Biggest Stat/Adjusted +/- differences (Scott S., 2008)
Posted: Mon Apr 18, 2011 4:53 am
Author Message
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Thu Sep 18, 2008 4:32 pm Post subject: Biggest Stat/Adjusted +/- differences over since 2003 Reply with quote
Using the following formula (20+adjusted +/-)/(20+statistical+/-), after removing half of the best and worst statistical plus minus seasons and averaging the season totals since 2003, here are the highest and lowest ratios among those with adjusted +/- ratings each season.
Code:
Scalabrine, Brian 130%
Murray, Ronald 129%
Dooling, Keyon 126%
Allen, Malik 122%
Wallace, Rasheed 121%
Collins, Jarron 118%
Wilcox, Chris 118%
Jeffries, Jared 117%
Collins, Jason 117%
Watson, Earl 113%
Moore, Mikki 113%
Randolph, Zach 113%
Szczerbiak, Wally 112%
Prince, Tayshaun 112%
Pargo, Jannero 112%
Bowen, Bruce 111%
Miller, Brad 111%
Mutombo, Dikembe 111%
Hassell, Trenton 110%
Thomas, Kurt 110%
Radmanovic, Vladimir 110%
Arroyo, Carlos 110%
Duncan, Tim 109%
Pierce, Paul 109%
Jamison, Antawn 109%
Foster, Jeff 108%
Rose, Malik 108%
Brown, Devin 108%
Ming, Yao 107%
Garnett, Kevin 107%
Code:
Murphy, Troy 75%
Mohammed, Nazr 78%
Swift, Stromile 78%
Camby, Marcus 81%
Bibby, Mike 83%
Posey, James 85%
Marion, Shawn 85%
Arenas, Gilbert 86%
Hughes, Larry 86%
Billups, Chauncey 86%
Williams, Jason 87%
Gooden, Drew 88%
Curry, Eddy 88%
Salmons, John 89%
Jefferson, Richard 90%
Knight, Brevin 90%
McInnis, Jeff 91%
Brown, Kwame 91%
Hamilton, Richard 91%
Wallace, Ben 92%
Stoudamire, Damon 92%
Skinner, Brian 93%
Butler, Caron 93%
Daniels, Antonio 93%
O'Neal, Jermaine 94%
Boykins, Earl 94%
Redd, Michael 94%
Lue, Tyronn 94%
Dampier, Erick 94%
Boozer, Carlos 94%
What are some of the explanations we can come up with. Murray, Moore ad Devin Brown had low 2003 minutes and should be eliminated. Hamilton and, possibly Rasheed Wallace can still be attributed, at least in part to the 2006 adjusted +/- ratings. Since statistical +/- is not pace adjusted (at least the one I used), that can influence some. Some rankings can be assumed to be a result of the impact of man defense (Bowen and Redd). Some players on the positive end seem to fit the scrappy, role player with intangibles. These results don't seem completely random to me, but given the length of time we are looking at, I was expecting slightly more obvious tendencies and patterns on both spectrums.
Back to top
View user's profile Send private message Visit poster's website
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Thu Sep 18, 2008 5:00 pm Post subject: Reply with quote
Defense is obviously the biggie. Statistical +/- is in many ways superior to a BoP-based "total player rating" because SPM is largely team-independent and DRtg is so team-dependent. However, defense is obviously a big part of the game -- can we justify ignoring it altogether? In reviving Dan's old SPM formula and applying it to every season since 1971, I came up with 2 sets of data: standard statistical +/- (league average forced to = 0) which I'll call "SPM1", but also a "team-based" SPM, where the minute-weighted team average of each player's SPM was forced to = the team's efficiency differential (I'll call this "SPM2"). Needless to say, the former misses on some defensive-minded players but is more consistent year-to-year and picks up more on individual performance rather than team performance; the latter gives defenders credit for being on good defensive teams, but overly credits bad players on good teams/punishes good players on bad teams. Here are the leaders from last year in each (split seasons, min 500 MP):
Code:
Player Year Ag Ht Pos Tm G Min SPM1
LeBron James 2008 23 80 F CLE 75 3027 13.95
Chris Paul 2008 22 72 G NOH 80 3006 10.02
Kobe Bryant 2008 29 78 G LAL 82 3192 8.83
Manu Ginobili 2008 30 78 G SAS 74 2299 8.71
Kevin Garnett 2008 31 83 F BOS 71 2328 7.96
A Stoudemire 2008 25 82 F PHO 79 2677 7.45
Pau Gasol 2008 27 84 F LAL 27 918 7.21
Dirk Nowitzki 2008 29 83 F DAL 77 2769 7.03
C Billups 2008 31 75 G DET 78 2522 6.86
Shawn Marion 2008 29 79 F PHO 47 1713 6.62
A Kirilenko 2008 26 81 F UTA 72 2217 6.30
Caron Butler 2008 27 79 F WAS 58 2314 6.02
Baron Davis 2008 28 75 G GSW 82 3196 5.85
Marcus Camby 2008 33 83 C DEN 79 2758 5.75
Josh Smith 2008 22 81 F ATL 81 2873 5.72
Jason Kidd 2008 34 76 G DAL 29 1011 5.56
Chris Bosh 2008 23 82 F TOR 67 2425 5.53
Tim Duncan 2008 31 84 F SAS 78 2651 5.50
Amir Johnson 2008 20 81 F DET 62 764 5.44
Andrew Bynum 2008 20 84 C LAL 35 1008 5.14
Code:
Player Year Ag Ht Pos Tm G Min SPM2
Kevin Garnett 2008 31 83 F BOS 71 2328 17.37
Chris Paul 2008 22 72 G NOH 80 3006 15.40
Paul Pierce 2008 30 78 F BOS 80 2874 14.15
