“Position-less” Basketball
The Process
In this era of basketball, point guards no longer have to be the smallest man on the court. Instead, teams can run with a 6'10" point guard by the name of Ben Simmons. Centers no longer play solely in the paint, but now the 7-foot “Splash Mountain”, Brook Lopez, can spread the floor on offense, clearing the paint for the 6'11" Giannis Antetokounmpo to slash into the paint as a point-forward. This past draft, the Cavaliers drafted a traditional point guard in the lottery for the second straight year. I’m not sure what the plan is in terms of playing time, but I’m sure both Colin Sexton and Darius Garland will share plenty of time on the court. The highest scoring bench player of all time, Lou Williams, is the size of a traditional point guard at 6'1", but he only played 25% of his possessions on the court as the Clippers’ PG. It is evident that teams no longer determine positions based on the size of the player. Instead they consider play style of each player and how they play off of their teammates.
With this said, I decided to find the different play styles through a statistical cluster analysis. Instead of focusing on the common PG, SG, SF, PF, an C positions, I want to objectively categorize the positions of “stretch big”, “3 and D”, “perimeter shooter”, etc. In order to do so, I gathered a number of stats from Basketball Reference and Cleaning the Glass, and created my own database of 2018–2019 NBA player stats (445 players). Combining the two separate databases was the most time consuming part of the process, because Cleaning the Glass’ player stats included less players, so I had to make sure to match each player correctly. Once I gathered all the stats, I carefully selected certain measures to be the basis of the cluster analysis. These stats are minutes played, field goal attempts per 36 mins, points per 36 mins, effective field goal percentage, usage rate, free throw rate, offensive rebound percentage, defensive rebound percentage, assist percentage, steal percentage, block percentage, turnover percentage, and the percent of shots at the rim, short mid-range, long mid-range, and three. These stats were chosen because they include every offensive category and every easily quantifiable defensive category. No one stat is directly correlated with another, so there is no perfect collinearity. Moreover, all of the stats are somewhat standardized so players with different usage rates can be compared.
If you don’t care about the nitty-gritty part of the process, you can probably skip this paragraph… Once all of the stats were collected and chosen, I used RapidMiner to produce an X-Means cluster analysis. I normalized all of the selected stats using a z-transformation, because they all had different scales. The variation in the scales would have caused certain stats to have greater weights than others, which was unwanted. I then performed multiple x-means cluster analysis. I finalized the model when 12 clusters were found and there was sufficient variation in the clusters.
The Results
As mentioned, 12 clusters were found. I went through each cluster to see where the players are above average and where they are below average. Here’s what I found:
Cluster 1: Quality Starters
This is one of the two categories that completely strays away from traditional positions. Players in this cluster range from point guards to centers, but they still have plenty in common. These are not your superstars, but they are still useful high usage players. They tend to produce more points and play more minutes than your average NBA player. They would not be the alpha on a championship team, but they are still necessary to compete with the best teams. These players are solid contributors and glue guys.
Example Players: CJ McCollum, Kris Middleton, Klay Thompson, Danilo Galinari, Lamarcus Aldridge
Above Average: FGA, PTS, USG%, Mins, Limiting TOs
Below Average: STL%
Nearly average in every other category
Cluster 2: Defensive Bigs
These (4) players are 4.87 standard deviations higher than the average player in BLK% and over 2 standard deviations higher in rim attempts and eFG%. For non-stat people, this means these players are off-the-chart in these categories. However, they barely take shots and are essentially not used on offense. They are solely used as rim protectors. In every tested cluster model, these 4 players were in their own cluster.
Example Players: Nerlens Noel, Mitchell Robinson, Ekpe Udoh, Robert Williams
Above Average: BLK%, Rim Freq, eFG%, ORB%, FTR
Below Average: 3 Freq, FGA, USG%, Long Mid Freq, AST%, Mins, Pts
Cluster 3: Attacking Defensive Bigs
These players are very similar to cluster 2, but they are used more on offense. Their calling card is still defense, but they have the ability to distract a defense. Their usage is higher than cluster 2, but they are still not even close to being the go-to guy. They provide high energy and are a lob-threat around the rim. They also get fouled at a high percentage (but that might be due to Hack-a-Shaq strategies).
Example Players: Ed Davis, Rudy Gobert, DeAndre Jordan, Kevon Looney
Above Average: FTR, Rim Freq, REB%, eFG%, BLK%
Below Average: 3 Freq, FGA, USG%, Long Mid Freq
Cluster 4: Pass-First Offensive Initiators
These players are the best set up men in the league. They clearly are capable of playing on the perimeter, as they above average in assists, steals, and shooting 3s. Because they have the ball the most, they are also very high on the turnover spectrum. They set up the plays, as there first option is not to score themselves (low point average). It is interesting that not every player in this cluster is a point guard. The list includes Evan Turner and Draymond Green. Teams might be able to use lineups that include those two players without a traditional point guard to exploit different match-ups.
