APBRmetrics

Thanks for adressing the concerns, Kirk.

KirkG wrote: In response to Mystic, I like your range/spread idea. I have been thinking about that a lot. The top 3 guys in that metric are: Nash (.62), Calderon (.57), and Ray Allen (.56); it's really quite amazing how far Nash is ahead of everyone else, but what does this metric really tell us?

As I said I would interpret it as the ability of players to find the right spot for the shot rather than just pulling the trigger whenever they feel it is ok. I would also try to use a dynamic coefficient in order to determine whether the respective region is a plus or minus for the player. Different zones should have different average efficiency values. Well, the latter might be only be useful under the assumption that each spot on the court is equally important/good in order to take a shot. It is not unlikely that this isn't the case. kjb pointed out that there seems to be a pattern in which players are taking shots.

Imho, there has to be a way to seperate between efficiency via shot selection and via being a good shooter. The way I see it, the results are not caused by being a good shooter alone. For example, Calderon is picking his shots more and is not willing to take tough shots, that will increase his own shooting efficiency, but will not help the team. Nash has games in which he is prone to that too. Well, I would propose that a shooter is not taking enough shots, if the team eFG% with him on the court is lower than his own eFG%. But well, the shot distribution might also be heavily influenced by team chemistry, some players seems to be able to live with less shots better than others.

Kirk, it would also be interesting to see the list of the players (over 20-25 minutes per game) who have the most area below .9. .85 or .8 pts per possession. It would also be interesting to see which players have the best an d worst spots maybe 3' * 3' or bigger. 82 games had that in the past.

Hi,

My analysis is not nearly as thorough as Professor Goldberry's, so please take these with a grain of salt, but I have some preliminary results:

With a box size of 3 x 3 ft, I have found the top and bottom 10 spots for players with at least 50 attempts from one spot for four different intervals: 0-4, 4-15,15-22,22-50. The columns are playerID, player name, number of attempts from the spot, efficiency.

1. 0 - 4
Best:
395,LeBron James,148,0.898649
625,Tim Duncan,51,0.803922
43,Antawn Jamison,111,0.801802
84,Brendan Haywood,84,0.773810
101,Caron Butler,83,0.771084
233,Fabricio Oberto,262,0.763359
630,Tony Parker,72,0.750000
532,Reggie Williams,78,0.743590
210,Dwight Howard,1993,0.736076
584,Shaquille O'Neal,1148,0.735192
Worst:
420,Malik Rose,128,0.421875
121,Chris Mihm,57,0.421053
123,Chris Quinn,114,0.421053
497,P.J. Brown,142,0.415493
86,Brevin Knight,112,0.410714
189,Derek Fisher,402,0.410448
151,Damon Stoudamire,106,0.405660
208,Drew Gooden,52,0.403846
442,Marvin Williams,63,0.396825
595,Speedy Claxton,53,0.320755
101,Caron Butler,79,0.303797

2. 4 -15
Best:
672,Zach Randolph,60,0.700000
639,Troy Murphy,74,0.662162
371,Kevin Garnett,51,0.647059
164,Darko Milicic,59,0.644068
119,Chris Kaman,83,0.626506
371,Kevin Garnett,53,0.622642
630,Tony Parker,158,0.620253
625,Tim Duncan,120,0.616667
668,Yao Ming,77,0.610390
13,Al Jefferson,64,0.609375
Worst:
451,Mehmet Okur,57,0.315789
371,Kevin Garnett,54,0.314815
164,Darko Milicic,75,0.306667
194,Desmond Mason,55,0.290909
668,Yao Ming,55,0.290909
210,Dwight Howard,90,0.288889
13,Al Jefferson,58,0.275862
559,Rudy Gay,51,0.274510
100,Carmelo Anthony,62,0.274194
348,Josh Smith,57,0.263158
672,Zach Randolph,64,0.218750

3. 15 - 22
Best:
442,Marvin Williams,63,0.619048
98,Carlos Boozer,62,0.612903
208,Drew Gooden,62,0.612903
371,Kevin Garnett,52,0.596154
646,Udonis Haslem,54,0.592593
200,Dirk Nowitzki,60,0.583333
200,Dirk Nowitzki,60,0.583333
200,Dirk Nowitzki,54,0.574074
176,David West,67,0.567164
324,Joe Johnson,60,0.566667
Worst:
534,Richard Hamilton,66,0.318182
275,Jamal Crawford,63,0.317460
100,Carmelo Anthony,54,0.314815
370,Kevin Durant,53,0.301887
61,Baron Davis,51,0.294118
370,Kevin Durant,55,0.290909
414,Luol Deng,56,0.285714
476,Monta Ellis,68,0.279412
414,Luol Deng,58,0.275862
395,LeBron James,50,0.260000
414,Luol Deng,50,0.240000

4. 22-50
Best:
47,Anthony Parker,64,0.640625
639,Troy Murphy,55,0.636364
122,Chris Paul,53,0.584906
605,Steve Nash,65,0.553846
266,J.R. Smith,52,0.538462
38,Andres Nocioni,71,0.535211
380,Kobe Bryant,59,0.525424
603,Steve Blake,97,0.515464
475,Mo Williams,51,0.509804
373,Kevin Martin,53,0.509434
Worst:
302,Jason Terry,51,0.274510
516,Rafer Alston,70,0.271429
61,Baron Davis,59,0.271186
527,Rasual Butler,63,0.269841
380,Kobe Bryant,69,0.260870
159,Danny Granger,60,0.250000
516,Rafer Alston,55,0.236364
43,Antawn Jamison,56,0.232143
298,Jason Kidd,52,0.230769
582,Shane Battier,70,0.214286
524,Rashard Lewis,64,0.187500

Thank you,
Michael

Thanks.

