Quick — which NBA player is most integral to his team’s offense? Which player shoulders the biggest offensive burden? And to what degree are those questions even equivalent?
Statistically, such concerns fall under the umbrella of “usage rate,” a term that colloquially describes an entire class of metrics tasked with quantifying the size of a player’s offensive role. Usage is one of the most accessible concepts in basketball analytics — rock-simple in its purview and relatable to anyone who’s ever played with a shameless ball hog or been a terrified freshman playing hot potato. In statsier circles, usage is a staple of player analysis, in part because it remains relatively constant amid a player’s shifting contexts and roles. At a glance, usage says more about how a player plays than most other basic basketball metrics.
One small problem: Nobody seems to agree about what exactly usage rate is, or should be, or how it is calculated. Many analytics-minded observers don’t even know there are different, competing versions of the statistic in popular use, much less that each variant has its own philosophy about what it means to “use” a possession. For a term so common to the modern hoops lexicon, that’s more than a little strange. So let’s have ourselves a little history lesson and learn much more than you ever wanted to know about usage rate, in all its permutations.
Usage through the years
Like many concepts in basketball analytics, usage rate can be traced back to Dean Oliver and John Hollinger, still probably the field’s two most influential figures. The notion that too much (or too little) offense could flow through an individual player is as old as the game itself, but it’s hard to find anyone formally putting a number on the phenomenon before the early-to-mid-2000s, when Hollinger published his inaugural “Pro Basketball Prospectus” and Oliver wrote the seminal “Basketball On Paper.” In fact, the thought of listing a player’s rate of possession-usage at all — let alone as something other than a purely negative indicator — was alien to many of the early hoops number-crunchers.
To understand why, it’s useful to look back at the primordial era of basketball metrics. NBA statheads cribbed many of their early concepts from baseball’s sabermetric movement — which effectively had a 25-year head start — including a tunnel-visioned focus on maximizing efficiency. Such a fixation makes sense in baseball, where a player’s susceptibility to making outs is unambiguously negative — you get 27 of them each game, to be guarded vigilantly — and you can draw a straight line between a player’s individual efficiency and his effect on the team. Hence the reasoning, as applied to basketball: If possessions, like outs, are the sport’s fundamental unit of opportunity, why would we celebrate a player’s propensity for using them up?
Basketball is more complicated than baseball, however. Possessions alternate between teams, so at least one player must always have a hand in “using” each of them. More importantly, teammates do not take turns with their opportunities like hitters going through a batting order: Any individual player is free to use as many (or as few) of the team’s possessions as he wants. This provides a lot of complex ways for an individual to help the team beyond his own personal efficiency statistics.
One of Oliver and Hollinger’s key insights was that the frequency with which a player generates offense — as proxied by usage rate — is a consideration that should always accompany (and temper) his efficiency metrics. “Some guys … are great shooters and passers, and rarely turn the ball over,” Hollinger wrote, introducing usage in the 2002 edition of his “Prospectus,” predicting the wars he’d fight over Carl Landry half a decade later. “If that’s the case, why don’t people regard them as superstars? The reason is that they cannot create their own shot as often as some other players can.” Usage rate was born out of the effort to quantify said ability to create.
Hollinger’s original conception of usage, which can still be found at ESPN.com today, was a relatively simple pace-adjusted rate of shots, assists and turnovers per 40 minutes. Oliver’s, while rooted in the same basic tenets, went to a far more complex place, accounting for the additional possession-extending nature of offensive rebounds and even parceling out fractional credit to the scorer and passer on an assisted basket. But at their most elemental, both attempt to individually account for all the actions that can spell an end to any team possession: made baskets, misses that aren’t rebounded by the offense, free throws and turnovers.
Neither Oliver’s nor Hollinger’s interpretation of usage, however, is the preferred version of 2015’s stathead. (At least, not according to this unscientific Twitter poll I conducted Tuesday.) Among the respondents who actually recognized differences between various flavors of usage, nearly twice as many said they use the Basketball-Reference.com (BBR) version as Hollinger’s. (Oliver’s version isn’t widely available online, except for college players.)
As the stats are used today, there isn’t much separating the three. Mention that a player’s usage rate or usage percentage is in the high-20s to low-30s and you call to mind a ball-dominant focal point of an offense; drop down an octave, into the low-to-mid-20s, and you instead have a player who creates a good deal of offense but doesn’t dribble the leather off the ball. Whichever version you prefer, usage is in common enough usage that it serves as shorthand for offensive hierarchy.
In most every practical application, breaking one or the other down to its atomic particles and recompiling them into the competing version will be pointless; you already get the idea. Still, it remains worthwhile to understand the differences, such as they are, and how those differences inform what it is you’re looking it. Why? Because BBR’s usage metric doesn’t include assists.