C Billups 2008 31 75 G DET 78 2522 14.11
Manu Ginobili 2008 30 78 G SAS 74 2299 13.80
Kobe Bryant 2008 29 78 G LAL 82 3192 13.68
LeBron James 2008 23 80 F CLE 75 3027 13.33
Amir Johnson 2008 20 81 F DET 62 764 12.69
Pau Gasol 2008 27 84 F LAL 27 918 12.06
James Posey 2008 31 80 F BOS 74 1821 11.93
R Wallace 2008 33 82 F DET 77 2346 11.92
A Kirilenko 2008 26 81 F UTA 72 2217 11.79
Rajon Rondo 2008 21 73 G BOS 77 2306 11.39
Ray Allen 2008 32 77 G BOS 73 2624 11.28
Dirk Nowitzki 2008 29 83 F DAL 77 2769 11.12
A Stoudemire 2008 25 82 F PHO 79 2677 11.11
Tim Duncan 2008 31 84 F SAS 78 2651 10.58
Shawn Marion 2008 29 79 F PHO 47 1713 10.28
K Perkins 2008 23 82 F BOS 78 1912 10.27
Andrew Bynum 2008 20 84 C LAL 35 1008 9.99
I'm sticking to SPM1, but it is interesting to see the effect of forcing the team's minute-weighted average to equal its PtDiff, rather than forcing the league's weighted avg to equal zero.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Thu Sep 18, 2008 5:31 pm Post subject: Reply with quote
"after removing half of the best and worst statistical plus minus seasons and averaging the season totals since 2003"
can you explain this part more please?
reason?
Back to top
View user's profile Send private message
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Thu Sep 18, 2008 5:35 pm Post subject: Reply with quote
DW I'd be interested in seeing more data on SPM2 if you are willing to post a document or post further here.
Back to top
View user's profile Send private message
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Thu Sep 18, 2008 9:05 pm Post subject: Reply with quote
I just removed half of the best and worst seasons in adjusted plus minus due to the high variability of the statistic. If I wanted to look in depth into this, I would sum the apm times a weight each season according to a credibility factor from the adjusted +/- standard deviation and then divide by statistical +/- adjusted as DW21 suggested after multiply each season by the same factor above. (These standard deviation aren't all available). As an estimate, I kept half of the extreme seasons to reflect a possible improvement in player performance, but removed a half to reflect possible statistical error. As a note, Garnett tops the list for average adjusted +/- of 9.6.
Who do we think is attributable to covering defense? On the over list, I find Bowen, Duncan, Garnett, Prince and Mutombo as all league defenders. But the bottom list has Marion, Hughes, Camby, Billups and BenWallace. Obviously, Bowen and Prince are the most 2 "non-statistical defenders".
DW21, How do you calculate statistical +/-? I got all variables from Dan R's latest coeffients (I think) on http://sonicscentral.com/apbrmetrics/vi ... .php?t=327
I did the same thing, but was getting 6.5 for Garnett and 8.9 for Lebron, etc. in 2008.
Back to top
View user's profile Send private message Visit poster's website
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Fri Sep 19, 2008 12:12 am Post subject: Reply with quote
Mountain: I just took the raw stat +/- used for SPM1 and, instead of adding the "league constant" that forces the league avg to equal zero, I added a "team constant" that forces the team's weighted average to equal its efficiency differential.
Scott: I use the weights in this article, after "translating" a player's stats to the 2002-2004 environment.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Fri Sep 19, 2008 2:59 am Post subject: Reply with quote
Ok Scott S., I can follow that logic. Just wanted to be sure.
davis21wylie2121, sorry, I meant interest in data
"on more players" but of course on your schedule and format.
And same thing Scott S. Your ratios for more players would be of interest to me as well.
Thanks to you both for the timely work. Data on more players is likely to be interesting for those players and helpful in trend analysis.
In Dan's original post that you reference Scott I want to highlight for possible further discussion a few things I see again:
offensive INTERCEP = -7.056284, that would be a per 40 minute value for a player who did absolutely nothing in the boxscore? And for defense = -3.683703.
So weighting half offense, half defense, a no box score impact player, shifted to per 48 minutes basis, comes in at -6.75? That happens to be roughly what a player at about the low 15th percentile pulls on adjusted +/-. Replacement player has other definitions but maybe this level of player (a player who on average would be roughly below the 1oth man per team on impact if players of different impacts were distributed equally thru the league) needs a name and should be kept in mind whether it is called replacement level or whatever.
