NBA Shot Selection Analysis

Peter Hwang, Calvin Li, Enson Soo

Introduction

Throughout the NBA's history, the types and distances of shots taken have drastically changed over time. With the introduction of the 3-point line in 1979, the NBA opened up a new opportunity for scoring. Although it took decades for players and teams to take advantage of this means of scoring, recent years seemed to have shown the merit and upside of utilizing the 3-point line. This also has become a large topic of discussion since, and an analysis could prove to be insightful on potential future trends.

In addition to understanding how the use of 3-point line has changed, we wanted to see how the shot selection or shot distance might impact a game. To guide our analysis, we broke down our tasks into 3 components:

Understanding the progression of the Three-Point Revolution
Investigating other factors that might influence shot selection
Determining if longer distance shots ultimately influence the outcome of the game

In our research into this topic, we found two reference papers that were relevant to our project. The first was Basketball Shot Types and Shot Success in Different Levels of Competitive Basketball by Frane Erčulj and Erik Štrumbelj, where they go into different shot selections made by players across different levels and leagues of basketball. Although the scope of this paper was slightly different than this project, it offered interesting insights on the variety of shot selections made and how the level and type of competition played into that. Our second paper was The Problem of Shot Selection in Basketball , written by Brian Skinner. Skinner's paper was a more relevant to our project, as he attempts to create a theoretical model on shot likelihood depending on several factors.

Data

Our analysis utilized a series of datasets compiled by another researcher scraping data from Basketball Reference. Each dataset contained all shots taken over the course of a season, from the 2000-2001 season through the 2019-2020 season, with at least 180,000 records and 18 attributes per season's dataset. Each record represents a single shot taken in a game, with attributes such as the unique game ID, the date of the game, and the play in which the shot was taken. However, for the use of our project and its visualizations, we used 11 of the 18 attributes, listed in detail in the following table:

Variable Name	Details
x	X position of the shot taken in relevance to a shot chart visualization
y	Y position of the shot taken in relevance to a shot chart visualization
winner	Winner of the game
loser	Loser of the game
time_remaining	Amount of time remaining in the quarter
quarter	Quarter in which the shot was taken
shots_by	Name of the player taking the shot
outcome	Result of the shot (made/missed)
attempt	Shot type attempted (2-pointer/3-pointer)
distance	Distance at which shot was taken (ft)
team	Team of the player shooting

Given this data, we needed to clean some of the columns to improve our analysis and make better visualizations. For example, we changed the 'distance' column from '11ft' format to just a numeric value and filtered out some data inconsistencies differing from the play of the game.

Overview of Data: Shots Attempted by Team

The following figure is a bar plot displaying the number of shots attempted by each team during the 2019-2020 NBA season. Overall, most teams had a total of 6,000 shots during this season, with Dallas, Portland, and Milwaukee having the highest number of shots attempted.

Shot Chart Comparison (2000 vs 2019)

The following figure is an interactive side-by-side shot chart comparison of teams from the 2000 and 2019 NBA seasons. For any selected teams, we can observe that the shot location distribution has drastically changed, when comparing the 2000 and 2019 NBA seasons.

We may consider the New York Knicks as an example. In the 2000 season, most of the shots taken were mostly equally distributed across the court, with a slightly higher density of mid-range shots and shots near the basket. In contrast, most of the shots taken in the 2019 season were mostly clustered either outside the 3-point arc or near the basket, and less mid-range shots. When looking at other teams this trend is similar. Therefore, this indicates that over the past 2 decades, NBA players have transitioned to shooting more 3-pointers and less mid-range shots. The change in strategy may be due to the fact that mid-range shots are only worth 2 points, which would make them less efficient than 3-pointers.

Shot Distribution for New York Knicks

In our side by side bar plot, we decided to examine the New York Knicks because there is a wide variety of players who play different positions. It also provided more insights into outliers that exist in terms of three point shots attempted. The side by side bar plot allows us to make a better comparison and understand the skew of shots attempted for both categories.

