 |
APBRmetrics
The statistical revolution will not be televised.
|
| View previous topic :: View next topic |
| Author |
Message |
DSMok1
Joined: 05 Aug 2009
Posts: 602
Location: Where the wind comes sweeping down the plains
|
 Posted: Fri Mar 26, 2010 1:30 pm Post subject: Adjusting Assists |
 |
|
Tom Haberstroh had an excellent article over at Hardwood Paroxysm a few days ago about "Dismantling the Assist". The basic concept was to look at each player's assists and to value them based on how much better an assisted shot from that range was than an unassisted shot.
This struck me as the wrong way to approach it. I believe the better way to attack it is to compare the value of the shot from that location to a generic unassisted shot. For dunks in particular, this makes a huge difference.
The groundwork for all this comes from 82games.com's "The Value of a Good Pass". From there it is possible to see how much value an assist at each location is worth. Unfortunately, these bins don't match the HoopData location tables for assists. I therefore had to do some math to approximate how many assists of each type were represented in the HoopData table. Approximately: 33% of "At Rim" assists were dunks (the rest were "Close Shots"), and 50% of "<10 feet" assists were "Close Shots" (the rest were 2pt Jumpers in 82Games.com parlance).
An unassisted shot has an EFG% of approximately 44% (from what I calculated). To value the assists, then: compare the potential assists' EFG% to the 44% baseline.
The new point value for potential assists to each area:
Dunk: 0.960 pts
Close Shots: 0.346 pts
2 Pt. Jumper: 0.036 pts
3 Pt. Jumper: 0.257 pts
I then summed the points added and normalized so the new, tweaked "assists" still sum to the same total across the NBA. Basically, 1 point added = 2 assists.
So here are the new, adjusted assist leaders (Ast+ is a HoopData metric; my new metric is the "Adj. Assists") (stats per 40 min, NOT pace adjusted):
| Code: | # Player Assists Ast+ Total Pts Adj. Assists Change (from Assists)
1 Steve Nash 13.4 15.0 7.0 14.0 0.6
2 Deron Williams 11.2 12.2 5.6 11.1 -0.1
3 Chris Paul 11.4 13.0 5.5 11.0 -0.4
4 Rajon Rondo 10.6 11.8 5.4 10.8 0.2
5 LeBron James 8.8 10.1 5.0 10.0 1.2
6 Baron Davis 9.4 10.2 4.7 9.4 0.0
7 Jameer Nelson 7.4 8.9 4.5 8.9 1.5
8 Russell Westbrook 9.1 10.1 4.4 8.7 -0.4
9 Jason Kidd 10.2 10.9 4.3 8.5 -1.7
10 Jose Calderon 8.6 9.6 4.1 8.2 -0.4
11 Darren Collison 8.3 9.4 4.0 7.9 -0.4
12 Devin Harris 7.7 8.3 3.9 7.8 0.1
13 Chris Duhon 7.4 8.5 3.9 7.8 0.4
14 Raymond Felton 6.5 7.4 3.8 7.5 1.0
15 Brandon Jennings 7.2 8.3 3.7 7.4 0.2
16 Gilbert Arenas 7.8 8.5 3.7 7.3 -0.5
17 Dwyane Wade 7.3 8.4 3.6 7.2 -0.1
18 Earl Watson 6.6 7.8 3.5 7.0 0.4
19 Tony Parker 7.2 8.4 3.5 7.0 -0.2
20 Manu Ginobili 6.9 7.6 3.5 7.0 0.1
21 Mike Conley 6.7 7.3 3.5 6.9 0.2
22 Andre Miller 7.1 8.0 3.4 6.8 -0.3
23 Mo Williams 5.9 6.4 3.4 6.7 0.8
24 Will Bynum 7.2 8.2 3.4 6.7 -0.5
25 Chauncey Billups 6.8 7.6 3.3 6.6 -0.2
26 T.J. Ford 6.1 7.1 3.2 6.4 0.3
27 Jarrett Jack 7.1 7.8 3.2 6.4 -0.7
28 Stephen Curry 6.2 6.9 3.2 6.3 0.1
29 Aaron Brooks 5.7 6.5 3.1 6.1 0.4
30 Larry Hughes 5.3 5.9 3.0 6.0 0.7 |
EDIT: I just found that Tom had an older post that looked at things more similarly to this (except without using the 44% baseline or using actual ASSISTED FG% for each location). Here that is: http://www.hardwoodparoxysm.com/2010/03/10/nba-hd-adjusting-how-we-measure-and-view-assists/. |
|
| Back to top |
|
 |
habetw4
Joined: 12 Nov 2009
Posts: 22
Location: CT
|
| Posted: Fri Mar 26, 2010 2:09 pm Post subject: |
|
|
I don't mean to reinvent the assist statistic because I don't think we have enough granular data, at least publicly, to do the makeover. The two articles were more of an exploration of various ways to record an assist and how that reflects the player's value as a ball distributor. There are many ways to look at the value of the assist. I hope the 82games study and two pieces gets us closer to finding the answer.