"4. Interestingly, even after accounting for points scored, players with more three point attempts tend to have higher offensive adjusted plus/minus ratings. This suggests that in addition to the points they score, the ability of three point shooters to spread the floor is very important for offenses. Also, note that more three point attempts is not associated with worse defense. It does not appear that the long rebounds from missed three pointers is typically leading to easy transition points."
Wonder what value 3 pt attempts for just the part over league average rate would get.
Top 8 teams in playoffs by regular season rank for 3pt shot frequency: Lakers 6th, Celtics 9th, Spurs 8th, Hornets 12th, Cavs 11th.... Magic 1st.... Piston 21st, Jazz 28th . Looks like a pattern pretty consisent with 3 point shooting being important apart from direct value. And the Pistons and Jazz being out of step with that pattern, though obviously you could win pretty well more than one way.
"11. Holding the other variables constant, players with more personal fouls tend to be more effective defenders."
With defensive PF = 0.309126, almost as big as defensive rebound on its own, and after the net of prorated FTA and PTS even bigger?
So Jerry Sloan's hacking defense, on average, is in line with the regression? Wonder what value fouls over league average rate would get?
And is Dave Berri's value for defensive rebounds carrying this value for defensive fouls?
At http://www.wagesofwins.com/CalculatingWinsProduced.html I see "value of personal fouls .... is calculated from the value of the opponent’s free throws made."
So if defensive PFs has positive value according to Dan"s regression that positive value appears to be going elsewhere in Wins Produced, right? I don't see fouls in the team defensive adjustment so I believe this value is going to defensive rebounds. So WP's value for defensive rebounds is a bundle of defensive rebounds and other defensive behavior (the value of the fouls themselves beyond the FT cost- and also and probably more importantly the active defensive play that doesn't result in fouls but is associated with players who pick up fouls).
Certainly this treatment blurs things / doesn't assign the credit accurately, but that value associated with defensive fouls by default. in both Dan's regression and Dave Berri's system, is legitimate player added defensive value, at least at league level. How much of that value of fouls is from big men vs perimeter? If it is more from bigs and interior players foul more (I think, but haven't checked) then the lumping of the defensive fouls with defensive rebounds may not be that bad and may help balance the image created by box score stats (and PER) more heavily covering offensive behavior. Still, to my understanding, it would be better if the positive defensive value of fouls (defensive basket protection effort) were in WP separately instead of lumped with defensive rebounds.
I also see later in the discussion from dlirag on page 3
"Adding in the defense coefficients, the point where TS% has no projected effect on overall efficiency is between 49% and 50%. Are these calculations right?
-0.525 / (-2 * 0.703) ~ 0.373 TS% [Off efficiency]
(-0.525 + -0.105) / (-2 * (0.703 + -0.0676)) ~ 0.496 TS% [Both off and def efficiency]"
I didn't see any response then but isn't this offense + defense impact according to dlirag's calculation basically in line with Dave Berri's Wins Produced treatment of FG misses or haven't I fully processed this?
Back to top
View user's profile Send private message
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Fri Sep 19, 2008 1:01 pm Post subject: Reply with quote
DW21:I don't know exactly why Dan R chose the method you used, but it seems it probably happened to fit his specific set of data. I don't think it makes too much sense to assess Dennis Rodman's defensive rebounding value to be far inferior to John Stockton's. However I do like the second degree TSA's to estimate shot creating ability.
Mountain: Love the observations/discussion. I thought I missed the boat on this conversation and would love to consider it more in detail.
For one, someone with no boxscores would be "predicted" to actually have a -13.5 (didn't exactly check the math) expected adj +/- since offensive and defensive adj +/- numbers should be summed and not added, to my understanding. Actually, this is outside the obseved range and is not really an assertion one should make.
As to the 3 point attempts, I have a few observations from tests I did. (Note that I was not happy enough with my results to finalize them.) Someone who can spread the floor adds great value to his team. An inside scoring presence allows the outside to open up and an outside shooting presence allows for more easy shots down low. The worst type of statistic I could find was an assist on a 2 point jumper. This clogs up the defense and offers the lowest expected point value based on the low percentage and low reward. Unassisted 3 pointers actually also capture a large portion of the ability to create a shot. Logically, how often do players shoot unassisted 3 pointers? I can only think of distress situations where a low shooting percentage can still be very valuable. I agree that a true test for a "3 point presence" would probably be more telling.
Initially, I was surprised to see the positive value of personal fouls. It seems counterintuitive to expect both drawing fouls and committing fouls to create value. However, upon considering this, it makes a lot of sense. How often are players told to foul? Many players primary value is in "using up fouls." Also, a foul isn't always called and provides a defensive advantage. Many players who get to the line a lot are a combination of poor free throw shooters and effective shooters in general. This means that if Shaq has any reasonable shot available, you should foul him. This adds value to fouling Shaq if it's done under the right circumstances (you still want to try to deny him the ball). It would take advanced statistics to extract the good fouls, bad fouls and a circumstance better than those 2 but worse than a block or a steal - a defensive missed shot or TO forced.
