The Warriors and APM
by philthiest on Aug 27, 2009 9:25 AM PDT
54 comments
I thought I'd share something I stumbled upon a little while ago: 2008-9 Adjusted Plus Minus ratings. For those of you who aren't familiar, Adjusted +/- (or just APM) is a metric for separating an individuals contribution to the team based on how the team does with him on the floor. APM has gotten some street cred as a stat recently, since it was referenced in a Micheal Lewis article. More info can be found here and here. These numbers were compiled by Steve Ilardi, and a little more about his method for these can be found here. To reduce noise, he used 6 seasons of data, weighting earlier seasons much less than the 2008-9 season. Here are the numbers for the Warriors:
Off Def Std Total
Team Player Min APM APM Err APM
GSW, NYK Harrington, Al 2,479 1.56 0.04 0.90 1.60
GSW Jackson, Stephen 2,184 1.59 -0.39 0.96 1.20
GSW Morrow, Anthony 1,433 4.09 -2.93 1.34 1.16
GSW Turiaf, Ronny 1,624 1.51 -0.49 1.09 1.02
GSW Biedrins, Andris 1,756 -1.05 1.93 1.07 0.88
GSW Wright, Brandan 649 -0.10 0.02 1.61 -0.08
GSW Maggette, Corey 1,494 1.26 -1.59 0.97 -0.33
GSW Belinelli, Marco 851 -2.33 1.81 1.49 -0.52
GSW, NYK Crawford, Jamal 2,379 2.57 -3.96 0.89 -1.39
GSW Ellis, Monta 834 -0.16 -1.51 1.14 -1.66
GSW Randolph, Anthony 1,106 -0.80 -0.89 1.43 -1.69
GSW Azubuike, Kelenna 2,262 -2.21 0.37 1.05 -1.83
GSW Watson, C.J. 1,776 0.04 -2.13 1.24 -2.10
GSW Kurz, Rob 427 -10.91 0.35 2.08 -10.57
Gut reactions:
- The Harrington-Crawford trade didn't work out in our favor.
- At first glance, it seems we are underrating Biedrins defense. While this might be true, he's still by no means an elite defender. The metric seems give an advantage to big men on defense (which seems logical). Here are the top 10 by Defensive APM:
Team Player Minutes Def APM
BOS Garnett, Kevin 1,642 8.74
MIL Bogut, Andrew 1,136 8.21
DET Johnson, Amir 906 8.05
HOU Hayes, Chuck 786 7.31
POR Przybilla, Joel 1,935 7.04
PHI Iguodala, Andre 3,269 6.64
DEN Hilario, Nene 2,490 6.56
HOU Ming, Yao 2,454 5.70
PHI Young, Thaddeu 2,580 5.70
LAC Camby, Marcus 1,925 5.66
- It also seems Belinelli might have a slightly underrated defender. CJ Watson and Jamal Crawford, though, seem to be right where we expect them.
- Anthony Morrow is a good offensive player. That pretty much goes without saying.
- Our best player (Stephen Jackson) was the 100th best player in the league.
- I think we're all going to miss Rob Kurz.