The issue essentially boils down to the chicken or the egg problem. Does the passer create the open shots for the offense or does the offense create open shots for the passer? More to the point, should we discount, says, Jason Kidd's assists because his personnel/system environment lends itself to mid range shots, which, in general, are the least efficient shots in the game? Of course not. But then again, we'd like to reward players for opening up opportunities for teammates to get higher percentage shots that they wouldn't have normally received. This is where point expectancy comes in. That, I think, is the ultimate piece to the puzzle.
Like in all player evaluation, we want to assign credit where credit is due. The contextual factor of assists creates a host of misattribution errors that makes player evaluation more difficult. At this point, I think the answer is a blend of different metrics (on/off court, player pairs, type of assist breakdowns, etc) before we are able to delve into point expectancy.
Sometimes, I think assist as a statistic is so flawed it isn't worth the effort and time. But then I feel as though I'm missing an opportunity to find inefficiencies if I ignore them all together.
_________________
I'm a twitterererer: @tomhaberstroh. |
|
| Back to top |
|
|
DSMok1
Joined: 05 Aug 2009
Posts: 602
Location: Where the wind comes sweeping down the plains
|
| Posted: Fri Mar 26, 2010 2:27 pm Post subject: |
|
|
| habetw4 wrote: | I don't mean to reinvent the assist statistic because I don't think we have enough granular data, at least publicly, to do the makeover. The two articles were more of an exploration of various ways to record an assist and how that reflects the player's value as a ball distributor. There are many ways to look at the value of the assist. I hope the 82games study and two pieces gets us closer to finding the answer.
The issue essentially boils down to the chicken or the egg problem. Does the passer create the open shots for the offense or does the offense create open shots for the passer? More to the point, should we discount, says, Jason Kidd's assists because his personnel/system environment lends itself to mid range shots, which, in general, are the least efficient shots in the game? Of course not. But then again, we'd like to reward players for opening up opportunities for teammates to get higher percentage shots that they wouldn't have normally received. This is where point expectancy comes in. That, I think, is the ultimate piece to the puzzle.
Like in all player evaluation, we want to assign credit where credit is due. The contextual factor of assists creates a host of misattribution errors that makes player evaluation more difficult. At this point, I think the answer is a blend of different metrics (on/off court, player pairs, type of assist breakdowns, etc) before we are able to delve into point expectancy.
Sometimes, I think assist as a statistic is so flawed it isn't worth the effort and time. But then I feel as though I'm missing an opportunity to find inefficiencies if I ignore them all together. |
I see what you were trying to do--more exploration than anything.
I would say that the assist is definitely a valid measure; there is a reason it is quite significant in the SPM regression. It does measure a player's impact in a certain dimension.
As for my version of "assists" and Jason Kidd: it appears that the best valuation would be somewhere in between pure assists (a success metric) vs. my adjusted assist (an efficiency metric). Similar to the difference between yards/carry in football and FirstDowns/carry. One measures "success"; the other "efficiency". True value is a synthesis of the two.
I suspect that using an adjusted version of assists such as this would allow a stronger correlation when doing an SPM-on-APM regression.