As far as the break even point for True shooting percentage, I already stated I believe there is definately at least situational value in creating a shot. However, there seems to be a slight tendency for good offensive players to have a worse defensive value than otherwise expected. I think it is ironic that the combined TS% breakeven is around average, but it might make sense in an economic mindset (one can have an excess benefit offensively until his defensive liability catches up). Just a guess, I'd have to look into it more.
Another thing to note, Dan R says players with more turnovers are, on average, worse defenders. However, the percentage of turnovers resulting from steals times the offensive value of steals are pretty much in line with the defensive value subtracted by turnovers.
I'll stop here because I could go on about this stuff forever.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Fri Sep 19, 2008 1:55 pm Post subject: Reply with quote
Scott thanks for the further discussion & correction about adding the offense and defense scores as is together. It was late and I did go off track. No NBA player is truly down in the do absolutely anything of value range. But saying a 15th percentile -6.75 guy on statistical +/- or adjusted is doing about 50% of what they'd have to do to have a neutral impact, which would be an accomplishment given the high level of competition, sounds about right. A replacement level impact, even though some of the guys at this level of impact are higher in the rotation.
Last edited by Mountain on Sat Sep 20, 2008 12:46 am; edited 2 times in total
Back to top
View user's profile Send private message
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Fri Sep 19, 2008 7:21 pm Post subject: Reply with quote
I figured to see exactly what was missing, I took NBA Live 2007 ratings for each of the 86(!) categories to see if I could find some consistencies from the top and bottom 30 (of 134). What I did was took the ratios of each rating to average and then took the average ratio of the top 30 over the average ratio of the bottom 30 and normalized to 100%. I can't say the results were what I expected, but I notice a few patterns. Here it is:
Code:
120% Shotblock
117% Offensive Rebounding
117% Offensive Rebounding Strength
116% Hands Inp Post
115% Offensive Box Out Effectiveness
115% Strength
114% Tipins/Putbacks
113% Offensive Sink Or Swim Effectiveness
113% Blocker
113% Defensive Post Strength
112% Post Strength
112% Defensive Rebounding Strength
112% InsidePresence
112% Post Savvy
111% Defensive Box Out Effectiveness
111% Defensive Rebounding
111% Dunk Ability
110% Inside Shooting Ability
110% Setting Screens
110% Offensive Post Footwork
109% Inside Shooting Likelihood
109% Defensive Sink Or Swim Effectiveness
109% Shotalter
108% Defensive Post Footwork
108% Inside
108% Dunk vs Layup Frequency
107% Fighting Through The Screen
106% Highflyer
105% Layup/Dunk Aggressiveness
104% Layup Ability
103% Midrange Shooting Likelihood
103% Midrange Shooting Ability
103% Freethrow Ability
103% Defense
102% Dunk/Layup Shot Off
101% Vertical
101% Shooter's Mentality
101% Scorer
100% Scoring With Contact
99% Shooting Touch
99% Overall
99% Athleticism
98% Foul Mistakes
98% Drawing Offensive Foul
98% Layup In Traffic
98% Mental Toughness
98% Drawing Defensive Foul
97% Helpside Effectiveness
97% Shooter
97% Conditioning
97% Hedging The Screen
97% Double Teamability
96% Closeout Effectiveness
96% Agility
96% Lateral Quickness
95% Using Screens
95% Recovery Quickness
95% Outside
95% Quickness
94% On Ball Effectiveness
94% Outside Shooting Ability
94% Fastbreak Anticipation
94% Acceleration
94% Passlane Defense Effectiveness
94% Goto Guy
94% Teamwork
93% Max Speed
93% Dribbling
93% Double Team Composure
92% Hustle
92% Offball Awareness
92% Tenacity
91% Ballcarrier Speed
91% Interception
91% Hands On Perimeter
90% Triple Threat Effectiveness
90% First Step
90% Pass Accuracy
90% Stealer
89% Clutch
89% Penetration Effectiveness
89% Playmaker
87% Difficult Pass Ability
87% Steal
85% Court Vision
74% Outside Shooting Likelihood
It seems some of the suggestions of the coefficient estimates (outside shooting) were over-valued (not that they aren't important). At the top of the list was inside presence related skills. The main thing I guess I was expecting were intagibles such as man defense and scrappiness, which did not seem to show through. I did expect post presence to be of importance, but thought that fouls drawn and shot attempts would reflect much of this in stat +/-. Of course many of the rankings in NBA Live 07 seem different than I would expect subjectively (I specifically looked at defense), but they did a solid job of predicting adjusted +/-.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Sat Sep 20, 2008 2:06 pm Post subject: Reply with quote
The differences in the findings in Dan's original article and the that later offensive / defensive split post appear to be many and significant. I'd go with the later over the former but remain somewhat leary of league based statistical +/- and would like to see more on team-based.