- The Warriors off-season additions speak for themselves:
DAL George, Devean 673 -6.66 -2.13 1.29 -8.78
ATL Law, Acie 553 -3.03 2.51 1.63 -0.53
ATL Claxton, Speedy NO DATA
- One big blowout listing of players everyone wants the Warriors to trade for, former Warriors, and the top-20:
PLAYERS IN EVERY FAR-FETCHED TRADE PROPOSAL
PHX Stoudemire, Amare 1,948 -2.38 2.13 1.05 -0.25
UTA Boozer, Carlos 1,155 -0.01 -1.98 1.07 -1.99
TOR Bosh, Chris 2,884 5.65 0.54 1.10 6.19
UTA Kirilenko, Andrei 1,805 1.39 3.29 1.00 4.68
NOTABLE FORMER WARRIORS
LAC Davis, Baron 2,248 1.63 0.49 0.96 2.12
CHA, PHX Richardson, Jason 2,411 0.05 -3.12 0.90 -3.07
ORL Pietrus, Mickael 1,329 0.55 0.20 0.93 0.76
IND Murphy, Troy 2,482 3.02 -2.15 1.04 0.87
IND Dunleavy, Mike 495 -2.46 0.38 1.18 -2.08
MIN Cardinal, Brian 909 3.72 1.67 1.40 5.39
NBA 2008-9 APM TOP 20
MIA Wade, Dwyane 3,048 10.66 2.96 1.16 13.61
BOS Garnett, Kevin 1,642 4.47 8.74 1.07 13.21
CLE James, LeBron 3,054 10.16 3.03 1.09 13.19
NOH Paul, Chris 2,888 9.36 3.35 1.42 12.71
PHX Nash, Steve 2,484 11.28 -2.45 1.11 8.83
LAL Odom, Lamar 2,203 3.28 5.52 0.95 8.81
PHI Iguodala, Andre 3,269 1.96 6.64 1.19 8.61
ORL Lewis, Rashard 2,859 5.17 2.94 0.96 8.11
HOU Ming, Yao 2,454 1.59 5.70 1.03 7.29
DAL Kidd, Jason 2,814 3.31 3.35 0.98 6.66
LAL Gasol, Pau 2,857 4.53 2.11 0.92 6.64
DAL Nowitzki, Dirk 2,977 4.83 1.67 1.16 6.50
PHI Young, Thaddeus 2,580 0.67 5.70 1.08 6.37
TOR Bosh, Chris 2,884 5.65 0.54 1.10 6.19
DET Johnson, Amir 906 -1.86 8.05 1.47 6.18
HOU Artest, Ron 2,343 0.83 5.01 0.93 5.83
SAS Parker, Tony 2,386 4.48 1.29 1.18 5.77
LAL Bryant, Kobe 2,824 7.57 -1.95 1.10 5.63
WAS Jamison, Antawn 3,025 4.26 1.31 1.11 5.58
SAS Duncan, Tim 2,454 1.81 3.60 1.22 5.40
So, what do you think? Are we wrong about Biedrins D? Are you excited about Morrow's offense? Do we want Amare? Boozer? Should we still try to get Baron back? Is Adjusted Plus Minus even a good metric for judging a players value?
This FanPost is a submission from a member of the mighty Golden State of Mind community. While we're all here to throw up that W, these words do not necessarily reflect the views of the GSoM Crew. Still, chances are the preceding post is Unstoppable Baby!
5 recs |
54 comments
Email
Print
Comments
www.basketballvalue.com
is another site that compiles APM.
APM is probably my one favorite, all-encompassing basketball stat.
Keep in mind though that it inherently contains a relatively large standard error. Even with that though, you can still weed out the truly good players from the truly bad ones.
"We Deserve"
by YaHeard on Aug 27, 2009 11:54 AM PDT reply actions 0 recs
basketballvalue.com is an awesome resource, especially because they offer in season APM numbers.
These numbers, however, is compiled with 6 years of data, which greatly lowered it’s standard error, although there are still some anomalies that snuck through
by philthiest on Aug 27, 2009 12:19 PM PDT up reply actions 0 recs
Keep in mind though that it inherently contains a relatively large standard error.
The SE for many players is often too large to be of any use at all. Almost none of the measures meet even rather loose criteria for statistical significance. When the error is larger than the estimate (not uncommon) the measurement has a serious flaw.
It correctly identifies most of the really good players and really horrible players (which probably didn’t need a number to help), but I wouldn’t use it to differentiate anything else. Honestly, the “unadjusted” net plus minus seems like a better measure in many many cases.
by jae on Aug 27, 2009 2:50 PM PDT up reply actions 0 recs
Honestly, the "unadjusted" net plus minus seems like a better measure in many many cases.
Raw +/- numbers are great for judging 5 man units, but as a stat for individual players, I always felt it told you about as much about a player as RBIs tell you about a hitter.
by philthiest on Aug 27, 2009 9:08 PM PDT up reply actions 0 recs
It correctly identifies most of the really good players and really horrible players (which probably didn’t need a number to help), but I wouldn’t use it to differentiate anything else.
So, you’re saying that it’s useful for backing up what we can see with our eyes (and basic statistics), but not useful for much of anything else. Great!
You have been DFiBrillated.
by Dubs fan in Boston on Aug 28, 2009 2:42 AM PDT up reply actions 0 recs
while APM is a valuable stat
i fail to see how brian friggin cardinal rates higher than STAT. he’s close to bosh according to this?
by bebopmonkey on Aug 27, 2009 12:16 PM PDT reply actions 0 recs
Confession
I really only included Cardinal in this post because his number was so high. He rated as the 21st player in the league and was almost on par with Duncan. But anyone with less than 1600 minutes should be looked at cautiously, and anyone with under 1000 is almost worth ignoring.
by philthiest on Aug 27, 2009 12:23 PM PDT up reply actions 0 recs
Question
Given Cardinal’s small sample size, how come his standard error isn’t any bigger than CP3’s or Al Harrington’s?