A few notable players that increased or decreased on this list:
| Code: | Player Assists Adj. Ast. Change (from Assists)
Jameer Nelson 7.4 8.9 1.5
LeBron James 8.8 10.0 1.2
Raymond Felton 6.5 7.5 1.0
Mo Williams 5.9 6.7 0.8
Vince Carter 3.9 4.7 0.8
Larry Hughes 5.3 6.0 0.7
Stephen Jackson 3.7 4.4 0.7
Carmelo Anthony 3.6 4.3 0.7
Kevin Garnett 3.6 4.3 0.7
Steve Nash 13.4 14.0 0.6
Player Assists Adj. Ast. Change (from Assists)
Jason Kidd 10.2 8.5 -1.7
Kirk Hinrich 5.5 4.3 -1.2
Tayshaun Prince 3.4 2.2 -1.2
Derrick Rose 6.4 5.5 -0.9
Rafer Alston 5.4 4.6 -0.8
Rodney Stuckey 5.6 4.9 -0.7
Jarrett Jack 7.1 6.4 -0.7
Mike Miller 4.5 3.8 -0.7
Rafer Alston 4.4 3.8 -0.6
John Salmons 2.9 2.3 -0.6 |
Interesting! |
|
| Back to top |
|
|
erivera7
Joined: 19 Jan 2009
Posts: 178
Location: Chicago, IL
|
| Posted: Sat Mar 27, 2010 4:28 pm Post subject: |
|
|
I like the idea of the "adjusted assist" ... that's certainly one metric, in theory, that could be effectively used to find out who really is a good passer vis-a-vis the shots he creates for his teammates. The assist is a flawed statistic but I don't see why it can't be improved upon so it better captures what people are looking for.
I like what I see from the list provided above, so far.
_________________
@erivera7
I cover the Orlando Magic - Magic Basketball |
|
| Back to top |
|
|
John Hollinger
Joined: 14 Feb 2005
Posts: 175
|
| Posted: Tue Mar 30, 2010 11:09 am Post subject: |
|
|
I think Tom just hit upon something very important which is why I've restrained myself from doing more in ratings to reflect 'assist quality' -- Jameer Nelson has about a hundred times more opportunity to make a pass leading to a dunk than Jason Kidd, simply because of who he plays with. (ok, slight exaggeration). If Jameer played with Caron Butler and Dirk while Kidd playe dwiht Dwight Howard I suspect the numbers would flip.
An interesting case study for the chicken-egg proposition would be to look at Nash in the two MVP years -- one with Amare, one without. Suspect he was setting up a lot more long Js the second year. |
|
| Back to top |
|
|
Ryan J. Parker
Joined: 23 Mar 2007
Posts: 708
Location: Raleigh, NC
|
| Posted: Tue Mar 30, 2010 11:13 am Post subject: |
|
|
I think Dean makes a good argument for how to value assist quality in BoP. This is one of the many things I lose sleep over (think impact on usage), but in general I think it's a good way of attacking the problem.
_________________
I am a basketball geek. |
|
| Back to top |
|
|
habetw4
Joined: 12 Nov 2009
Posts: 22
Location: CT
|
| Posted: Tue Mar 30, 2010 11:35 am Post subject: |
|
|
| Quote: | | I think Dean makes a good argument for how to value assist quality in BoP. This is one of the many things I lose sleep over (think impact on usage), but in general I think it's a good way of attacking the problem. |
His discussion in BoP was the inspiration for my assist analysis. IIRC, at its core it is a point expectancy argument. The greater the change in point expectancy, the more credit we assign to the passer. We don't have the luxury of within-play point expectancy so a lot of assumptions would have to be made.
_________________
I'm a twitterererer: @tomhaberstroh. |
|
| Back to top |
|
|
DSMok1
Joined: 05 Aug 2009
Posts: 602
Location: Where the wind comes sweeping down the plains
|
| Posted: Tue Mar 30, 2010 11:55 am Post subject: |
|
|
Good discussion, guys.