Confusion reigns
Full disclosure: I used to work for Sports-Reference, the company that runs BBR, so I’m close to the situation. And now, a scene from my former life running the company blog at a time when BBR founder Justin Kubatko and I staged nerd fights about this (and other statistical barnacles):
ME: “Why do we use Hollinger’s definition of usage instead of Dean Oliver’s?”
JUSTIN: “That’s not Hollinger’s. That’s mine.”
ME: “It’s not what he uses at ESPN? I thought it was the same definition.”
JUSTIN: “No. His multiplies assists by a third.”
ME: “I see. But I guess the question still stands.”
JUSTIN: “Mine is basically percentage of team plays used. What the heck is his actually measuring?”
ME: “It’s trying to measure possessions, and failing. But Oliver’s formula gives us real possessions.”
JUSTIN: “They’re not real, either! They’re estimates — better than Hollinger’s, but estimates.”
ME: “I’m confused. This is Hollinger’s fault.”
For most players, this distinction is largely irrelevant; among qualified players1 this season, the correlation between BBR usage and Oliver’s more full-bodied formula is 0.98. But for certain types of players, it can matter: It’s the difference, for instance, between claiming that DeMarcus Cousins carries the league’s biggest offensive burden (as he does under BBR’s formula) and giving the distinction to Russell Westbrook (No. 1, according to Oliver and Hollinger). One measures pure scoring affinity; the others factor in ballhandling responsibility while still strictly accounting for the player(s) who served as the conduit for every possession’s end.
Neither approach is perfect. Playmaking is obviously a massive part of “creating” offense, and cutting it out entirely isn’t ideal. But just stapling assists onto a scoring metric misses huge chunks of what you’re trying to capture. Plus, heavy ballhandlers tend to have higher turnover rates than would be predicted from how often they end possessions, which suggests that even a completist accounting method such as Oliver’s is missing some fundamental aspect of how passers create shots for others.
So with the advent of player-tracking data from SportVU, Seth Partnow of NylonCalculus.com set out to detect the invisible. He developed a statistic called True Usage, which incorporates “assist chances” (so-called “hockey assists,” plus passes that would have been scored as assists if the shot had been made) into the usage mix. The resulting leaderboard is decidedly skewed toward point guards and other primary ballhandlers, like LeBron James. If we’re truly interested in measuring a player’s offensive burden, that probably makes for a more accurate usage framework.
The problem, of course, is that the old-hat usage figures are now entrenched in not only the analytic lexicon, but also the updating leaderboards on big industry portals like Basketball-Reference and ESPN. It’s hard to change hearts and minds without first winning over the APIs.
From one stat to many
Then again, maybe the entire concept of a one-number “usage rate” has outlived its usefulness, particularly in an age of hyper-detailed SportVU possession stats. We can now see how long a player holds the ball, how often he passes, how many points those passes create — every conceivable piece of the puzzle is out there, if you know where to look. And just about every basketball analytics expert I consulted told me that they preferred a modular approach to usage, with different formulas to measure different aspects of a player’s offensive responsibility.
“I don’t use just one usage stat,” Oliver told me. “I do have a shot usage, a field goal usage, and a possession usage stat. Depending on the question being asked, I will look at the one that makes the most sense.”
Jacob Rosen, who writes about analytics for Nylon Calculus and the Cleveland sports blog Waiting For Next Year, concurs that today’s all-in-one usage metrics are inadequate. “Like any type of basketball stat, it’s the balance of wanting to push everything into one metric,” Rosen said. “In some ideal world, you’d have a stat that measures the dimensions of possession time, passes, potential assists, turnovers, shots, free throws, etc. But they’re on somewhat different planes of existence.”
As a possible alternative to a one-size-fits-all usage formula, Rosen wondered if usage rate’s next step would be to incorporate player typologies, such as the Position-Adjusted Classification (PAC) system developed by current Cleveland Cavaliers Director of Analytics Jon Nichols. “In my mind, having those different dimensions would be more accurate,” Rosen said. “You could perhaps do a PAC definition just with usage-based things alone (i.e., passing, possession, turnovers, shots).”
Given the state of today’s tools of observation, Partnow’s True Usage may have struck the best balance between the all-encompassing and the customizable, if not the most widely used and understood.
“To me the ideal is True Usage,” Nylon Calculus writer2 Ian Levy said. “It is as accurate a measure as there is of the quantity of a player’s offensive responsibilities. But the real benefit is that you can parse out the different components to see what comes from playmaking, scoring, turnovers. That’s the ideal — [a] good holistic measure [that’s] also parsable into components for descriptive uses.”
If so, maybe we should all just turn our attention toward rebranding campaigns for the other myriad versions of usage rate — “Possession Rate”? “Scoring Attempt Frequency”? — or pester the bosses at ESPN or Basketball-Reference for one more column in the Advanced Stats tab. That is, until basketball’s next data revolution comes and brings with it an even more accurate way to measure offensive workload … which we can promptly christen “usage rate” and start all over again.