"We Deserve"
by YaHeard on Aug 27, 2009 2:06 PM PDT up reply actions 0 recs
rec'd, btw
this belongs at the top
"We Deserve"
by YaHeard on Aug 27, 2009 2:07 PM PDT up reply actions 0 recs
I’m guessing Paul having a higher SE than Cardinal comes from Paul playing the majority of the time with the same people. If you look at Paul’s top 5 man units, you’ll notice, among other things, that they all include West. Cardinal, though he played less minutes, played on much more varied units.
by philthiest on Aug 27, 2009 9:22 PM PDT up reply actions 0 recs
brian cardinal is responsible for one of my favorite box scores from last year. in one game he played something like 8 minutes (i think) and had zeroes across the board, including plus/minus in a game where the wolves were trounced. good times.
heart of a champion, will of the warrior.
by cap'n hack on Aug 27, 2009 1:29 PM PDT up reply actions 0 recs
PLAYERS IN EVERY FAR-FETCHED TRADE PROPOSAL
1. Bosh +6.2
2. Kirilenko +4.7
3. Stoudemire -0.3
4. Boozer -2.0
Interestingly, that’s about the order and degree to which I covet them on the Warriors. Too bad #1 (and probably #2, alas) aren’t really available. As for #3 and #4 … sure, they’re both coming off uncharacteristically mediocre seasons, but in light of these numbers, their dicey contract situations (i.e. either one-year rentals or long-term pricy extensions), and the bloated asking price that goes along with being an ALL-STAR, I just can’t see why either of them would excite anybody.
"I’M TAKING YOU TO THE HOLE, WONDERLIC!" [turns ball over] -- markdash
Get off my lawn, hipsters! -- Missing Barry
by Sleepy Freud on Aug 27, 2009 12:31 PM PDT reply actions 0 recs
p.s.
Awesome post, Philthiest. Rec.
"I’M TAKING YOU TO THE HOLE, WONDERLIC!" [turns ball over] -- markdash
Get off my lawn, hipsters! -- Missing Barry
by Sleepy Freud on Aug 27, 2009 12:31 PM PDT up reply actions 0 recs
I'm not a big fan of +/-...
Always appreciate stats, but +/- to me is one of those stats that I don’t look at much b/c it does not take the differences between the other 4 players each player is playing with. Scrubs can have better +/- than starters because they usually play in the game with and against other scrubs.
Check out Goallineblitz - Free Football MMORPG
Build players, Build teams, watch games...
http://goallineblitz.com/game/signup.pl?ref=4892220
by FLAxwless on Aug 27, 2009 9:43 PM PDT reply actions 0 recs
Adjusted +/- takes the strength of the opposing lineup on the floor into account.
by philthiest on Aug 27, 2009 9:53 PM PDT up reply actions 0 recs
What about the quality of teammates?
Check out Goallineblitz - Free Football MMORPG
Build players, Build teams, watch games...
http://goallineblitz.com/game/signup.pl?ref=4892220
by FLAxwless on Aug 27, 2009 10:36 PM PDT up reply actions 0 recs
Yep, it takes that into account too. A brief example:
Derek Fisher:
Raw /- +8.61
Adjusted +/- 1.63 gave most of the credit for Fisher’s time on the court (and the Lakers success) to Odom (8.81), Pau (6.64), and Kobe (5.63).
(numbers are per 100 possessions)
Adjusted +/
by philthiest on Aug 27, 2009 11:20 PM PDT up reply actions 0 recs
I see...
I’ll look into it more but just seeing the std error being so huge makes the statistics seem unreliable. But always appreciate new stats.
