I’ve been thinking a lot recently about “fit” when it comes to building a basketball team.
Too often, statheads are guilty of plugging in players’ metrics and expecting a certain outcome without giving a thought to how the skills of each player mesh together. Dean Oliver’s “Skill Curves” are a simple way of modeling these teammate interactions, but they often fall short because Possession % is not always an adequate explanation for how a player produced points at a given efficiency.
A better way to look at fit would be to grade players’ skill sets and see how certain combinations of skills predict lineup efficiency. That’s a project for a later date, but today I thought I’d take a stab at generating ratings for players based on their skills. Here’s the process:
Generate projections for each player. In this case, I used Basketball-Reference’s Simple Projection System and regressed to positional means instead of league means. I also left out the age adjustment, since I want to estimate player skills in year Y, not year Y+1.
Decide which skills to measure, and how to measure them. I settled on the following “tools” as base skill sets for an NBA player:
Shooting - Pure shooting ability; measured by Free Throw Percentage
Scoring - Ability to create & make shots; measured by Points per 36 minutes
Defense - Size + blocks, steals, & other stats that predict defensive +/-; measured by Defensive SPM
Rebounding - Ability to grab missed shots; measured by Total Rebounds per 36 minutes
Determine a grading scale. First, I calculated each player’s percentile rank (among players at the same position that season) in each category. Then, in true scouting tradition, I set up a 2-8 scale for each skill rating, corresponding to the following percentiles:
2 = 0-3%
3 = 4-13%
4 = 14-35%
5 = 36-65%
6 = 66-87%
7 = 88-97%
8 = 98-100%
The average player at a position will have a rating of 5 in a given skill. Remember, players are rated relative to others at the same position, so a 5 Floor Game point guard is still likely to be a better passer/ballhandler than an 8 Floor Game center.
The most “complete” players (highest simple sum of ratings) in the dataset seem to be Larry Bird and Kevin Garnett. In 2005, KG was an 8 (relative to other PFs) in Scoring, Floor Game, Rebounding, and Defense, with a 7 in Shooting, for a total of 39. Bird actually had three separate 39 seasons — 1984, 1985, and 1989; in the former two he was 8 across the board except in Scoring, and in the latter he pushed his Scoring to 8 (with his Defense falling to 7)… Chris Paul’s 2009 and 2010 seasons were phenomenally complete, as he posted 8s in Scoring, Floor Game, & Defense, to go with 7s in Shooting & Rebounding… LeBron James is known for his all-around talent, which is borne out with 8s in Scoring/Floor Game and 7s in Rebounding/Defense. But the well-known hole in his game is also exposed here: he’s merely a 5 Shooter… Kobe Bryant’s complete skills seem to have peaked in 2004, when he was an 8 Scorer, 7 Shooter/Rebounder, and 6 in Floor Game/Defense. In 2011 he remained an 8 Scorer (as he’s been every year since 2002), but his Shooting/Rebounding have fallen to 6s, and his Floor Game & Defense only score 5s… Arvydas Sabonis was a surprisingly great all-around player. Even at ages 34-35 in 1999 & 2000, he was an 8 Shooter relative to other centers, with 7s across the board in the other categories… During the prime of his first stint in the league, Michael Jordan was consistently an 8 Scorer/6 Shooter/7 Floor Game/8 Defender, with his rebounding varying between 7 & 8. When he returned from retirement, MJ’s Floor Game & Defense fell to 6, and in the twilight of his Wizards days he was a 7 Scorer/5-6 Shooter/6-7 Floor Game/7 Rebounder/5-6 Defender.