I thought my initial evaluation wasn't granular enough, so I went back and calculated an approximate unassisted and assisted FG% for each team at each location range. Estimating the overall unassisted FG% for each team gave me a baseline for each assist to each bin for each team. Now, Jason Kidd gets more credit for midrange assists; Nelson no longer gets overwhelming credit for dishing the ball to Howard.
A bye-product of this approach is that good passers on bad teams get adjusted upwards because of all the potential assists that were missed by their poor-shooting teammates. Devin Harris and Rip Hamilton are the biggest beneficiaries of this adjustment--both of their teams are about 41% estimated unassisted EFG%.
Until we know true unassisted FG% and assisted FG% for each range for each team, though, this will still just be an estimate. Perhaps Kidd's dishes to the midrange make a huge difference in Dirk's FG% for that range... maybe...
The new top 20:
| Code: | Player Assists Ast+ Total Pts Adj. Ast. Change (from Assists)
Steve Nash 13.4 15.0 5.3 13.5 0.1
Chris Paul 11.4 13.0 4.4 11.1 -0.3
Deron Williams 11.2 12.2 4.1 10.4 -0.8
LeBron James 8.8 10.1 3.9 10.0 1.2
Rajon Rondo 10.6 11.8 3.8 9.7 -0.9
Baron Davis 9.4 10.2 3.8 9.6 0.2
Jason Kidd 10.2 10.9 3.7 9.4 -0.8
Devin Harris 7.7 8.3 3.5 8.8 1.1
Russell Westbrook 9.1 10.1 3.4 8.6 -0.5
Gilbert Arenas 7.8 8.5 3.4 8.5 0.7
Brandon Jennings 7.2 8.3 3.3 8.3 1.1
Jose Calderon 8.6 9.6 3.3 8.3 -0.3
Darren Collison 8.3 9.4 3.2 8.0 -0.3
Jameer Nelson 7.4 8.9 3.0 7.7 0.3
Dwyane Wade 7.3 8.4 3.0 7.6 0.3
Raymond Felton 6.5 7.4 2.9 7.3 0.8
Andre Miller 7.1 8.0 2.8 7.0 -0.1
Earl Watson 6.6 7.8 2.7 7.0 0.4
Manu Ginobili 6.9 7.6 2.7 6.9 0.0
Mike Conley 6.7 7.3 2.7 6.8 0.1 |
And the new "biggest movers":
| Code: | Player Assists Ast+ Total Pts Adj. Ast. Change (from Assists)
LeBron James 8.8 10.1 3.9 10.0 1.2
Richard Hamilton 5.1 5.5 2.5 6.2 1.1
Brandon Jennings 7.2 8.3 3.3 8.3 1.1
Devin Harris 7.7 8.3 3.5 8.8 1.1
Raymond Felton 6.5 7.4 2.9 7.3 0.8
Carmelo Anthony 3.6 4.0 1.7 4.4 0.8
Andrew Bogut 2.4 2.8 1.3 3.2 0.8
Stephen Jackson 3.7 4.0 1.8 4.5 0.8
Mo Williams 5.9 6.4 2.6 6.7 0.8
Lou Williams 5.3 5.8 2.4 6.1 0.8
Kevin Garnett 3.6 4.1 1.7 4.3 0.7
Gilbert Arenas 7.8 8.5 3.4 8.5 0.7
Andray Blatche 2.5 2.8 1.2 3.2 0.7
Josh Smith 4.9 5.6 2.2 5.5 0.6
Andre Iguodala 5.8 6.3 2.5 6.4 0.6
Player Assists Ast+ Total Pts Adj. Ast. Change (from Assists)
Chauncey Billups 6.8 7.6 2.4 6.2 -0.6
Mehmet Okur 2.4 2.5 0.7 1.8 -0.6
Boris Diaw 4.3 4.9 1.4 3.7 -0.6
Beno Udrih 5.7 6.3 2.0 5.0 -0.7
Derrick Rose 6.4 7.1 2.3 5.7 -0.7
Kirk Hinrich 5.5 5.8 1.9 4.8 -0.7
Tony Parker 7.2 8.4 2.5 6.4 -0.8
Tyreke Evans 6.0 6.9 2.1 5.2 -0.8
Jarrett Jack 7.1 7.8 2.5 6.3 -0.8
Jason Kidd 10.2 10.9 3.7 9.4 -0.8
George Hill 3.8 4.3 1.2 3.0 -0.8
Deron Williams 11.2 12.2 4.1 10.4 -0.8
Rajon Rondo 10.6 11.8 3.8 9.7 -0.9
Paul Pierce 3.7 4.0 1.1 2.8 -0.9
Tayshaun Prince 3.4 3.8 0.9 2.3 -1.1 |
|
|
| Back to top |
|
|
Ryan J. Parker
Joined: 23 Mar 2007
Posts: 708
Location: Raleigh, NC
|
| Posted: Tue Mar 30, 2010 12:37 pm Post subject: |
|
|
| habetw4 wrote: | | Quote: | | I think Dean makes a good argument for how to value assist quality in BoP. This is one of the many things I lose sleep over (think impact on usage), but in general I think it's a good way of attacking the problem. |