Check out Goallineblitz - Free Football MMORPG
Build players, Build teams, watch games...
http://goallineblitz.com/game/signup.pl?ref=4892220
by FLAxwless on Aug 28, 2009 12:08 AM PDT up reply actions 0 recs
Yeah, it’s really a mathematical estimate of a players affect on a game isolated from his teammates and opponents. Raw plus/minus carries so much noise, and APM is an attempt to filter through that noise. It’s not perfect, as the error shows, but as the top 20 in the league shows, it passes the “laugh test,” and also provides some interesting results.
by philthiest on Aug 28, 2009 12:13 AM PDT up reply actions 0 recs
it passes the "laugh test,"
That means it’s not a stupid/poorly conceived measure. It doesn’t mean that it provides any useful data beyond what you can get from basic statistics like FG%, PPG, Team wins, etc. Furthermore, when you try to use it for any sort of decision making that isn’t a no brainer already, the error pops up as on the order of magnitude of the signal.
Raw plus/minus carries so much noise, and APM is an attempt to filter through that noise.
An attempt perhaps, but not a great one. Probably in no small part due to the constant problem of sample size that we have with basketball.
PS. Sorry, I’m just tired of these fancy statistics that keep telling us that Kevin Garnett is good and that Jamal Crawford is bad. We can see it on the court and in the win column already.
The thing about baseball is that there are very few variables. There’s a pitcher and a batter. Each is either a lefty or a righty, but really, that’s about it. You have isolated occurances that are pretty free from other variables and those other situational variables are always known (men on base, 2 outs, 0-2 count). Statistics are easy to apply and useful because the lab setup is relatively clean (i.e. no noise or at least the noise is known and measurable). In basketball, it’s completely different. You’ve got constant motion, and improvisation between 10 guys who are constantly interacting with the ball and each other which constantly changes the situation. Everybody is constantly making decisions and taking actions that have drastic impacts on the statistical measurements. These are not only vast, but impossible to quantify. This is why the simplest statistics are typically the best.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 28, 2009 3:03 AM PDT up reply actions 0 recs
Just because they are not perfect does not mean they are useless...
Like you pointed out basketball is a game that is constantly changing and very difficult to quantify the value of the player in a single number. However, just because it is difficult we should not stop our pursuit of advanced basketball statistics. Even with the imperfect basketball statistics we do have there is useful knowledge to be gained at the very least in helping you find diamonds in the rough.
Also I believe that most of the cutting edge statistics (the most accurate ones) are being held secret by the NBA teams. We’ll see how the Rockets do this year as they are a pioneer in using basketball statistics. Their roster looks like crap so if they win even without Yao then they have to be onto something.
Check out Goallineblitz - Free Football MMORPG
Build players, Build teams, watch games...
http://goallineblitz.com/game/signup.pl?ref=4892220
by FLAxwless on Aug 28, 2009 9:57 AM PDT up reply actions 0 recs
ven with the imperfect basketball statistics we do have there is useful knowledge to be gained at the very least in helping you find diamonds in the rough.
For every real diamond in the rough, you’ll find that “piece of glass that looks like a diamond by APM” named Brian Cardinal.
The Rockets are major proponents of using statistical analysis, but I’d expect that other teams are on the bandwagon too… just probably not as vocal about it. We’ll see. The other thing the Rockets have on the rest of us statisticians is that they know EXACTLY what their players are supposed to do on every play, so they know who’s screwing up what. From what I’ve heard, they’ve got somewhere between 5-10 statisticians working for them. If this is the case, they can easily collect all kinds of data (percentage of time a player improvises successfully/unsuccessfully, percentage of time a player helps off his man successfully/unsuccessfully, etc.) that the rest of us don’t have access to and will never have access to either because of the sheer effort or the need to know what was supposed to happen vs. what did happen.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 29, 2009 11:58 AM PDT up reply actions 0 recs
There's always going to be more misses then swings...
There will always be more “misses” then “hits” but when you’re talking about bench players there’s not much wrong you can do unless you pull a Jerry West and overpay them. Otherwise you lose very little in taking a risk on these “diamonds in the rough” if you pay them accordingly to market value. Just because you think a player will play more then his market value, doesn’t mean you should pay him more then his market value.
Check out Goallineblitz - Free Football MMORPG
Build players, Build teams, watch games...
http://goallineblitz.com/game/signup.pl?ref=4892220
by FLAxwless on Aug 31, 2009 2:29 PM PDT up reply actions 0 recs
This is why the simplest statistics are typically the best.
Points per game is one one the simplest, most beloved, and most oft-cited statistics out there. And it would tell you that Jamal Crawford over the last two seasons has been better than Kevin Garnett.
I dunno, to me adding one more interesting weapon to the fight against regressive “traditional” hoops statistical analysis can only be a good thing.