Jefferson, Randolph and Garnett each have a 3*3 spot apparently within the 4-15 foot zone where they rank on the best and worst lists. Might need to refine their sense of of their range? Too late I am sure.

Somewhat surprised Durant has 2 spots on the worst list from 15-22 feet.

Hi,

Number of 3x3 squares (with at least 50 shots) with points per possession below 0.9, 0.85, and 0.8. The columns are playerID, player name, average number of shots in the squares, average efficiency, and the number of squares.

1. 0.9

13,Al Jefferson,7.478571e+01,0.388707,14
380,Kobe Bryant,6.733333e+01,0.400433,12
100,Carmelo Anthony,7.627273e+01,0.379670,11
176,David West,6.872727e+01,0.391886,11
200,Dirk Nowitzki,6.572727e+01,0.413501,11
390,LaMarcus Aldridge,6.109091e+01,0.412619,11
370,Kevin Durant,5.522222e+01,0.369062,9
210,Dwight Howard,9.987500e+01,0.411032,8
534,Richard Hamilton,6.737500e+01,0.386817,8
625,Tim Duncan,7.775000e+01,0.392953,8

2. 0.85
13,Al Jefferson,7.276923e+01,0.385859,13
100,Carmelo Anthony,7.544444e+01,0.365305,9
176,David West,6.237500e+01,0.375856,8
380,Kobe Bryant,7.150000e+01,0.380214,8
625,Tim Duncan,7.775000e+01,0.392953,8
200,Dirk Nowitzki,6.142857e+01,0.400192,7
370,Kevin Durant,5.585714e+01,0.347876,7
534,Richard Hamilton,6.857143e+01,0.379123,7
43,Antawn Jamison,6.633333e+01,0.349401,6
211,Dwyane Wade,7.983333e+01,0.382381,6

3. 0.8

13,Al Jefferson,6.028571e+01,0.363992,7
100,Carmelo Anthony,8.283333e+01,0.344368,6
370,Kevin Durant,5.650000e+01,0.335342,6
476,Monta Ellis,6.483333e+01,0.354375,6
534,Richard Hamilton,6.533333e+01,0.374128,6
43,Antawn Jamison,6.120000e+01,0.334498,5
176,David West,5.980000e+01,0.358278,5
390,LaMarcus Aldridge,55,0.385384,5
164,Darko Milicic,6.175000e+01,0.325064,4
211,Dwyane Wade,7.350000e+01,0.368756,4

Thank you,
Michael

You are fast. Thanks again.

Jefferson heads the worst list on all 3 cuts using # of bad squares.

Kobe 2nd worst on the .9 cut, 4th on .85 but not top 10 on the .8 cut.

Anthony with an average rank of 2.3 on the worst list.

Durant on all three worst lists. Dirk the first two.

Guys taking or have to take a heavy usage role apparently often heaving up low efficiency shots from somewhere.

Hi,

I have spent some time thinking about how Professor Goldsberry creates his metric for how to evaluate shooting range and such. To me, it could make the most sense to cluster different player's pockets of shots together. I think these clusters should be adaptive to different players, as they will tend to change from player to player. For this reason, I have begun by creating a plot (similar to Professor Goldberry's) of Steve Nash's number of shots for different points on the floor.



I then ran a simple clustering algorithm to find clusters of shots (demanding that at least seven shots have been taken at a point to count towards a cluster).



I then plotted the points per attempt for the spots within one of these clusters.



I am wondering if there is anything to be done with this method in terms of classifying efficiency or looking at how a player does against certain teams, which will have their own points per attempt distribution. Maybe one can correlate player performance against certain teams depending on what their defensive efficiency is at the player's favorite shooting locations.

Thank you,
Michael

Clustering shots is a useful simplification of the discrete data. I think it would be easier to get the main message. With all the individual shots separately marked I am not sure how well people (statisticians, coaches, players, etc.) would process the message of the data for a player or in comparison to others.

Graphics can sometimes be useful but the usefulness of a graphic should always be checked.
Sometimes a table or a spreadsheet is quicker and / or more useful than a chart or a whole much of charts isolating each and every trend, IMO. But upper management seems to often like to look at charts instead of greater detail.

Zero comments about the conference from those who attended?

Kirk,

I'm currently trying to create a shot chart like the one in your paper. Do you have some tips on how to layer the color-coded squares onto a shot chart? Right now, I'm using Python but advice for any programming language would be appreciated. Thanks.

Hi,

Quick question: do people have thoughts about a good metric for determining the best shooter in the NBA? Simply range/spread seems to get rid of (some) information about efficiency with cutting it as simply asking if a location is below or above 1.

Michael

Some range is nice but I not sure how much indirect value elite range has. Maybe an analysis of player range vs lineup +/- would aid the consideration of the importance of range. Cumulative lineup range. star range, multiple player range and range diversity.

I listened to the Basketball Analytics panel a few weeks ago and it seemed pretty much like a replay of all the previous ones- the panelists saying very little beyond the surface. If they can't / won't say much new, I'd suggest the conference go to a new generation of speakers and perhaps mainly folks not currently working for teams. Maybe draw upon the list of past research papers presenters and perhaps a few folks from the blogosphere. It would give up "big names" for deeper, freer discussion.

It could potentially be a meetup for authors of different player and team evaluation metrics (including perhaps Wins Produced). As much as folks like Dean Oliver and Mike Zarren downplay their value to or relative use by insiders, there are a lot of ways to use these meaningfully as a base level of information: to find rough estimates of value; to find league averages; to construct age curves; to track year to year change; to find degree of rarity on performance levels by position or player type; to see what the very best are giving extra compared to your guy or the league average; to determine which stat-lines are being over and under bid for, who is doing the over-bidding, for what and whether it makes sense in their overall context; etc.