (the only player on the list, outside of Trenton Hassell and maybe Charlie Bell, I can find that is supposedly a good defender but is shown to be a poor defender based on last year's role and performance)
Joel Pryzbilla: 60.97
Antawn Jamison: 30.66

FULL SPREADSHEET:

http://www.mediafire.com/?4v5yrbyznm3

(column BF is the final ratings)

And I am done...
Back to top
View user's profile Send private message Send e-mail
kjb



Joined: 03 Jan 2005
Posts: 865
Location: Washington, DC

PostPosted: Wed Aug 22, 2007 9:20 am Post subject: Reply with quote
Can you email the spreadsheet to me at kevinbroom@realgm.com? The web filter at my job won't let me look at your spreadsheet online. Thanks.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
KeeneKaufmanWheeler



Joined: 02 Aug 2006
Posts: 72


PostPosted: Wed Aug 22, 2007 10:28 am Post subject: Reply with quote
I've tried to implement DCS into a defensive component of my player rankings, and I think I've managed to do so rather successfully.

I had already converted DCS into a relative measure when factoring in team influences on DCS rankings.

Next, in order to convert DCS to someting that equates to my player ranking system I had to estimate how much "better" the best defender is than the worst defender. This required a little leap of faith, but I figure that if a 50% shooter would get shut down by the best defender he would probably shoot about 35% but would kill the worst defender to the tune of about 65% shooting; then logically there should be about a 30% spread between the best and worst defenders. I had used the Version 1 DCS scores (1-100) and so by adding 200 to the DCS value I had about the 30% spread that I wanted.

I could then easily convert DCS to something compatible to my IPM score (where 1.0 is superstar level, .75 is average for a decent starter and roughly .55 is replacement level).

Once this was done it was simply a matter of adding IPM to DCS to come up with an offense+defense player ranking.

However, I had to weight the addition. Basketball is different from other games in that the rules are so heavily weighted toward offense that I felt that defensive prowess could not be given equal weight to offensive. If the best defender holds a 30 point scorer to 20 but scores zero himself then his impact on the game was still -20 points. He would have been better off allowing 35 but scoring 20.

I settled on weighting the combined score as 80% IPM and 20% DCS. I then added the results to come up with a combined IPM score.

The results were in line with what I wanted. It gave a boost to the rankings of defensive-minded players (such as Dalembert and Ben Wallace) while hammering one dimensional players like Eddy Curry. However, that did not massively change the rankings at the superstar level, moving players like Nowitzki and Gasol down a bit while giving a small boost to Marion.

While overall I really like the rankings, there are a couple of imperfections. Guard performance seems to be somewhat affected by the quality of their big men (which may explain the Rafer Alston phenomenom and TJ Ford's high defensive rating). Players like Shane Battier, despite getting a big boost in the rankings, still seem undervalued (although I am starting to doubt how important Bruce Bowen really is to the Spurs).

It does provide an interesting insight, though, at how Golden State has a bunch of really good individual defenders despite a poor overall defensive ranking.

Rankings available at http://briefcase.yahoo.com/ipmhoops in the file "IPM-DCS Combined Rankings"
Back to top
View user's profile Send private message Visit poster's website
Fighting Fitzpatrick



Joined: 21 Aug 2007
Posts: 1


PostPosted: Wed Aug 22, 2007 1:09 pm Post subject: I want to use DCS for my replays. What's the DCS formula? Reply with quote
Hi. I am doing some replays of the 2005-06 Milwaukee Bucks with my Replay Basketball tabletop board basketball game,and I would like to know the formula for DCS so I can rate my players with it. I have a feeling this stat will be useful for my replays. Can you help me out? Thanks!
_________________
I am a Fighting Fitzpatrick...Fight me I'm Irish and I am a Bucks fan...light it up!
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Thu Aug 23, 2007 11:39 am Post subject: Reply with quote
Unadjusted, it's just a combination of three z-scores: defensive rating (adjusted for position), counterpart PER (adjusted for position), and +/- (not adjusted for position). Add the three scores and multiply them by -10.
Back to top
View user's profile Send private message AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527


PostPosted: Sat Aug 25, 2007 11:47 am Post subject: Reply with quote
By Defensive Win Shares Ben Wallace was #2 in league. By DCS he is 29th. At his price, his defensive rank matters a lot. Is he the 2nd most impactful in the game or 5th best on his team? He slips on DCS because of counterpart score where he ranks 53rd in league. That seems to be getting closer to fair and full assessment to me, though I'd weight counterpart even heavier.