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Thu Sep 18, 2008 4:32 pm Post subject: Biggest Stat/Adjusted +/- differences over since 2003 Reply with quote
Using the following formula (20+adjusted +/-)/(20+statistical+/-), after removing half of the best and worst statistical plus minus seasons and averaging the season totals since 2003, here are the highest and lowest ratios among those with adjusted +/- ratings each season.
Code:
Scalabrine, Brian 130%
Murray, Ronald 129%
Dooling, Keyon 126%
Allen, Malik 122%
Wallace, Rasheed 121%
Collins, Jarron 118%
Wilcox, Chris 118%
Jeffries, Jared 117%
Collins, Jason 117%
Watson, Earl 113%
Moore, Mikki 113%
Randolph, Zach 113%
Szczerbiak, Wally 112%
Prince, Tayshaun 112%
Pargo, Jannero 112%
Bowen, Bruce 111%
Miller, Brad 111%
Mutombo, Dikembe 111%
Hassell, Trenton 110%
Thomas, Kurt 110%
Radmanovic, Vladimir 110%
Arroyo, Carlos 110%
Duncan, Tim 109%
Pierce, Paul 109%
Jamison, Antawn 109%
Foster, Jeff 108%
Rose, Malik 108%
Brown, Devin 108%
Ming, Yao 107%
Garnett, Kevin 107%
Code:
Murphy, Troy 75%
Mohammed, Nazr 78%
Swift, Stromile 78%
Camby, Marcus 81%
Bibby, Mike 83%
Posey, James 85%
Marion, Shawn 85%
Arenas, Gilbert 86%
Hughes, Larry 86%
Billups, Chauncey 86%
Williams, Jason 87%
Gooden, Drew 88%
Curry, Eddy 88%
Salmons, John 89%
Jefferson, Richard 90%
Knight, Brevin 90%
McInnis, Jeff 91%
Brown, Kwame 91%
Hamilton, Richard 91%
Wallace, Ben 92%
Stoudamire, Damon 92%
Skinner, Brian 93%
Butler, Caron 93%
Daniels, Antonio 93%
O'Neal, Jermaine 94%
Boykins, Earl 94%
Redd, Michael 94%
Lue, Tyronn 94%
Dampier, Erick 94%
Boozer, Carlos 94%
What are some of the explanations we can come up with. Murray, Moore ad Devin Brown had low 2003 minutes and should be eliminated. Hamilton and, possibly Rasheed Wallace can still be attributed, at least in part to the 2006 adjusted +/- ratings. Since statistical +/- is not pace adjusted (at least the one I used), that can influence some. Some rankings can be assumed to be a result of the impact of man defense (Bowen and Redd). Some players on the positive end seem to fit the scrappy, role player with intangibles. These results don't seem completely random to me, but given the length of time we are looking at, I was expecting slightly more obvious tendencies and patterns on both spectrums.
Back to top
View user's profile Send private message Visit poster's website
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Thu Sep 18, 2008 5:00 pm Post subject: Reply with quote
Defense is obviously the biggie. Statistical +/- is in many ways superior to a BoP-based "total player rating" because SPM is largely team-independent and DRtg is so team-dependent. However, defense is obviously a big part of the game -- can we justify ignoring it altogether? In reviving Dan's old SPM formula and applying it to every season since 1971, I came up with 2 sets of data: standard statistical +/- (league average forced to = 0) which I'll call "SPM1", but also a "team-based" SPM, where the minute-weighted team average of each player's SPM was forced to = the team's efficiency differential (I'll call this "SPM2"). Needless to say, the former misses on some defensive-minded players but is more consistent year-to-year and picks up more on individual performance rather than team performance; the latter gives defenders credit for being on good defensive teams, but overly credits bad players on good teams/punishes good players on bad teams. Here are the leaders from last year in each (split seasons, min 500 MP):
Code:
Player Year Ag Ht Pos Tm G Min SPM1
LeBron James 2008 23 80 F CLE 75 3027 13.95
Chris Paul 2008 22 72 G NOH 80 3006 10.02
Kobe Bryant 2008 29 78 G LAL 82 3192 8.83
Manu Ginobili 2008 30 78 G SAS 74 2299 8.71
Kevin Garnett 2008 31 83 F BOS 71 2328 7.96
A Stoudemire 2008 25 82 F PHO 79 2677 7.45
Pau Gasol 2008 27 84 F LAL 27 918 7.21
Dirk Nowitzki 2008 29 83 F DAL 77 2769 7.03
C Billups 2008 31 75 G DET 78 2522 6.86
Shawn Marion 2008 29 79 F PHO 47 1713 6.62
A Kirilenko 2008 26 81 F UTA 72 2217 6.30
Caron Butler 2008 27 79 F WAS 58 2314 6.02
Baron Davis 2008 28 75 G GSW 82 3196 5.85
Marcus Camby 2008 33 83 C DEN 79 2758 5.75
Josh Smith 2008 22 81 F ATL 81 2873 5.72
Jason Kidd 2008 34 76 G DAL 29 1011 5.56
Chris Bosh 2008 23 82 F TOR 67 2425 5.53
Tim Duncan 2008 31 84 F SAS 78 2651 5.50
Amir Johnson 2008 20 81 F DET 62 764 5.44
Andrew Bynum 2008 20 84 C LAL 35 1008 5.14
Code:
Player Year Ag Ht Pos Tm G Min SPM2
Kevin Garnett 2008 31 83 F BOS 71 2328 17.37
Chris Paul 2008 22 72 G NOH 80 3006 15.40
Paul Pierce 2008 30 78 F BOS 80 2874 14.15
C Billups 2008 31 75 G DET 78 2522 14.11