His discussion in BoP was the inspiration for my assist analysis. IIRC, at its core it is a point expectancy argument. The greater the change in point expectancy, the more credit we assign to the passer. We don't have the luxury of within-play point expectancy so a lot of assumptions would have to be made. |
Yeah I knew you were doing this, just wanted to point that out to JH 
Ultimately he's right in that assist totals aren't that useful, even if you do adjust them.
_________________
I am a basketball geek. |
|
| Back to top |
|
|
Roland_Beech
Joined: 14 Jan 2005
Posts: 43
|
| Posted: Thu Apr 01, 2010 6:12 pm Post subject: |
|
|
| John Hollinger wrote: | | An interesting case study for the chicken-egg proposition would be to look at Nash in the two MVP years -- one with Amare, one without. Suspect he was setting up a lot more long Js the second year. |
um, yes:
05-06:
222 assisted 3's
307 assisted 2pt Jump shots
209 assisted close shots (not dunks)
88 assisted dunks
04-05:
284 assisted 3's
169 assisted 2pt Jump shots
206 assisted close shots (not dunks)
202 assisted dunks
... |
|
| Back to top |
|
|
bchaikin
Joined: 27 Jan 2005
Posts: 686
Location: cleveland, ohio
|
| Posted: Thu Apr 01, 2010 8:48 pm Post subject: |
|
|
A bye-product of this approach is that good passers on bad teams get adjusted upwards because of all the potential assists that were missed by their poor-shooting teammates.
fwiw you might want to rethink this line of approach - if you look at say players up and through the age of 30 (just to even out the field as many do not play past that age) over the last 15 years of the nba, a player for example like brevin knight threw for assists at a better ast/min rate than did players like jason kidd, steve nash, andre miller, and a host of others. his ast/min rate was bettered by only chris paul and deron williams...
yet the teams he played on averaged only about 30 wins a season. do you really think adjusting his assists upwards would make him in the minds of some the best "passer" in the game over the last decade and a half?... |
|
| Back to top |
|
|
BobboFitos
Joined: 21 Feb 2009
Posts: 192
Location: Cambridge, MA
|
| Posted: Thu Apr 01, 2010 11:35 pm Post subject: |
|
|
| bchaikin wrote: | A bye-product of this approach is that good passers on bad teams get adjusted upwards because of all the potential assists that were missed by their poor-shooting teammates.
fwiw you might want to rethink this line of approach - if you look at say players up and through the age of 30 (just to even out the field as many do not play past that age) over the last 15 years of the nba, a player for example like brevin knight threw for assists at a better ast/min rate than did players like jason kidd, steve nash, andre miller, and a host of others. his ast/min rate was bettered by only chris paul and deron williams...
yet the teams he played on averaged only about 30 wins a season. do you really think adjusting his assists upwards would make him in the minds of some the best "passer" in the game over the last decade and a half?... |
why is it hard to believe brevin knight *could* be one of, if not the best, pure passers in the entire game? brevin was 5'10 (optimistically speaking) without a lot of athletic ability and 0 shooting ability. He had unbelievably quick hands, but even with a great STL%, wasn't ever considered a positive on defense. (And possibly was a detriment) But he had an 11 year career playing over 18,000 minutes. Clearly he was doing *something* well, right?