The adjustments made by KKW and the ref in their versions add to the discussion and I could support some of the motivations and approaches. Not sure if consensus is being sought or if enough voices will be expressed to help reach it. Several flavors is ok too for those very interested at least. A combined table ranking top 25 or 50 (to capture outliers from top 25 in some methods) by each of the methods now out there would be very helpful for considering player rank and for considering next steps with the concept, if any of the principals were willing to work it up.

Still hope for future arrival of adjusted defensive +/-. Counterpart and team would be in natural balance there and the unity and simplicity might make it easier to get broader audience viewing and acceptance.
Back to top
View user's profile Send private message
thref23



Joined: 13 Aug 2007
Posts: 90


PostPosted: Sun Aug 26, 2007 11:27 pm Post subject: Reply with quote
Mountain wrote:
By Defensive Win Shares Ben Wallace was #2 in league. By DCS he is 29th. At his price, his defensive rank matters a lot. Is he the 2nd most impactful in the game or 5th best on his team? He slips on DCS because of counterpart score where he ranks 53rd in league. That seems to be getting closer to fair and full assessment to me, though I'd weight counterpart even heavier.


IMO defensive win shares are pointless in certain respects to compare players across different teams. The Bulls may have had more defensive win shares than any other team last year - very significantly more than, say, Milwaukee. DWS is also impacted significantly by playing time.

I know that in my own customized version of DCS, Luol Deng is ranked #5, Ben Wallace was ranked #29. Even if DCS was adjusted per team and per minutes played and fulfilled its intended purpose, who's to say Deng was not more impactful in putting the Bulls in a position where other players could then put the team over the edge and win a game.

Does anybody think adjusted win shares should be a category in DCS? I'm not sure its necessary but could make some sense logically. I'll also point out that #29 - Ben's rank in my formula - is almost exactly midway in between 2 and 53.

I'd be curious to run the formula for previous season's to see where he would have stood. Prior to last year he was known as an elite defensive player, so any dropoff last year could have caused people to forget what he brings to the table. Same could be said about Kirilenko, who was as elite defender last year according to the counterpart PER category (Kirilenko is ranked #16 by my formula).
Back to top
View user's profile Send private message Send e-mail
Jon Nichols



Joined: 18 Aug 2005
Posts: 370


PostPosted: Sun Aug 26, 2007 11:53 pm Post subject: Reply with quote
thref23 wrote:
Mountain wrote:
By Defensive Win Shares Ben Wallace was #2 in league. By DCS he is 29th. At his price, his defensive rank matters a lot. Is he the 2nd most impactful in the game or 5th best on his team? He slips on DCS because of counterpart score where he ranks 53rd in league. That seems to be getting closer to fair and full assessment to me, though I'd weight counterpart even heavier.


IMO defensive win shares are pointless in certain respects to compare players across different teams. The Bulls may have had more defensive win shares than any other team last year - very significantly more than, say, Milwaukee. DWS is also impacted significantly by playing time.

I know that in my own customized version of DCS, Luol Deng is ranked #5, Ben Wallace was ranked #29. Even if DCS was adjusted per team and per minutes played and fulfilled its intended purpose, who's to say Deng was not more impactful in putting the Bulls in a position where other players could then put the team over the edge and win a game.

Does anybody think adjusted win shares should be a category in DCS? I'm not sure its necessary but could make some sense logically. I'll also point out that #29 - Ben's rank in my formula - is almost exactly midway in between 2 and 53.

I'd be curious to run the formula for previous season's to see where he would have stood. Prior to last year he was known as an elite defensive player, so any dropoff last year could have caused people to forget what he brings to the table. Same could be said about Kirilenko, who was as elite defender last year according to the counterpart PER category (Kirilenko is ranked #16 by my formula).


I'm currently working on the data for the last few years, and I'm going to try to put them all together on an easy-to-use web site in the future. You're right about Wallace and Kirilenko. In the years previous to 2007, they were at the very, very top of the DCS rankings.
Back to top
View user's profile Send private message AIM Address
supersub15



Joined: 21 Sep 2006
Posts: 273


PostPosted: Mon Aug 27, 2007 6:10 am Post subject: Reply with quote
Has anybody thought about adding charges drawn? This stat is available now through 82games.com and can influence the DCS. It is an important defensive tactic, even more than a block, which is accounted for.