Manu Ginobili 2008 30 78 G SAS 74 2299 13.80
Kobe Bryant 2008 29 78 G LAL 82 3192 13.68
LeBron James 2008 23 80 F CLE 75 3027 13.33
Amir Johnson 2008 20 81 F DET 62 764 12.69
Pau Gasol 2008 27 84 F LAL 27 918 12.06
James Posey 2008 31 80 F BOS 74 1821 11.93
R Wallace 2008 33 82 F DET 77 2346 11.92
A Kirilenko 2008 26 81 F UTA 72 2217 11.79
Rajon Rondo 2008 21 73 G BOS 77 2306 11.39
Ray Allen 2008 32 77 G BOS 73 2624 11.28
Dirk Nowitzki 2008 29 83 F DAL 77 2769 11.12
A Stoudemire 2008 25 82 F PHO 79 2677 11.11
Tim Duncan 2008 31 84 F SAS 78 2651 10.58
Shawn Marion 2008 29 79 F PHO 47 1713 10.28
K Perkins 2008 23 82 F BOS 78 1912 10.27
Andrew Bynum 2008 20 84 C LAL 35 1008 9.99
I'm sticking to SPM1, but it is interesting to see the effect of forcing the team's minute-weighted average to equal its PtDiff, rather than forcing the league's weighted avg to equal zero.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Thu Sep 18, 2008 5:31 pm Post subject: Reply with quote
"after removing half of the best and worst statistical plus minus seasons and averaging the season totals since 2003"
can you explain this part more please?
reason?
Back to top
View user's profile Send private message
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Thu Sep 18, 2008 5:35 pm Post subject: Reply with quote
DW I'd be interested in seeing more data on SPM2 if you are willing to post a document or post further here.
Back to top
View user's profile Send private message
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Thu Sep 18, 2008 9:05 pm Post subject: Reply with quote
I just removed half of the best and worst seasons in adjusted plus minus due to the high variability of the statistic. If I wanted to look in depth into this, I would sum the apm times a weight each season according to a credibility factor from the adjusted +/- standard deviation and then divide by statistical +/- adjusted as DW21 suggested after multiply each season by the same factor above. (These standard deviation aren't all available). As an estimate, I kept half of the extreme seasons to reflect a possible improvement in player performance, but removed a half to reflect possible statistical error. As a note, Garnett tops the list for average adjusted +/- of 9.6.
Who do we think is attributable to covering defense? On the over list, I find Bowen, Duncan, Garnett, Prince and Mutombo as all league defenders. But the bottom list has Marion, Hughes, Camby, Billups and BenWallace. Obviously, Bowen and Prince are the most 2 "non-statistical defenders".
DW21, How do you calculate statistical +/-? I got all variables from Dan R's latest coeffients (I think) on http://sonicscentral.com/apbrmetrics/vi ... .php?t=327
I did the same thing, but was getting 6.5 for Garnett and 8.9 for Lebron, etc. in 2008.
Back to top
View user's profile Send private message Visit poster's website
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Fri Sep 19, 2008 12:12 am Post subject: Reply with quote
Mountain: I just took the raw stat +/- used for SPM1 and, instead of adding the "league constant" that forces the league avg to equal zero, I added a "team constant" that forces the team's weighted average to equal its efficiency differential.
Scott: I use the weights in this article, after "translating" a player's stats to the 2002-2004 environment.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Fri Sep 19, 2008 2:59 am Post subject: Reply with quote
Ok Scott S., I can follow that logic. Just wanted to be sure.
davis21wylie2121, sorry, I meant interest in data
"on more players" but of course on your schedule and format.
And same thing Scott S. Your ratios for more players would be of interest to me as well.
Thanks to you both for the timely work. Data on more players is likely to be interesting for those players and helpful in trend analysis.
In Dan's original post that you reference Scott I want to highlight for possible further discussion a few things I see again:
offensive INTERCEP = -7.056284, that would be a per 40 minute value for a player who did absolutely nothing in the boxscore? And for defense = -3.683703.
So weighting half offense, half defense, a no box score impact player, shifted to per 48 minutes basis, comes in at -6.75? That happens to be roughly what a player at about the low 15th percentile pulls on adjusted +/-. Replacement player has other definitions but maybe this level of player (a player who on average would be roughly below the 1oth man per team on impact if players of different impacts were distributed equally thru the league) needs a name and should be kept in mind whether it is called replacement level or whatever.
