Examining the
shot location
battle in the NBA
August 4, 2016
Pace and Space
NBA
Term coined by Miami Heat coach
Erik Spoelstra in 2012 (won
championship that season)
Result of the analytics movement
and use of player tracking analysis
in recent years
Space the floor, stretch the
opposing defense, choose shot
location based on greatest
expected value
Expected value of
shot distance: 20122013 NBA
Distance from
Basket
Field Goal
Percentage
Expected
Point Value
per shot
Percent Added
Value
0-3 ft
64.6%
1.29
+68%
3-10 ft
39.8%
0.80
+4%
10-16 ft
41.7%
0.83
+8%
16-23 ft
38.3%
0.77
--
Threes
35.9%
1.08
+40%
NBA Team Average 3-point Attempts per Game
26
24.1
24
22
20
18
16
14
12
14.7
.477
2001-2002 NBA Effective FG%
.502
2015-2016 NBA Effective FG%
All-time High
Effective FG%: Adjusts for the fact that a 3-point field
goal is worth one more point than a 2-point field goal
Does this actually
lead to winning more
games?
How signifcantly does a teams shot location, and a
teams ability to defend high value shot locations,
affect wins?
Does it
matter?
Please We
need to win
some
games.
What are the
factors to solve
this problem?
Offense: Shot selection distances and shots
made
Percentage of team shots from certain distance
ranges away from rim and the corresponding field
percentages of those shots
Defense: Allowed shot distances and shots
made
Percentage of shots allowed to opponents from
certain distance ranges away from rim and the
corresponding field goal percentages of those
201573 wins (most
wins all time)
2016
Offense
Defense
% of Field
Goals
Attempted
Field Goal %
% of
Opponent
FGA
Opponent
FG%
0-3 ft
.301
.660
.277
.639
3-10 ft
.125
.362
.177
.345
10-16 ft
.093
.423
.118
.386
16 <3pt
.119
.452
.164
.389
3-point
.362
.416
.263
.332
100%
100%
Then
what?
Evaluate these factors as potential Key
Performance Indicators and use them to build a
Win Expectancy Model using multiple regression
Use data from every NBA team over the last five
full regular seasons (82 games)
5 years, 30 teams, 21 statistics 150
observations, 3150 unique data points
Challenge
s
Too many variables = Less reliable
Multicollinearity: Percentages of shots
taken/allowed from distance ranges are
dependent on each other
Ex: Increasing rate of 3-pt shots will decrease
rate of other shots
Solution: Determine most significant distance
range for offense and defense, then build a
model using backward elimination
First Analysis
Results
Offense: 3-PT attempt rate has highest correlation to wins
(0.28)
Defense: 0-3 ft shots allowed has highest correlation to
wins (-0.38)
Notable: Defensively, allowing 0-3 ft shot attempts and 3point shot attempts has a negative correlation to wins.
Allowing mid-range shots between 3 ft and 3-point line has a
positive correlation to wins
Weird: Offensively, all field goal attempt rates besides 3pointers, including 0-3 ft shots, have a negative correlation
Force more mid-range
shots on defense!
Shoot more
threes on
offense!
Multiple Regression Analysis
Predicted Wins = 0 + 1(X1) + 2(X2) + + X(XX)
P < 0.1
R2 = .97
Predicted Wins = 0 + .239(3-PT attempt rate) + 1.348(0-3ft FG%) + .519(3-10ft FG%)
+ .534(10-16ft FG%) + .692(16-3PT FG%) + 1.586(3-PT FG%)
.551(0-3ft shots allowed rate) 1.027(0-3ft FG% allowed)
1.142(3-10ft FG% allowed) 1.423(3-PT FG% allowed)
P<
0.05
R2 = .
97
Predicted Wins = 0 + 1.570 (0-3ft FG%) + .596(3-10ft FG%) + .860(16-3PT FG%)
+ 1.682(3-PT FG%) 1.074(0-3ft FG% allowed) 1.114(3-10ft FG%
allowed)
1.805(3-PT FG% allowed)
Most predicted wins
2016 Golden State Warriors
68 predicted wins 73 actual wins
Fewest predicted
wins
2013 Charlotte Bobcats
15 predicted wins 21 actual wins
Most overachieving
2015 Chicago Bulls
50 wins, 32 predicted wins 18 residual wins
Most underachieving
Not cool.
2016 Philadephia 76ers
10 wins, 29 predicted wins -19 residual wins
Conclusions
Shooting from specific distance ranges away from the rim to
optimize expected point value per shot has a small-to-moderate
correlation to wins, but this analysis was unable to prove a
strong significance -- too many other factors in the game
Its more significant to be able to make and defend shots FG%
and FG% allowed proved to be very significant
Looking ahead: Consider other ways to evaluate shot distances
since their dependencies would affect a regression analysis.
Evaluate independent variable relationships further with VIF test.
Look at regressions of shooting statistics with entirely different
types of statistics (pace, ball movement, etc)
Thank
you
Any questions?