"I’M TAKING YOU TO THE HOLE, WONDERLIC!" [turns ball over] -- markdash
Get off my lawn, hipsters! -- Missing Barry
by Sleepy Freud on Aug 28, 2009 10:03 AM PDT up reply actions 0 recs
This is why the simplest statistics are typically the best.
I didn’t say always, did I? I didn’t reference PPG as a good and useful stat, did I? Is APM the worst statistic? No, certainly not. However, what, exactly does it bring to the table?
If you had a player who played on a winning team his whole career, had good shooting percentages, rebound rates, and didn’t turn the ball over a ton, his team performed better when he was on the floor than not (by measuring on court vs. off court +/-), and never caused controversy or shot people (or himself) at night clubs, but had a bad APM, would you really bat an eye at adding him to your team or considering him a “good player?” I wouldn’t. Nor would I look at APM as a first step “litmus test” before looking at the other stuff. So, again, please tell me what APM brings to the table because I can’t see it.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 29, 2009 11:52 AM PDT up reply actions 0 recs
In basketball, it’s completely different. You’ve got constant motion, and improvisation between 10 guys who are constantly interacting with the ball and each other which constantly changes the situation. Everybody is constantly making decisions and taking actions that have drastic impacts on the statistical measurements. These are not only vast, but impossible to quantify. This is why the simplest statistics are typically the best.
This is EXACTLY why I love APM. Much like baseball can be broken into a series of discrete events (plate appearances) basketball can do this as well (possessions). Baseball IS relatively clean, with a finite set of interactions, and basketball IS noisy, with dozens of interactions happening at once. But, those “simple stats” that you refer to, points, assists, rebounds, field goal percentage, etc., only give credit to what happens at the end of the possession. They completely ignore what can be up to 24 seconds of basketball.
For a concrete example: Biedrins dunks the ball. He gets credit for two points, and it will raise his FG%. But he dunked the ball because Monta Ellis hit him with a nice pass. Ellis gets assigned an assist. But Ellis only made the pass because Biedrins was open, because Biedrins man rotated over to guard Ellis, who was set free by an Anthony Randolph screen. And this whole time, Anthony Murrow was standing at the three point line, and his presents, and because he’s already knocked down three open three-pointers, kept his man from rotating down to Biedrins. Randolph and Morrow get no credit for what happened on that play from “simple stats,” but there contributions were felt on the play.
Now, back to APM, it doesn’t care about who actually put the ball in the hoop. It’s calculated by taking a dataset that has the 10 men on the court for every possession in every NBA game, than puts that data into the Large Hadron Collider, spins it around, causing collisions and deflections, and watches as a value is assigned for every player in the league.*
Of course, basketball is a team game, and a players value is inherently tied to his teammates (and opponents). APM, however intuitively realized this. The value that it gives us is the mean of standard distribution, and we also get a standard error. Now we have a range that the player usually falls into. So, yes, there is still noise, but we can still come to conclusions about the value of a player. And, I would postulate, those conclusions are probably more in line with reality than most other metrics.
So, you say simple stats are typically best, Adjusted Plus Minus night not be the most simple stat to calculate, but it’s meaning, which boils down to how much more or less points does a team score when this player is on the floor, is about as simple a concept as you can get in basketball.
*This probably isn’t how regression analysis actually works, but it’s how I like to think of it.
by philthiest on Aug 28, 2009 8:13 PM PDT up reply actions 1 recs
That’s a great way to describe it.
I look at APM vs “Simple Stats” like I look at baseball’s ERA vs “Other Pitching Stats”.
With ERA, you validate a pitcher by measuring what essentially matters: how many earned runs does he give up? The bottom line, basically. All the other stuff (hits, WHIP, K-BB, K, HR) does not really say anything about how good a pitcher is at pitching. It tells you what kind of pitcher he is, but not necessarily how good he is. ERA does.
Similarly, APM evaluates a basketball player’s value based on what really matters: how well the team does when he is on the court, with all other players on the court factored in. All the other stats (PPG, RPG, BPG, FT%, FG%, TS%) tell you what kind of basketball player he is, but not how good he is. APM does.
"We Deserve"
by YaHeard on Aug 29, 2009 2:28 AM PDT up reply actions 0 recs
ERA is not a very good pitching stat.
by Missing Barry on Aug 29, 2009 12:20 PM PDT up reply actions 0 recs
Curious.. care to explain why?
