3 and D Players, A Necessity in Today’s NBA?
What’s caused the popularity of 3 and D players? What do they add on the court?
About Writer and His Motivation for the Project:
Doyeop Kim
I’m a senior at the Wharton School studying Finance. I like to watch a lot of basketball in my free time. Hearing countlessly about Three and D players in sports media made me curious about their true value.
This analysis was done in conjunction to Data Project 2 in OIDD 245 with Professor Tambe.
The Growing Interest in Three and D Players is a Well-Documented Phenomenon
“3-and-D is simple: They excel at shooting 3s and defending. Shooting and defensive versatility are premium skills in this era, and the 3-and-D mold is becoming more of a necessity than a speciality for role players at wing and forward.”
— Kevin O’Connor
“The NBA has evolved into a game of small-ball, which uses a smaller, quicker lineup with tendencies to play sound defense but most importantly, attack by shooting from the perimeter. They have been around for some time, but their particular need has been evident over the past few years.” — Brad Washington
“For NBA teams hoping to mitigate the risk of facing mismatches and seeing their weak-link defenders targeted possession after possession, franchises will be coveting the 3-and-D archetype”
Data Analysis Part 1 — Defining and Creating a Categorical Variable for Three and D Players
Our criteria was based Bleacher Report’s Criteria in the article Breaking Down the Best 3-and-D Guys in the NBA Entering 2013–2014.
Ability to Shoot Three Pointers:
We created a categorical variable called “goodat3” which was found by Multiplying 3 Pointers Attempted and 3 Point Percentage. If the player belonged in the top 50% of this metric, they were given 1 in the categorical variable “good at 3.” Much of this analysis was carried out with Tidyverse function of filter.
Ability to Defend:
We created another categorical variable called “goodatd” which was found by first testing if the player belonged in the top 50% of defensive rating (which accounts for team defense), and then if the player belonged in the top 50% of individual defense (which was found by calculating the different between opponent’s average points per game relative to when the player was defending this player). If they belonged in 50% of both these variables, they were given 1 in the categorical variable “goodatd.” Much of this analysis was carried out with Tidyverse function of filter.
Positional Fit:
Despite a very general definition of Three and D players, centers usually are not treated as Three and D players. Therefore, players who had any minutes at Center, with the label “Center-Forward” or “Center,” were disregarded.
Three and D Variable:
If all three categorical variables above were met to be 1, the player received a 1 in the categorical variable defining that they were a Three and D player. The image below represents 10 of the 89 players who qualified.
Data Analysis Part 2 — Examining the Growing Popularity of Three and D Players by Average Age-Adjusted Salary
First, utilizing the group_by function in Tidyverse, I created four categories of players: Three and D players, Shooters who only qualified for the “goodat3” variable, Defenders who only qualified for the “goodatd” variable, and Neither which could not meet either of the conditions.
Second, because the NBA Players Association and NBA have negotiated contracts so that it is largely decided by the number of years in the league, I created an age-adjusted salary by dividing the annual salary of player in 2019–2020 season by their age subtracted by 19 (the first age of eligibility in the NBA).
The Welch Two Sample T-Test proved that there was a statistically significance increase in the average age-adjusted salary of Three and D players and their counterparts which a p-value of 0.00853.
By utilizing GGPLOT to plot the average age-adjusted salary across the four groups, you can see there is a significant difference between those who can shoot three-pointers and not. Then, there is an added boost in salary for players who can play defense in addition to shooting threes.
Data Analysis Part 3 — Examining the Growing Popularity of 3 and D players through their positions in the NBA Draft over Time.
NBA rookies are drafted through the NBA Draft that occurs annually. Players that teams value more are drafted with earlier picks in the draft.
Therefore, in our analysis, we examined the average draft positions of Three and D players over the past 10-years from 2010 to 2019, and found that there was a notable tendency to draft Three and D players earlier in the draft.
Data Analysis Part 4 — Analyzing the Value of Three and D Players Through The Importance of the “Corner Three” and the Proficiency of Three and D Players at the “Corner Three.”
Corner Three’s refer to the three-point field goals made in the right and left corners of the basketball court. Due to the fact that this is the closest three-point shot to the rim, NBA teams who have been growingly taking advantage of data analytics have been taking advantage of this shot.
Proving the Value of Corner Three Pointers
First, we created a variable of corner three points made by adding the columns of corner threes in the left corner and right corner per player. Second, we ran a linear regression against the number of wins that the team the player belonged to had. As demonstrated below, we found that there was a significantly positive correlation at a p-value of 1.55e^-11 between the number of corner threes made per game and the wins of a team.
Three and D Players Exceed at Shooting Corner Three Pointers
As you can see in the graph below, there is a notable ability of Three and D players to shoot a high percentage with corner three percentages.
Data Analysis Part 5 — Analyzing the Value of Three and D Players Through the Importance of “Catch and Shoot” Threes and the Proficiency of Three and D Players at the “Catch and Shoot” Threes.
There are countless ways to score the basketball: isolation, slashing to the rim, off a pick, cutting to the rim, and numerous others. But the connotation that most Three and D players don’t create shots for themselves, but they score after shooting perimeter shots from a pass set up by another player.
Proving the Value of the “Catch and Shoot” Threes
As demonstrated below, I ran a linear regression of the number of catch and shoot 3 pointers that players made versus the number of wins in a team. There was a significantly positive correlation as proved by the p-value of 3.76 e^-11.
Three and D Players Exceed at Shooting “Catch and Shoot” Three Pointers
Data Analysis Part 6 — Can Three and D Players Succeed Alone?
If so much of the value of Three and D players is shooting shots created by others, at what point is there too many Three and D players on a single team?
