In the previous post we established that the FG% depends on the closest defender's distance, the shot distance and whether the player dribbled or not before the shot. In this post I'm going to use a shot quality map to visualize the shot selection of certain players.
First, let's start with a map of the eFG% as a function of the defender distance and the shoot distance:
The eFG% is divided into five categories: good (>54.5), above average (51.5-54.5), average (48.5-51.5), below average (45.5-48.5) and bad (<45.5). The exact eFG% is noted on the pixel. Shots between 22 feet - 24 feet can either be a long 2 or a corner 3. I choose the eFG% of corner 3s for that distance since it is a much more common shot. We can see that "good" shots are mainly close to the basket and open 3s. Long 2s, between 16 feet - 24 feet, are "bad" even when the shot is wide open. A highly contestant shot (defender closer than 3 feet) is "bad" unless the shot is taken right under the basket. Midrage jumpers (8 feet - 16 feet) are also "bad" unless the shot is wide open. For wide open midrange shots the data is a little noisy, due lower number of shots, but is in the average-above average range.
Now we want to create a shot quality map that contains information about the number of shots taken at a certain distance and a certain defender's distance. The x axis is the shot distance, the y axis is the closest defenders and the area enclosed between lines of different color contains the number of shots taken. This is called a contour plot (more information about contour plots). If you are old school and used to read topographical maps these plots will look familiar.
Reminder on how to read these plots: defender's distance above 8 feet means wide open shot. Top left are wide open layups, bottom right contestant layups, top right wide open threes and bottom right contestant threes. Corner 3s are at 22 feet and the 3 point arc is at 24 feet.
For example, below is a shot quality map for all the catch and shot shots taken since the 2013 season:
There are two main clusters - close to the basket where the majority of the shots are taken about 3 feet away from the basket with a defender's distance of about 2.5 feet, three point shots with the peak value at shot distance of 24 feet and a defender's distance of 5 feet. Barely any shots are taken from midrange (8 feet-16 feet). There is a secondary peak at 22 feet mostly because of corner threes.
If we look at the shot quality map for shots taken off the dribble since 2013 we get a different distribution:
The vast majority of shots are taken close to the basket with the most frequent shot appearing at a distance of 5 feet with a defender's distance of about 2 feet. There are much more (relative) midrange jumpers compared to catch and shot. Three point shots are still fairly popular but the secondary peak at 22 feet is gone since it is rather difficult to take a corner 3 off the dribble.
We can look at all of the shots in the NBA and check the typical defender distance for a particular shot distance:
The red line represents the most frequent closest defender distance for a particular distance. This is called the mode in statistics. The blue area represents the middle 50% of the shots (between the 25th to the 75th percentiles). The obvious trend is that the closer the shot distance the closer the defender is to the shooter.
Shot Quality Maps - Individual Players:
In order to be able to compare an individual player to the rest of the NBA I'm overlaying the previous plot, of the mode and the middle 50 percent of shots, with the individual shot quality maps. As a quick reference compare the contours to the red line which represents the most frequent defender distance for a particular shot distance.
DeAndre Jordan takes most of his shots from very (very) close range with the most frequent defender distance between 2-3 feet. This map shows that his shooting range is within 6 feet with most of his shots taken at a defender distance slightly higher than most NBA players.
Stephen Curry takes most of his shots from around 24 feet with a defender's distance of about 4 feet. We can see that this shot is considered harder than the majority of the NBA shots from that distance (most of the shots are under the red line which represents the mode). We can also see that his range extends all the way back to 30 feet. Insane!
Kobe Bryant likes to take contestant shots as can clearly be seen in the map. His favorite spots are 3s from 25 feet out and long 2s from 16-18 feet. We can see that Kobe takes most of his shots when the defender is much closer compared to the rest of the NBA. Is this because he is the Lakers main threat and the defense can stay focused on him or is it bad shot selection? You decide.
We can also explore differences between different seasons. Let's take Chris Bosh before and after LeBron:
Bosh two favorite spots in 2013-14 were close to the basket and 24 feet away. We can see that most of the 3s are taken when the defender is fairly far away (most of the shots are above the red line). Also, his long 2s are also relatively contestant.
Last season (2014-15) Bosh took considerably less shots close to the basket. Another notable difference is that his 3s were defended a little closer compared to the previous season. This makes sense since LeBron is not there to draw attention but why did he stop shooting close to the basket?
Conclusions:
- The shot quality map is a quick way to get an idea of the shot selection of an individual player or a team. This is a much faster way to interpret the data compared to a typical table.
- Kobe takes the most difficult shots in the NBA.
- DeAndre Jordan takes the easiest shots in the NBA.
Contour vs. heatmap
Both a contour plot and a heatmap can be plotted to represent the same data. Here is an example:
which one do you prefer?
Do you enjoy this blog? If you enjoy this blog please let me know. Please comment if you have any thoughts. If there is enough interest in this blog, I can convert it into a website with some of the statistics updating on a regular basis.
Follow new posts on twitter: @eyal_shafran
Technical details:
- why did I choose the mode and not the median? When plotting a heatmap or contour plot our eye focuses on the "hotspots" which are equivalent to the mode. Therefore, for a proper comparison of the contour\heatmap plots to the rest of the NBA the mode needs to be plotted and not the median.
- The raw data is in scatter format. I used kernel density estimation with a Gaussian kernel in order to convert the data to the plotted format.