_________________
-Rob |
|
| Back to top |
|
|
bchaikin
Joined: 27 Jan 2005
Posts: 686
Location: cleveland, ohio
|
| Posted: Fri Apr 02, 2010 1:16 am Post subject: |
|
|
why is it hard to believe brevin knight *could* be one of, if not the best, pure passers in the entire game?... Clearly he was doing *something* well, right?
well then it depends on what your definition of what a "pure passer" is. if its a player that throws a ton of passes and gets alot of assists simply because he throws a ton of passes, then i'd say he could be...
the fact is that in today's game 56%-57% of all FGM have an associated assist. over the past 15 years that percentage goes up to close to 60%. if you assume say just 1/10 of all FGM are clearly on one-on-one plays where a basket is scored and there was no assist (and i'm guessing on that 1/10 number), then close to 2/3 of all of the other FGM had an associated assist - i.e. something like 2/3 of all other FGM preceded by a pass had that pass recorded as an assist...
so the question becomes what proportion of those assists on those other 2/3 of FGM were due more to the passing ability of the passer rather than the scoring ability of the player receiving the pass? or...
The issue essentially boils down to the chicken or the egg problem. Does the passer create the open shots for the offense or does the offense create open shots for the passer?
however if your definition of what a "pure passer" is is not simply just someone who gets alot of assists because he throws alot of passes, but someone who either (1) gets a higher percentage of assists per pass that he throws, or (2) increases the FG% of the players he throws passes to moreso that other passers do, then there should probably be a better measure than simply the number of assists thrown for by a player...
...because there are and have been many players who throw for alot of assists simply because they throw a ton of passes, as something like 2/3 of all FGM that were preceded by a pass had that pass recorded as an assist...
again for example looking at the past 15 seasons and all those players that have (a) played at least 3000 total minutes up to and through the age of 30, and (b) have thrown for assists at a rate of (again an arbitrary number here) say 9 ast/40min, you will find not only names like chris paul, deron williams, jason kidd, and steve nash, but also names like brevin knight, robert pack, sergio rodriguez, and jamaal tinsley...
are there people out there that consider players like pack, rodriguez, and tinsley to have been some of the best "passers" over the last decade and a half? or just as players who threw for a ton of assists because they handled the ball so much, shot infrequently, and threw a ton of passes?... |
|
| Back to top |
|
|
DSMok1
Joined: 05 Aug 2009
Posts: 602
Location: Where the wind comes sweeping down the plains
|
| Posted: Fri Apr 02, 2010 7:59 am Post subject: |
|
|
| bchaikin wrote: |
are there people out there that consider players like pack, rodriguez, and tinsley to have been some of the best "passers" over the last decade and a half? or just as players who threw for a ton of assists because they handled the ball so much, shot infrequently, and threw a ton of passes?... |
Because I am looking at where the assist went to, I think the problem is resolved--there is almost no credit for an assist to a midrange jumper but there is a lot of credit for an assist to a layup. I don't care if they didn't shoot much--if their assists were going for dunks and layups, they should get some credit for it!
Interestingly, the players that get the biggest boost percentage-wise are the post players. Last year, Yao and KG both got a 30% boost because of where their assists went to. The ones hurt the most are players who just made passes to the midrange--Stuckey, Chalmers, Prince, Lee.
Oddly enough, despite the ones helped often being post players, Andrew Bynum is one hurt the most by the adjustment. |
|
| Back to top |
|
|
habetw4
Joined: 12 Nov 2009
Posts: 22
Location: CT
|
| Posted: Fri Apr 02, 2010 9:04 am Post subject: |
|
|
Are we to assume that there aren't good passers who throw a ton of passes? Are the two mutually exclusive? Surely, there are players who specialize in ball distribution just like there are those who exclusively specialize in rebounding, guarding the opponent's superstar, or defending the rim.
_________________
I'm a twitterererer: @tomhaberstroh. |
|
| Back to top |
|
|
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
Powered by phpBB © 2001, 2005 phpBB Group
|
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
