Author Message
Crow
Joined: 20 Jan 2009
Posts: 816
PostPosted: Sun Jan 16, 2011 2:23 am Post subject: Comparing EZPM100 to Adjusted +/- Reply with quote
I pulled 29 top guys (from all positions but ended up with more PGs somewhat on purpose). I computed change in EZPM100 from last season to this season and made a short post with it at Evan's site under the Centers thread.
But then I also combined those figures in a weighted almost 1.5 season EZPM100 and decided to compare those values to the 1.5 season Adjusted +/- estimated at basketballvalue and share this here.
(There might be a slight difference in coverage of this season but this is just a starter exercise.)
The r2 for this small sample was .42 but one metric tries to measure everything and one in a developing metric that tries to measure most everything that can be pulled from the boxscore and play by play so differences are expected and worth looking at I think.
So here is the data:
Code:
....................1.5 yr
LAST TEAM EZ100 APM APM-EZPM
Rose CHI 3.4 1.1 -2.3
Felton CHA 3.1 -4.8 -7.9
WestbrookOKC 3.2 0.7 -2.5
Boozer UTA 2.9 1.1 -1.8
Paul NOH 9.0 7.5 -1.5
Carter ORL 3.6 5.3 1.7
Horford ATL 3.6 6.4 2.8
Pierce BOS 3.2 2.6 -0.6
Nelson ORL 3.2 1.7 -1.5
Lawson DEN 2.7 -1.9 -4.6
Anthony DEN 3.1 4.7 1.6
WilliamsUTA 5.3 7 1.7
Lowry HOU 2.9 4.6 1.7
Miller POR 4.1 5.7 1.6
Gasol LAL 4.8 3.4 -1.4
Nash PHX 6.1 10.1 4.0
Kidd DAL 3.7 4.5 0.8
GinobiliSAS 5.8 4.9 -0.9
Camby LAC 3.3 2.3 -1.0
Bosh TOR 3.3 3.8 0.5
Bryant LAL 4.4 3.8 -0.6
Bogut MIL 3.0 6 3.0
Rondo BOS 5.4 0.3 -5.1
Howard ORL 5.7 3.3 -2.4
Duncan SAS 4.9 5.5 0.6
Wade MIA 7.9 10.4 2.5
Billups DEN 4.2 -4.9 -9.1
Durant OKC 5.5 13.7 8.2
James CLE 10.0 15.1 5.1
For 7 guys the difference was 4 or more. That could be mostly noise or more likely noise and accounting of some notable uncounted impacts. 16 (or a bit more than half) were with 2 points.
A comparison with a RAPM metric might be tighter. A comparison with a later version of EZPM100 might make it tighter too. But the differences probably won't go away and may have real meaning in some or many cases. Attempts to understand those differences could use everything else.
Last edited by Crow on Sun Jan 16, 2011 1:37 pm; edited 3 times in total
Back to top
View user's profile Send private message
Mike G
Joined: 14 Jan 2005
Posts: 3597
Location: Hendersonville, NC
PostPosted: Sun Jan 16, 2011 9:33 am Post subject: Reply with quote
Here they are sorted by 'most favored by EZPM on the left,
and 'most favored by APM' on the right:
Code:
LAST TEAM EZ100 APM APM-EZPM TEAM EZ100 APM APM-EZPM
Billups DEN 4.2 -4.9 -9.1 Durant OKC 5.5 13.7 8.2
Felton CHA 3.1 -4.8 -7.9 James CLE 10.0 15.1 5.1
Rondo BOS 5.4 0.3 -5.1 Nash PHX 6.1 10.1 4.0
Lawson DEN 2.7 -1.9 -4.6 Bogut MIL 3.0 6.0 3.0
Westbrook OKC 3.2 0.7 -2.5 Horford ATL 3.6 6.4 2.8
Howard ORL 5.7 3.3 -2.4 Wade MIA 7.9 10.4 2.5
Rose CHI 3.4 1.1 -2.3 Carter ORL 3.6 5.3 1.7
Boozer UTA 2.9 1.1 -1.8 Williams UTA 5.3 7.0 1.7
Paul NOH 9.0 7.5 -1.5 Lowry HOU 2.9 4.6 1.7
Nelson ORL 3.2 1.7 -1.5 Anthony DEN 3.1 4.7 1.6
Gasol LAL 4.8 3.4 -1.4 Miller POR 4.1 5.7 1.6
Camby LAC 3.3 2.3 -1.0 Kidd DAL 3.7 4.5 0.8
Ginobili SAS 5.8 4.9 -0.9 Duncan SAS 4.9 5.5 0.6
Pierce BOS 3.2 2.6 -0.6 Bosh TOR 3.3 3.8 0.5
Bryant LAL 4.4 3.8 -0.6
At first glance it seems ezPM "likes" PGs, while APM favors offensive stars?
The average for this sample is -0.26; likely due to these being above-average players, and APM having a much wider range of values.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
back2newbelf
Joined: 21 Jun 2005
Posts: 274
PostPosted: Sun Jan 16, 2011 10:14 am Post subject: Reply with quote
Here's single season RAPM if someone wants to compare
http://stats-for-the-nba.appspot.com/rankings11
Back to top
View user's profile Send private message
EvanZ
Joined: 22 Nov 2010
Posts: 287
PostPosted: Sun Jan 16, 2011 10:30 am Post subject: Reply with quote
back2newbelf wrote:
Here's single season RAPM if someone wants to compare
http://stats-for-the-nba.appspot.com/rankings11
Thanks, that's useful. I am thinking it makes more sense right now to compare the offensive and defensive components separately, especially since I don't yet have counterpart defense in the model.
Also, I really appreciate Crow and others for looking into this kind of stuff. Very Happy
_________________
http://www.thecity2.com
http://www.ibb.gatech.edu/evan-zamir
Back to top
View user's profile Send private message
Crow
Joined: 20 Jan 2009
Posts: 816
PostPosted: Sun Jan 16, 2011 12:51 pm Post subject: Reply with quote
Thanks Mike for kindly sorting and making a clean table.
Thanks also to new2newbelf for the RAPM data. The partial single season RAPM file has Anthony as a tiny negative on offense. I think it is fair to say this is not a "normal" year for him. The distraction, but also maybe impacts from Billups' fall-off on Adjusted +/- despite the effort to try to isolate player impacts.
I agree with Evan that it would be to compare the offensive and defensive components separately. I didn't have an APM for the same timeperiod to match-up with EZPM100 to that. Will look at the partial season RAPM, though a 1.5 or 2 season version is still my preference.
I decided to compare the guys on the two sides of the table with negative and positive differential between Adjusted +/- and EZPM100 on usage, TS%, 3 pointers made and Assist %.
I found that there was hardly any difference between the negative and positive differential sides in average usage and 3 pointers made. TS% and Assist% were each actually a couple of points higher for the guys with negative differential between Adjusted +/- and EZPM100. So it doesn’t appear these things are a strong source of advantage as they are not overcoming the other stuff pulling them down.
The net differential between the two sides is over 5 points. There are big things out there affecting this. What? Maybe counterpart defense, the assignment of defensive credit between perimeter and interior, the rebounding credit, error in Adjusted +/- or need for adjustment in some other weight in EZPM100. Something or probably several things. It would take further work and probably a solid multiple regression on the the whole league and maybe for several seasons.