"We Deserve"
by YaHeard on Aug 29, 2009 1:16 PM PDT up reply actions 0 recs
Mostly because it’s heavily dependent on defense, which can, and often does, have a fairly large effect on it.
by Missing Barry on Aug 29, 2009 4:26 PM PDT up reply actions 0 recs
Even beyond defense, the predictive value of ERA isn’t terribly high. It’s not awful, but looking at the “why” parts (walks/hits/homeruns/strikeouts) tend to better predict future success. WHIP, K’s, walks do say a whole lot about how good a pitcher is at pitching. Baseball, for all that people seem to marvel at how well it works with statistics, seems to have a large “luck” component. Balls hit hard but right at someone aren’t hits; lucky looping flyballs that fall are. There appears to be very, very little consistency on a pitcher’s batting average against for balls in play from season to season, even though this can have a huge influence on runs allowed.
by jae on Aug 29, 2009 10:00 PM PDT up reply actions 0 recs
Correct, though I also think you could more strongly acknowledge that one of the big factors why ERA doesn’t predict well is because of defensive factors. But yes, another one of the most influential factors from season to season is BABIP, which, while it’s ridiculous when people say it “isn’t in a pitchers control”, is largely not in a pitchers control.
On another note, I’m really looking forward to some studies on pitch/FX and batted ball profiles/BABIP. Nothing has been done on that front yet and it’s clearly one of the next big research areas in pitching so we can more fully understand what pitchers can control and influence better. Right now the research in that area is limited (which is one of the main reasons I hate FIP as a stat, though I do recognize how well it performs statistically).
by Missing Barry on Aug 29, 2009 11:42 PM PDT up reply actions 0 recs
Please explain how ERA is not simple
You take the number of runs given up, divide by the number of innings pitched, and multiply by the number of innings in a game. That’s plain and simple. APM is more like VORP. See below. See philthiest’s references before proclaiming what it is or isn’t.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 29, 2009 1:27 PM PDT up reply actions 0 recs
ERA doesn’t accurately separate the pitcher’s contribution from that of the defense. Metrics like FIP, QERA, and tRA are better. I wouldn’t say that it’s “not a very good pitching stat,” just that there are others that are better.
by philthiest on Aug 29, 2009 2:53 PM PDT up reply actions 0 recs
Sorry
When I said simple statistics, I meant things like FG% (which also judges the player’s ability to effectively decide between good/bad shot), Rebounding rates (going for a rebound is never bad), team W/L record, and regular old +/- compared when the player was on the court or off the court. I did not mean PPG or any of the other crap.
In your Biedrins dunk situation, explain how all of your influences are captured better by APM than by regular +/-. Explain how APM is better than comparing +/- from when the player is on the court vs. off the court. One thing that would be interesting is taking the +/- of players in the league, figuring out who’s playing with whom and adjust it by saying LBJ is playing with a bunch of scrubs but they still win, so he really is better than his +/- would indicate. This APM seems to include “average home court advantage” and probably some other hokey stuff.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 29, 2009 12:14 PM PDT up reply actions 0 recs
One thing that would be interesting is taking the +/- of players in the league, figuring out who’s playing with whom and adjust it by saying LBJ is playing with a bunch of scrubs but they still win, so he really is better than his +/- would indicate.
That IS the Adjusted Plus/Minus stat.
"We Deserve"
by YaHeard on Aug 29, 2009 1:15 PM PDT up reply actions 0 recs
Please actually read philthiest's references before claiming to know what APM is or isn't
It’s not what I was suggesting. It includes BS like “home court advantage.” And if it was, the measurement would NOT be in “offensive APM” and “defensive APM” because straight +/- is just a point differential.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 29, 2009 1:24 PM PDT up reply actions 0 recs
There isn’t any “home court advantage” adjustment. Where are you seeing that in any of my references?
The Offensive APM refers to how many more (or less) points (per 100 possessions) a team scores when a player is on the court than we he is off of it. Defensive APM refers to how many less (or more) points the opponent scores (per 100 possessions) when a player is on the court than when he is off of it.
by philthiest on Aug 29, 2009 3:29 PM PDT up reply actions 0 recs
http://www.82games.com/comm30.htm
It’s in the first section of the first “here”. Maybe it’s a bad link, maybe I didn’t read down to the bottom or something. :-P
Other links make it sound like what I was thinking but only explain only that “the model relies on the same basic mathematical/ statistical approach currently in widespread use by medical researchers and other scientists all over the world.”