Example Players: Lonzo Ball, Draymond Green, Kyle Lowry, Elfrid Payton, Rajon Rondo
Above Average: AST%, STL%, 3 Freq
Below Average: PTS, ORB%, eFG%, Limiting Turnovers
Cluster 5: Defensive Stretch Bigs
These players are still more valuable on the defensive side of the ball, but they have the ability to stretch the floor as well. They don’t necessary stay around the rim on offense. They have a decent mid range game and can still be a force around the rim. These bigs are harder to guard in ball screen actions, because they are a threat to roll or pop off of the screen. Their defense is not as good as cluster 2, but better than cluster 3.
Example Players: Mo Bamba, Pau Gasol, Al Horford, Serge Ibaka, Myles Turner
Above Average: BLK%, Long Mid Freq, DRB%, ORB%
Below Average: 3 Freq, STL%, MP
Cluster 6: Offensively Involved Bigs
These bigs are generally solid defensive/above average offensive players, but there are exceptions. This was the hardest category to name. This category of bigs has the highest offensive usage, but they tend to remain in the paint. These bigs have the potential to be a second or third option on offense without stretching the floor. They are still a force inside defensively, but to a less extend than the other “big” categories.
Example Players: Marvin Bagley, Clint Capela, John Collins, Andre Drummond, Montrezl Harrell, Ben Simmons
Above Average: OREB%, Rim Freq, DRB%, BLK%, eFG%
Below Average: 3 Freq, Long Mid Freq, AST%
Cluster 7: Defensive Wings/Forwards
These players tend to play on the perimeter and are not necessarily an offensive threat. They can guard multiple positions and have great length. They are definitively better on the defensive end, but not significantly above or below average in any category. Their highest z-score is STL% while their lowest is PTS.
Example Players: Kyle Anderson, Mo Harkless, Wes Iwundu, Zach Collins
Above Average: STL%, Rim Freq, FTR
Below Average: PTS, FGA, Long Mid Freq, 3 Freq
Cluster 8: Attack Guards
These players primarily play on the perimeter and have the ability to run an offense (evident by their above average AST%). If a “pass-first offensive initiator” is not in the game, these players can run the offense without the team missing a beat. They can both finish around the rim and shoot from 3. These players are high energy and play on both ends of the court.
Example Players: Jimmy Butler, De’Aaron Fox, Shai Gilgeous-Alexander, Chris Paul, Derrick Rose
Above Average: AST%, USG%, FGA, Long Mid Freq, STL%
Below Average: DRB%, ORB%, eFG%, BLK%
Cluster 9: Just Shooters
These players just shoot. They have the highest 3 Freq. Other than 3 Freq and Long Mid Freq, they are below average in every other category. If a team needs someone to space the floor and nothing else, these are the guys for it. They stay on the wing and if they have an open shot, they should take it. For a team looking for shooters, players in this category are option B.
Example Players: Channing Frye, Solomon Hill, Ryan Anderson
Above Average: 3 Freq, Long Mid Freq
Below Average: Everything else
Cluster 10: Better Shooters
These players are also highly qualified 3-point shooters, but they can do more on offense than just spot up and shoot. They tend to get more minutes than “Just Shooters” because they convert and have a better shot selection than cluster 9(according to eFG%). For a team solely looking for shooters, players in this category are option A.
Example Players: Malik Beasley, Avery Bradley, KCP, Seth Curry, Eric Gordon, Kyle Korver
Above Average: 3 Freq, Long Mid Freq, FGA, MP
Below Average: Rim Freq, ORB%, DRB%, BLK%
Cluster 11: 3-and-D
These players are great shooters and above average defenders. They tend to be bigger and longer than the other “shooter” categories. They know how to use their size to their advantage. They also have the highest eFG% out of all of the shooter categories.
Example Players: Trevor Ariza, Patrick Beverly, Robert Covington, Danny Green, Brook Lopez
Above Average: 3 Freq, eFG%, BLK%, MP
Below Average: USG%, AST%, Short Mid Freq
Cluster 12: Superstars
These players are the best in the league — plain and simple. They are above average in every category other than BLK%, ORB%, Rim Freq, and 3 Freq. In these below average categories, they are as close to average as you can get (within .26 standard deviations). These are the alphas on a championship team.
Example Players: Giannis, Steph, AD, KD, Harden, Kyrie, Lebron, Westbrook
Surprising Players: Trae Young, Lou Williams, Zach Lavine, Mike Conley
What’s Next?
This was just the beginning. Later in the summer I want to expand on this study by looking at the best lineups and rosters from this past season. Are there certain player-types that are found in the best lineups? How many players of each cluster were on the best teams? How do lineups without traditional point guards and centers perform? Once I answer these questions, my findings can help create lineup optimization and creativity along with a process to efficiently construct a roster.
This was the first publication of one of my personal studies. I plan to continue to publish different projects I have been/will be working on throughout the summer and upcoming school year. I hope you enjoy.
For more information on my model and process or for comments and suggestions, I can be reached through the following methods:
Email: bcstarks11@gmail.com
Twitter: @realbenstarks
LinkedIn: Benjamin Starks
— Ben Starks: Men’s Basketball Student Manager of Analytics for the University of Georgia
