Offense and defense ratings adjusted for position bias

Introduction

A few weeks ago, I took a look at players’ offensive and defensive ratings accounting for team performance as a whole. After doing so, I noticed that players’ performance seemed to correlate greatly with their positions, particularly on defense. I didn’t give this much thought at the time, but started thinking about it more in depth over the weekend.

Here are the questions I asked: What if some defensive and offensive ratings are biased towards certain positions? It would make sense that a big tall guy standing near the rim is more impactful on defense than a small guy on the perimeter. But does that necessarily mean he’s a better defender, or just plays a more important defensive role? And, if the latter, how can we more effectively compare the defensive performance of two players at two different positions?

Preliminary analysis and measures of interest

I started by collecting the data of all players who have played at least 25 games and at least 22 minutes per game. Then I disaggregated the data by position. Here are the results:

Position	Count	Average DRtg	Average ORtg	Net
C	23	102.35	108.91	6.57
PF	37	104.30	107.84	3.54
PG	39	106.90	106.36	-0.54
SF	43	106.12	105.30	-0.81
SG	42	107.71	107.21	-0.50

These results are hugely important. Just by virtue of their position, centers automatically look amazing compared to other players. This is true not only defensively, but also offensively (possibly due to offensive rebounding presence?). In fact, the average center has a net rating of 6.57. Power forwards also receive a bump in their stats due to their position/role. The other three positions are actually a net negative.

These results beg us to ask, if a center or a power forward has a good looking rating, is he actually good, or just playing a position that the stats bias positively?? Conversely, if a guard/wing has a bad looking rating, is he actually bad, or just playing a position that the stats bias negatively? And how the heck can we compare players across positions given these biases?

My solution is to create two stats, adjusted defensive rating and adjusted offensive rating. These two stats are calculated by subtracting a player’s individual rating by the average rating of the player’s position. I do this in order to eliminate the position bias inherent within these stats. In effect, it creates a measure similar to baseball’s WARP (wins above replacement player). Players’ performance are measured relative to an average player at their positions.

Technical information

The above chart shows all players’ adjusted defensive ratings on the x-axis and adjusted offensive ratings on the y-axis. Filters on the top left allow you to filter the chart by team and position. A tab on the upper left allows you to switch to a table view, where you can then sort all the players by their adjusted defensive and offensive ratings, and their adjusted net ratings (adjusted offense minus adjusted defense). Remember, for defense, negative numbers are better.

Limitations

All data is unweighted. Number of games and minutes played are not considered in players’ ratings or averages.

Observations

Oh man, where to even start. I guess let’s begin with some boring but important things. First, centers are the most valuable position. In particular, centers who do “centery” things have the most valuable roles. Just having them on the floor, even if they’re average at what they do, is a huge boon to team performance. The “stretch 5s” like Bargnani and Pau Gasol don’t pass the measurement test. That the Houston Rockets have had injured big men the whole season might be even more important than initially thought.

Second, the range of offensive performance (-20 to +20) is twice that of defensive performance (-12 to +8). So a player can be good or bad on defense, but can be REALLY GOOD or REALLY BAD on offense.  For example, a team receives a 20 point defensive bump per 100 possessions by upgrading from Andrea Bargnani to Roy Hibbert, but a 40 point offensive bump by upgrading from Jeff Taylor to Kevin Durant.

The best defender in the league is now Paul George. The best offensive player in the league is still Kevin Durant. Durant is also the best net player in the league. Durant also might be a golden god.

Dwight Howard suddenly looks much less impressive now that his performance has been adjusted. He’s a little above average defensively (though still the best on the team) and is below average offensively. James Harden, on the other hand, is redeemed a bit. He’s one of two players who are above average defensively and offensively, with Terrence Jones being the other. And, no matter how one slices and dices the data, Patrick Beverley’s supposed defensive prowess still does not show up. I’m starting to think that the critics who claim that his defense is more showy than effective are correct. Unfortunately for the PG position, Jeremy Lin is just as bad.

In conclusion:

 

View this discussion from the forum.