As I said earlier, it still just confirms things that are already known and the measurement has enough noise that it’s questionable to use it as proof of anything “surprising.”
You have been DFiBrillated.
by Dubs fan in Boston on Aug 29, 2009 6:52 PM PDT up reply actions 0 recs
http://www.82games.com/comm30.htm
It’s in the first section of the first "here". Maybe it’s a bad link, maybe I didn’t read down to the bottom or something. :-P
I don’t see anything in that entire article that mentions an adjustment for home court advantage.
As I said earlier, it still just confirms things that are already known and the measurement has enough noise that it’s questionable to use it as proof of anything "surprising."
Why does scientifically proving observational evidence have no value? I would think it’s a good thing…
However, I can provide an example of where APM can lead us to conclusion that other stats wouldn’t. Let’s talk about the guy everyone on this site seems to be clamoring for, Amare.
APM was no fan of Amare last year, but than again most metrics weren’t. He didn’t get along with the Suns new coach, we was injured, he missed 29 games, etc. But last year isn’t the reason that we want to trade for Amare. The reason we want Amare is that we believe that in our system and style of play he can reproduce what he did in 07-08. Amare was a monster that year, he lead the league in TS%, was in the top ten in PPG, RPG, BPG, FG%, all the marks of a good big man. The composite stats loved him as well, 3rd in the league in PER, 6th in Dean Oliver’s ORating, 2nd in the league in Basketball-References Win Shares, and Wages of Wins had him at 16.3 Wins Produced, and .291 WP/48. A fantastic season, truly this monster will make any team better. Right?
Well, if you look at adjusted plus/minus for that year, you see a shockingly different story. Amare rates out to be +1.76. That puts him behind Andris Biedrins’s +2.33, the guy who is almost always included in trade rumors for Amare.
As I said before, most boxscore stats only look at how a play ends, and don’t give credit to the other players on the floor who helped the play get there. So when the thing a player excels at is ending plays, as Amare Stoudemire does, any metric that is based on boxscore stats is going to favor that player. Adjusted Plus Minus isn’t based on boxscore stats, it simply tried to isolate the affect of that player being on the floor vs. that player being off of it. APM estimates the Suns were 1.76 points per 100 possesions better than their opponents with Amare Stoudemire on the floor and that they were 6.77 points per 100 better with Steve Nash on the floor than when he was off it. You can come to a conclusion that Amare’s ability to finish a play was a result of Steve Nash being on the floor with him.
There’s actually a lot more on the similarities between Stoudemire and Biedrins, and the impact losing Baron Davis had on Biedrins and how moving Stoudemire away from Steve Nash will greatly decrease his affectiveness at the one thing he does really well, finishing possessions, but I’ll save that for another fan post because this comment is getting pretty long…
by philthiest on Aug 29, 2009 8:17 PM PDT up reply actions 0 recs
I don’t see anything in that entire article that mentions an adjustment for home court advantage.
Re-read it. It’s there, although I’m not sure if it’s relevant. The initial method that Rosenbaum wrote up isn’t necessarily what you see in more recent numbers. The proponents of APM have tweaked it to use more data from prior seasons, though weighing more recent seasons more heavily. Reasonable, but how much do we weight each season? How do we account for players improving at different rates?
There are multiple ways to do the ‘adjusting’. The Michael Lewis article about Daryl Morey quoted him as saying they made adjustments. Don’t take this to necessarily mean that he’s using the same “adjusted plus minus” that Rosenbaum wrote up or that appears at 82game.come or basketballvalue.com. There are other ways of weighting the adjustments, perhaps ones with better predictive value.
And the predictive value is the real issue. If adjusted plus minus varies highly from season to season (and for many players, it does) either this suggests that players’ play is so highly variable that relying on these type of data is limited or it suggests that the result of APM is capturing something external that similarly has less utility in planning for the future.
For what it’s worth, I looked at Stoudemire with and without Nash. There was not much of a difference in his scoring efficiency and indeed he demonstrated that without Nash, he found ways to score almost as effectively without being fed the ball on assisted baskets.
