WIP.
The notion of the traditional five positions in the NBA is dead.
Nominally, Steph Curry is a point guard, but does he fit the archetype of a point guard? Traditionally, the point guard initiates the offense, frequently has the ball in his hands, distributes the ball proficiently, attacks the basket, and maybe shoots from deep at an acceptable rate. Think Chris Paul, Steve Nash, Jason Kidd. Curry is distinctly different — he only sometimes initiates the offense, and frequently receives off-ball screens for catch and shoot 3s. That sounds a lot like a shooting guard to me.
Curry isn’t unique. Many players have nominal positions that don’t quite fit how they play on court.
Jabari Parker — small forward or a power forward?
Kent Bazemore — shooting guard or small forward? Is there a difference?
Kristaps Porzingis — is he a center or a power forward? Both?
Giannis Antetokounmpo — small forward, power forward, or point guard?
So, in reality, the five nominal positions are not as discrete as one may imagine. Players don’t fall neatly into these five buckets, instead they straddle multiple roles and positions. Players are often described as combo guard, combo forward, wing, big. Furthermore, a modifier to describe their specific on-court role is usually attached. Wings are called 3&D, two-way. Bigs can be rim protectors, stretch-4s, stretch-5s.
Often these labels are thrown around haphazardly — subjectively rather than objectively. Luckily, there is a ton of data available from the NBA that can help us label players objectively.
Since the 2013-14 season, the NBA has been collecting player tracking data through SportVU cameras. We now not only know the results of a player’s play (points, rebounds, assists, box score things, etc), but also how they play. Do they like to play in isolation? Pick and rolls? How often are they shooting off the dribble? Spot ups?
With this data, we can find the right clustering algorithm (hopefully unsupervised) to objectively redefine positions and roles in the NBA. Having these new position definitions opens up many opportunities to analyze per-lineup data. Which positions play well together? Which don’t?
Why is this relevant?
Helps with roster construction during free agency. Is the guy you’re going after really going to fit your team? Helps with deciding what the rotation will be like.