The bar plot demonstrated that almost every player contributes to shooting three pointers. One insight is that the five players with the most three pointers: Julius Randle, RJ Barrett, Bobby Portis, Marcus Morris, and Kevin Knox are all forwards. Forwards, unlike guards, traditionally only take midrange shots, layups, or dunks. Therefore, irregardless of position, every player takes three pointers. This is especially apparent when the offense of the team primarily runs through a specific player. However, there will always be outliers such as Mitchell Robinson who has taken zero three point shots. Ultimately his role as a center is to grab rebounds and make defensive plays, and offense isn't their primary role.

Shot Distances Over Time

To see how the shot distances have changed over time, we can take a look into how each team's average shot distance of a season has changed from 2001 to 2019. The following figure is an interactive line plot that allows users to view the average shot distance trend for a selected team. Generally, for each team, it is clear that over the past two decades, shots have are taken further and further away from the basket on average. A great case of this is the Golden State Warriors, who have been a frontrunner in revolutionizing the 3-point game in the mid to late 2010s.

Distribution of 3-pointers: Time Remaining in Game

We also wanted to explore the potential factors that influence shot selection. The following histogram represents the distribution of the number of 3-pointers taken based on the time remaining in the game. Each color of the histogram bars represents a quarter of the game, and each bar represents a minute in the game.

We can observe that there are peaks at the end of each quarter, with the highest peaks being at the last minute of the 3rd and 4th (final) quarter. This indicates that players tend to shoot more 3-pointers at the end of each quarter, with the most 3-pointers taken towards the end of the game. In contrast, we can also observe that there are less 3-pointers taken towards the beginning of the game and during the beginning of each quarter. This is most likely due to the players facing less time pressure to score. However, towards the end of each quarter, players may feel compelled to shoot more 3-pointers because this is a strategy to maximize points before the quarter/game ends. Furthermore, a successful 3-pointer can significantly impact the score, especially towards the end of the game or if the shot is game-winning. Therefore, these observations suggest that the time remaining in each quarter and each game is a factor that influences a player's shot selection.

Understanding Shot Distance: Wins vs Losses

Initially we aggregated all of the teams' wins and teams' losses and eventually compared the shots made or missed. The following figure is a side-by-side box plot that compares the distribution of shot distances based on whether teams win or lose and between made and missed shots.

There is a negligible difference between both distributions. The slight difference is that the interquartile range for when teams win is slightly larger for shots made, albeit only one foot farther for the third quartile. Therefore, shot distance is a poor predictor of winning since all shot distances average out and the distributions are essentially the same for when teams win or lose.

Shot Types for Best and Worst Teams

We created a side-by-side bar plot comparing the Milwaukee Bucks and the Cleveland Cavaliers, the best and worst teams in the Eastern Conference, respectively. This comparison was an exploration on how taking 2-pointers vs. 3-pointers may contribute to wins for a team. Upon analyzing this visualization, although the percentage of made 3-pointers was fairly similar between both teams at around 35%, the Bucks made a significantly higher number of 3-pointers than the Cavaliers. Although it is not a significant indicator of contributing to games won, this could be a fair determinant of winning.

Summary and Additional Work

Overall we conducted a study to understand how the NBA has transformed in terms of the shot selection. We accomplish this in several ways by comparing the shot charts between 2000 and 2019. We also used a line chart to identify a trend in average shot distance for all teams.

Another finding is that at the end of each quarter, players tend to opt to take more three pointers. We use a histogram to visualize the distribution. Our last conclusion is that there is an almost equal distribution for shot distances between when teams win and lose. Therefore shot distance is a poor determinant of winning. However, when comparing the best team and the worst team in 2019, the team with the best record has more attempted and made 3-pointers. Therefore, there is some correlation in that aspect.

There are many tangents that we can take in future work. For instance, we can examine other factors influencing shot selection such as if a team is trailing or leading. Teams may opt to take more three pointers if they are trailing behind the opposing team. Also, there are various scenarios and circumstances such as the type of play: fast-break, defensive rebound, shot-clock running out, etc. This can severely alter shot selection as defense might be more pressing and driving to the basket might be difficult.

Other questions that we didn't answer is if there are wider margins of winning if the team takes more three pointers. Ultimately, the goal of any sport is to win each game. Another aspect is to see if a specific cluster or area of the shot chart has a higher shot accuracy. Identifying a trend like this can help players and coaching staff understand where they should practice their shots more.

Github Repo Link