Defense Doesn’t Pay in the NBA

Earlier this month ESPN unveiled a new advanced basketball metric called “real plus-minus” (RPM). The statistic was developed by Jeremias Engelmann and is trumpeted as an improvement over its predecessors, adjusted plus-minus and real adjusted plus-minus. One of the unique features of RPM is that it splits a player’s impact into offensive RPM (ORPM) and defensive RPM (DRPM). While ESPN has yet to unveil the exact methodology for RPM, early reactions to the stat can be read here, here and here. According to ESPN:

“RPM estimates how many points each player adds or subtracts, on average, to his team’s net scoring margin for each 100 possessions played. The RPM model also yields separate ratings for the player’s impact on both ends of the court: offensive RPM (ORPM) and defensive RPM (DRPM).”

Naturally, we were interested in visualizing the relationship between RPM and NBA salaries. With the exception of a few notable outliers (Stoudemire and Boozer), we found a strong relationship between RPM and annual salary. Once you take into account players who signed large contracts before suffering serious injuries (Stoudemire) and players on the end of lengthy contracts whose skills may have deteriorated (Pau Gasol), on average NBA teams appear to reward higher RPMs with higher salaries. However, as you’ll see in this first graphic there are a lot of players who fall well-below or well-above the line of best fit.


To understand what’s happening here we have to unpack RPM into its offensive and defensive components. The bottom line may not come as a surprise. From a salary perspective, offense trumps defense, by a lot. Merging ESPN’s RPM data with salary information from Basketball-Reference, we estimate that an additional point in ORPM is worth on average $1.4M extra a year in salary while an additional point in DRPM only earns an $700K on average. This makes defense a bargain when you consider that a point scored should be equivalent to a point prevented.

The results are even more striking when we control for position. In essence, we’re now asking for the average relationship between salary and RPM within position. Using the same specification but now with controls for position, we find that a point in ORPM is still valuable at $1.5M (and highly statistically significant) while a point in DRPM falls to $300K (and is no longer statistically significant). To reiterate this, if we compare two players at the same position we find that difference in their salaries is heavily driven by their production on the offensive end and has no connection to their defensive impact. For those who have argued that defense is under-rewarded in the NBA, this may not be a smoking gun, but it is notable.


We can also rank players in the NBA according to value by constructing their predicted salary based on ORPM, DRPM, position and tenure and comparing this to their actual salaries. To be conservative, we dropped anyone who played fewer than 50 games this season. Below is a list of the top and bottom five players by value in three categories: guards, forwards and centers.

Top 5 Players by Position

Guards Fowards Centers
Name Value Name Value Name Value
Vince Carter $6,422,378 Nick Collison $11,015,718 Audray Blatche $8,663,985
Ray Allen $5,464,209 Chris Andersen $9,654,286 Ryan Hollins $6,032,525
Goran Dragic $5,185,118 Mike Miller $8,263,340 Nazr Mohammed $5,131,789
Derek Fisher $4,658,847 Matt Barnes $7,520,160 Alexis Ajinca $4,917,366
Kyle Lowry $4,465,802 Matt Bonner $7,254,578 BeJuan Blair $4,725,142

Bottom 5 Players by Position

Guards Fowards Centers
Name Value Name Value Name Value
Joe Johnson -$13,011,606 Amare Stoudemire -$18,847,008 Dwight Howard -$10,757,813
Dwyane Wade -$12,158,871 Carlos Boozer -$12,417,897 Pau Gasol -$9,139,959
Deron Williams -$8,344,061 Rudy Gay -$11,260,331 Chris Bosh -$9,090,025
Eric Gordon -$7,023,669 Carmelo Anthony -$10,155,075 DeMarcus Cousins -$7,399,312
John Wall -$6,838,249 Blake Griffin -$8,622,789 Roy Hibbert -$5,980,450

We’ll post links to the full lists by position shortly.