One of the problems with relying heavily on Stoudemire’s “APM” is that he played so many of his minutes with Nash. If two good players spend much time on the court together, the one who does better when not paired will see almost all of the benefit, the other will be penalized, even if it’s a statistical anomaly of small sample. This is accounted for to a degree in the obscene standard errors, but in general, a method that routinely produces obscene standard errors has flaws beyond those reported therein.
by jae on Aug 29, 2009 10:21 PM PDT up reply actions 0 recs
One of the problems with relying heavily on Stoudemire’s "APM" is that he played so many of his minutes with Nash. If two good players spend much time on the court together, the one who does better when not paired will see almost all of the benefit, the other will be penalized, even if it’s a statistical anomaly of small sample.
The Steve Ilardi guy mentions something like this in his “future directions” section. The problem will always be that they’re trying to make assessments of players’ impact using statistical measures of a highly variable signal without a complete understanding of where the noise comes from (thus it’s impossible to completely negate) AND using a relatively small sample size. Sure, you’ve got 10K total sample points (possessions) for LBJ and the others that play a lot, which is a decent amount, but not that big. But they’re breaking that dataset into smaller groups of datapoints (those played with Andy V, etc.) while adjusting expectations on that performance based on variables that may only have <1K datapoints (i.e. the reserves) in the entire set.
With cancer research, they have billions of datapoints on the effects of smoking, etc. With Marco Belinelli’s +/-, you’ve only got a handful. It’s hard to make a statistically relevant argument unless you have enough data to support a rigorous statistical analysis.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 30, 2009 10:37 AM PDT up reply actions 0 recs
Well, if you look at adjusted plus/minus for that year, you see a shockingly different story. Amare rates out to be +1.76. That puts him behind Andris Biedrins’s +2.33, the guy who is almost always included in trade rumors for Amare.
Fine, Andris may have been better that year. But you can’t be remotely sure because the error in Andris’ measurement is 1.62 (or possibly 65% of the signal) and the error in STAT’s measurement is 1.48 (or 85% of the signal). STAT may be somewhere around 3 while Andris may be somewhere around 0.7 or vice versa. See where the statistic goes to ish?
You have been DFiBrillated.
by Dubs fan in Boston on Aug 30, 2009 10:43 AM PDT up reply actions 0 recs
And that's only the expected error
Any flaws in the measurement strategy notwithstanding.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 30, 2009 10:44 AM PDT up reply actions 0 recs
And sorry
It was section II, it’s got the equation and then states that “b0=home court advantage” on a subsequent line explaining the equation.
You have been DFiBrillated.
by Dubs fan in Boston on Aug 30, 2009 10:45 AM PDT up reply actions 0 recs
(going for a rebound is never bad)
except for when you know your team mate will get it and you should be heading down court ? or when you know you’ll foul the guy that has better position,?? or when you know Rudolf is gonna puppyfvck you if you get anywhere near him???
Standing on the moon
Where talk is cheap and vision true
Standing on the moon
But I would rather be with you
Somewhere in San Francisco
On a back porch in July
Just looking up to heaven
At this crescent in the sky
by Skeptic con Urquell on Aug 29, 2009 4:02 PM PDT up reply actions 0 recs
D'oh
Forgot about sbnation’s strikeout syntax…
by philthiest on Aug 27, 2009 11:22 PM PDT up reply actions 0 recs
Scrubs can have better +/- than starters because they usually play in the game with and against other scrubs.
I do not believe there’s really much support for this. Very few teams make wholesale changes such that backups match up exclusively with other backups.
by jae on Aug 28, 2009 11:56 PM PDT up reply actions 0 recs
APM is my initials
soon Apm will be associated with warriors and general managment when I work their.
by adamm on Aug 31, 2009 12:56 AM PDT reply actions 0 recs
when I work their?
When you work their whats?
Standing on the moon
Where talk is cheap and vision true
Standing on the moon
But I would rather be with you
Somewhere in San Francisco
On a back porch in July
Just looking up to heaven
At this crescent in the sky
by Skeptic con Urquell on Aug 31, 2009 7:13 AM PDT reply actions 0 recs
That's a bad, bad question
You have been DFiBrillated.
by Dubs fan in Boston on Aug 31, 2009 5:59 PM PDT up reply actions 0 recs
How is Kurz bottom of Warriors?!
He was our best player
by tafkasam on Sep 2, 2009 11:54 AM PDT reply actions 0 recs
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
Certain photos copyright © 2009 by Associated Press or Getty Images. Any commercial use or distribution without the express written consent of Associated Press and Getty Images is strictly prohibited.