"4. Interestingly, even after accounting for points scored, players with more three point attempts tend to have higher offensive adjusted plus/minus ratings. This suggests that in addition to the points they score, the ability of three point shooters to spread the floor is very important for offenses. Also, note that more three point attempts is not associated with worse defense. It does not appear that the long rebounds from missed three pointers is typically leading to easy transition points."
Wonder what value 3 pt attempts for just the part over league average rate would get.
Top 8 teams in playoffs by regular season rank for 3pt shot frequency: Lakers 6th, Celtics 9th, Spurs 8th, Hornets 12th, Cavs 11th.... Magic 1st.... Piston 21st, Jazz 28th . Looks like a pattern pretty consisent with 3 point shooting being important apart from direct value. And the Pistons and Jazz being out of step with that pattern, though obviously you could win pretty well more than one way.
"11. Holding the other variables constant, players with more personal fouls tend to be more effective defenders."
With defensive PF = 0.309126, almost as big as defensive rebound on its own, and after the net of prorated FTA and PTS even bigger?
So Jerry Sloan's hacking defense, on average, is in line with the regression? Wonder what value fouls over league average rate would get?
And is Dave Berri's value for defensive rebounds carrying this value for defensive fouls?
At http://www.wagesofwins.com/CalculatingWinsProduced.html I see "value of personal fouls .... is calculated from the value of the opponent’s free throws made."
So if defensive PFs has positive value according to Dan"s regression that positive value appears to be going elsewhere in Wins Produced, right? I don't see fouls in the team defensive adjustment so I believe this value is going to defensive rebounds. So WP's value for defensive rebounds is a bundle of defensive rebounds and other defensive behavior (the value of the fouls themselves beyond the FT cost- and also and probably more importantly the active defensive play that doesn't result in fouls but is associated with players who pick up fouls).
Certainly this treatment blurs things / doesn't assign the credit accurately, but that value associated with defensive fouls by default. in both Dan's regression and Dave Berri's system, is legitimate player added defensive value, at least at league level. How much of that value of fouls is from big men vs perimeter? If it is more from bigs and interior players foul more (I think, but haven't checked) then the lumping of the defensive fouls with defensive rebounds may not be that bad and may help balance the image created by box score stats (and PER) more heavily covering offensive behavior. Still, to my understanding, it would be better if the positive defensive value of fouls (defensive basket protection effort) were in WP separately instead of lumped with defensive rebounds.
I also see later in the discussion from dlirag on page 3
"Adding in the defense coefficients, the point where TS% has no projected effect on overall efficiency is between 49% and 50%. Are these calculations right?
-0.525 / (-2 * 0.703) ~ 0.373 TS% [Off efficiency]
(-0.525 + -0.105) / (-2 * (0.703 + -0.0676)) ~ 0.496 TS% [Both off and def efficiency]"
I didn't see any response then but isn't this offense + defense impact according to dlirag's calculation basically in line with Dave Berri's Wins Produced treatment of FG misses or haven't I fully processed this?
Back to top
View user's profile Send private message
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Fri Sep 19, 2008 1:01 pm Post subject: Reply with quote
DW21:I don't know exactly why Dan R chose the method you used, but it seems it probably happened to fit his specific set of data. I don't think it makes too much sense to assess Dennis Rodman's defensive rebounding value to be far inferior to John Stockton's. However I do like the second degree TSA's to estimate shot creating ability.
Mountain: Love the observations/discussion. I thought I missed the boat on this conversation and would love to consider it more in detail.
For one, someone with no boxscores would be "predicted" to actually have a -13.5 (didn't exactly check the math) expected adj +/- since offensive and defensive adj +/- numbers should be summed and not added, to my understanding. Actually, this is outside the obseved range and is not really an assertion one should make.
As to the 3 point attempts, I have a few observations from tests I did. (Note that I was not happy enough with my results to finalize them.) Someone who can spread the floor adds great value to his team. An inside scoring presence allows the outside to open up and an outside shooting presence allows for more easy shots down low. The worst type of statistic I could find was an assist on a 2 point jumper. This clogs up the defense and offers the lowest expected point value based on the low percentage and low reward. Unassisted 3 pointers actually also capture a large portion of the ability to create a shot. Logically, how often do players shoot unassisted 3 pointers? I can only think of distress situations where a low shooting percentage can still be very valuable. I agree that a true test for a "3 point presence" would probably be more telling.
Initially, I was surprised to see the positive value of personal fouls. It seems counterintuitive to expect both drawing fouls and committing fouls to create value. However, upon considering this, it makes a lot of sense. How often are players told to foul? Many players primary value is in "using up fouls." Also, a foul isn't always called and provides a defensive advantage. Many players who get to the line a lot are a combination of poor free throw shooters and effective shooters in general. This means that if Shaq has any reasonable shot available, you should foul him. This adds value to fouling Shaq if it's done under the right circumstances (you still want to try to deny him the ball). It would take advanced statistics to extract the good fouls, bad fouls and a circumstance better than those 2 but worse than a block or a steal - a defensive missed shot or TO forced.