Defining the Criteria for Playmakers
Playmakers are players who have the ability to create shots for themselves and to create shots for others. Just as I created a variable based on a series of criteria for a Three and D player, I created a variable for a player who can be defined as a playmaker based on a criteria used in Bleacher Report’s Article: The Best Playmakers in the NBA, According to the Numbers.
I created one categorical variable on whether the player’s field goals made un-assisted was in the top 40% of the players in the NBA. I created a second categorical variable on whether the player’s points created off of assists was in the top 40% of the players in the NBA. Thirdly, I created a third categorical variable on whether the player’s NET Rating was in the top 40% of players in the NBA. If all of these categorical variables conditions were met, the player was labeled as a Playmaker.
Summarizing by Team
By utilizing group_by based on Team and summarise, I created a dataframe with the number of Three and D players, number of Playmakers, and the wins for each team. I created a fourth variable titled “proportion” which was the proportion of the number of Three and D players to the number of Playmakers.
Running Regression Part 1: Just Having More and More Three and D Players Hurts Your Team
As you can see in the regression below between the number of wins for a team and the number of Three and D players, there is a statistically negative correlation with a p-value of 0.00016.
Running Regression Part 2: Just Having Playmakers Doesn’t Help Either
As you can see in the linear regression below between the number of team wins and the number of Playmakers on a team, there is only a statistically insignificant positive correlation with p-value of 0.216.
Running Regressions Part 3: Teams Need a Mix with a Skew Towards Playmakers
First, I filtered out teams with no Playmakers or Three and D players. As you can see in the regression below, the proportion of Three and D players to Playmakers on a team has a statistically negative correlation with a p-value of 0.0131. This indicates that of the teams with Three and D players and Playmakers, it is better to have more playmakers than Three and D players. In other words, Three and D players are valuable role-players but cannot be core to your team.
Data Analysis Part 7— Creating a Model to Find the Ideal 3-D Player in College
Three and D players have been notoriously hard to find through the draft. As mentioned in an article by The Ringer prior to the 2017 NBA Draft, “what isn’t simple is finding them, as successes have come from random sectors of the draft.”
Therefore, in this part of my data analysis, I attempted to create a logistic regression model to project whether a player will turn into a Three and D player based on their college-level statistics.
Creating Training and Test Data Sets within NBA Data
RealGM.com, which was utilized for college-statistics, only contained data from the 2007–2008 data. Therefore, within the original NBA players data set, I created a training data from players drafted in the 2013–2019 draft (therefore players who could have played the between 2012–2019 in college). I also created test data from players drafted between the 2008–2012 Draft.
Scraping Data from RealGM.com
Unlike NBA data, which was copied and pasted into an excel sheet from NBA.com because the statistics only contained data from the 2019–2020 season, I needed college-statistics across 1500 players per year and across 11 seasons. Furthermore, the html of the website had two components I needed to alter, the page and the season. Therefore, I had to create an extensive for-loop to account for the different pages, and within, I used l-apply with RVEST to extract data on players across the 11 seasons. The college statistics were inner joined with the nba training and test data sets respectively so that there were two dataframes with nba players drafted in 2008–2012 drafts with their college statistics and nba players drafted in 2013–2019 drafts with their college statistics.
Finding an Ideal Model to Fit Accuracy and Precision
Out of all the three-pointer metrics, three pointers made had the most statistically significant correlation with whether a player will become a Three and D player based on p-value of 1.22 e^-09. Furthermore, there was an added statistically correlation between the Minutes Played Per Game of a player at p-value of 0.0362.
Based on this model, we started testing on the data set NBADATATEST. To measure accuracy, we first had to set the threshold for the “probability” needed from the model to be labeled as a prediction of a Three and D player. Initially, we set this level highly at 0.7 which gave us an accuracy of near 75%. However, we realized this threshold also gave us a precision of 1.5%. In other words, we were eliminating too many players who could become Three and D players. Therefore, we set the level at 0.4 instead which slightly compromised our accuracy to 73.7%. However, our precision went up to 42.1%. We felt that this model reflected the idea of how hard it is to scout players who will become Three and D players, but tried to best create a optimized model that will not have scouts looking everywhere in college and also mostly accurate.
Data Analysis Part 8 — Who in the 2020 NBA Draft is a top Three and D prospect?
After taking the same steps as in Data Analysis Part 7, I scraped the college-data of players in the 2019–2020 NCAA season. From using predict and the model created above, we came up with the following top 10 players to be considered order in the probability of becoming a Three and D player.
Data Sources Utilized
Player statistics on NBA players were copied and pasted into Excel from NBA.com
ESPN contained the salary of NBA players utilized in Data Analysis Part 2. Utilizing a combination of RVEST and LAPPLY, I was able to web scrape the players names and their respective salaries into a single data frame. Then, utilizing inner join, I joined this with the data frame of NBA player statistics based on the players’ names.
As described in Part 7, there was extensive web scraping necessary from RealGM.com to acquire statistics on college players. Each html was organized so there was a changing component for the “page” of players and for the “season” of the data.
Such as the following: “https://basketball.realgm.com/ncaa/stats/2018/Advanced_Stats/Qualified/All/Season/All/points/desc/1/.”
Therefore, I created a for-loop that would go through all the pages 1–15 of the college statistics of each season. Then, within the loop, I used a combination of RVEST and LAPPLY that had a range of seasons 2008–2019. Therefore, each time the loop ran, it would be extracting content from a season and a page of that season.
Thank you for Reading the Article and Here are Some Next Steps I’d Like to Proceed With:
- Further Examination into Different Player Types to Determine the Ideal Proportion of Each on a Single Basketball Team
- Looking into Three and D Players who came undrafted and through the G-League or International Leagues
