Post by eric on May 12, 2016 13:55:45 GMT -6
It's obvious that different positions take shots from different places just by looking at three point attempts, but there's a big difference between an 18 foot jumper and a dunk. How much of those do positions take, relatively speaking? I developed a way to parse pbp data with enough speed to make such observations feasible, and here we go.
To recap briefly, pbp data does not give shot distance but rather breaks shots into eleven sectors:
top of the key, right wing, left wing for three
top of the key, right wing, left wing, right baseline, left baseline, inside, layup, dunk for two
I combined all the threes into one zone and combined layup and dunk. Here were the results for a league of all 50s players, very fast, outside, starting PG SG and C scoring options:
point guard
shooting guard.............................................small forward
power forward.............................................center
Red means that position took less than the average, green means more, white means the same. PF and C are virtually identical. SF is like a PG that takes more threes instead of upper long twos. SG is an entirely different matter altogether. It behaves like a big that takes threes, but also is more likely to take wing twos than top of the key twos where the big is vice versa.
.
We can also bin by putting all threes in "three", layup dunk and inside in "close", and the rest in midrange, which gives us...
The error bars on these range from .003 to .006 depending, so we're seeing lots of actual phenomena. It's really surprising to me how different SG and SF are.
.
What I'm gonna do next is stick with the all 50s league but go PG/SF/C scoring options to see if that's what's causing the odd disparity between wings, then very slow, then balanced, then inside, then start messing with the league composition. The experiment I have set up now only cares about what position takes the shot, but next week I can make a second one that parses our league's PBP to find out how each player is shooting from each area.
To recap briefly, pbp data does not give shot distance but rather breaks shots into eleven sectors:
top of the key, right wing, left wing for three
top of the key, right wing, left wing, right baseline, left baseline, inside, layup, dunk for two
I combined all the threes into one zone and combined layup and dunk. Here were the results for a league of all 50s players, very fast, outside, starting PG SG and C scoring options:
point guard
shooting guard.............................................small forward
power forward.............................................center
Red means that position took less than the average, green means more, white means the same. PF and C are virtually identical. SF is like a PG that takes more threes instead of upper long twos. SG is an entirely different matter altogether. It behaves like a big that takes threes, but also is more likely to take wing twos than top of the key twos where the big is vice versa.
.
We can also bin by putting all threes in "three", layup dunk and inside in "close", and the rest in midrange, which gives us...
pg sg sf pf c loc
0.146 0.210 0.161 0.256 0.263 close
0.669 0.597 0.611 0.569 0.571 midrange
0.184 0.193 0.228 0.174 0.166 threeThe error bars on these range from .003 to .006 depending, so we're seeing lots of actual phenomena. It's really surprising to me how different SG and SF are.
.
What I'm gonna do next is stick with the all 50s league but go PG/SF/C scoring options to see if that's what's causing the odd disparity between wings, then very slow, then balanced, then inside, then start messing with the league composition. The experiment I have set up now only cares about what position takes the shot, but next week I can make a second one that parses our league's PBP to find out how each player is shooting from each area.