As far as the break even point for True shooting percentage, I already stated I believe there is definately at least situational value in creating a shot. However, there seems to be a slight tendency for good offensive players to have a worse defensive value than otherwise expected. I think it is ironic that the combined TS% breakeven is around average, but it might make sense in an economic mindset (one can have an excess benefit offensively until his defensive liability catches up). Just a guess, I'd have to look into it more.
Another thing to note, Dan R says players with more turnovers are, on average, worse defenders. However, the percentage of turnovers resulting from steals times the offensive value of steals are pretty much in line with the defensive value subtracted by turnovers.
I'll stop here because I could go on about this stuff forever.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Fri Sep 19, 2008 1:55 pm Post subject: Reply with quote
Scott thanks for the further discussion & correction about adding the offense and defense scores as is together. It was late and I did go off track. No NBA player is truly down in the do absolutely anything of value range. But saying a 15th percentile -6.75 guy on statistical +/- or adjusted is doing about 50% of what they'd have to do to have a neutral impact, which would be an accomplishment given the high level of competition, sounds about right. A replacement level impact, even though some of the guys at this level of impact are higher in the rotation.
Last edited by Mountain on Sat Sep 20, 2008 12:46 am; edited 2 times in total
Back to top
View user's profile Send private message
Scott S
Joined: 10 Feb 2008
Posts: 46
Location: East Rutherford, NJ
PostPosted: Fri Sep 19, 2008 7:21 pm Post subject: Reply with quote
I figured to see exactly what was missing, I took NBA Live 2007 ratings for each of the 86(!) categories to see if I could find some consistencies from the top and bottom 30 (of 134). What I did was took the ratios of each rating to average and then took the average ratio of the top 30 over the average ratio of the bottom 30 and normalized to 100%. I can't say the results were what I expected, but I notice a few patterns. Here it is:
Code:
120% Shotblock
117% Offensive Rebounding
117% Offensive Rebounding Strength
116% Hands Inp Post
115% Offensive Box Out Effectiveness
115% Strength
114% Tipins/Putbacks
113% Offensive Sink Or Swim Effectiveness
113% Blocker
113% Defensive Post Strength
112% Post Strength
112% Defensive Rebounding Strength
112% InsidePresence
112% Post Savvy
111% Defensive Box Out Effectiveness
111% Defensive Rebounding
111% Dunk Ability
110% Inside Shooting Ability
110% Setting Screens
110% Offensive Post Footwork
109% Inside Shooting Likelihood
109% Defensive Sink Or Swim Effectiveness
109% Shotalter
108% Defensive Post Footwork
108% Inside
108% Dunk vs Layup Frequency
107% Fighting Through The Screen
106% Highflyer
105% Layup/Dunk Aggressiveness
104% Layup Ability
103% Midrange Shooting Likelihood
103% Midrange Shooting Ability
103% Freethrow Ability
103% Defense
102% Dunk/Layup Shot Off
101% Vertical
101% Shooter's Mentality
101% Scorer
100% Scoring With Contact
99% Shooting Touch
99% Overall
99% Athleticism
98% Foul Mistakes
98% Drawing Offensive Foul
98% Layup In Traffic
98% Mental Toughness
98% Drawing Defensive Foul
97% Helpside Effectiveness
97% Shooter
97% Conditioning
97% Hedging The Screen
97% Double Teamability
96% Closeout Effectiveness
96% Agility
96% Lateral Quickness
95% Using Screens
95% Recovery Quickness
95% Outside
95% Quickness
94% On Ball Effectiveness
94% Outside Shooting Ability
94% Fastbreak Anticipation
94% Acceleration
94% Passlane Defense Effectiveness
94% Goto Guy
94% Teamwork
93% Max Speed
93% Dribbling
93% Double Team Composure
92% Hustle
92% Offball Awareness
92% Tenacity
91% Ballcarrier Speed
91% Interception
91% Hands On Perimeter
90% Triple Threat Effectiveness
90% First Step
90% Pass Accuracy
90% Stealer
89% Clutch
89% Penetration Effectiveness
89% Playmaker
87% Difficult Pass Ability
87% Steal
85% Court Vision
74% Outside Shooting Likelihood
It seems some of the suggestions of the coefficient estimates (outside shooting) were over-valued (not that they aren't important). At the top of the list was inside presence related skills. The main thing I guess I was expecting were intagibles such as man defense and scrappiness, which did not seem to show through. I did expect post presence to be of importance, but thought that fouls drawn and shot attempts would reflect much of this in stat +/-. Of course many of the rankings in NBA Live 07 seem different than I would expect subjectively (I specifically looked at defense), but they did a solid job of predicting adjusted +/-.
Back to top
View user's profile Send private message Visit poster's website
Mountain
Joined: 13 Mar 2007
Posts: 1527
PostPosted: Sat Sep 20, 2008 2:06 pm Post subject: Reply with quote
The differences in the findings in Dan's original article and the that later offensive / defensive split post appear to be many and significant. I'd go with the later over the former but remain somewhat leary of league based statistical +/- and would like to see more on team-based.