Dec 10, 2013; Toronto, Ontario, CAN; San Antonio Spurs guard Manu Ginobili (20) goes to pass the ball as Toronto Raptors guard Terrence Ross (31) defends at the Air Canada Centre. San Antonio defeated Toronto 116-103. Mandatory Credit: John E. Sokolowski-USA TODAY Sports
One of the biggest difficulties in the field of basketball analytics is how to assign credit to individual players. In fact, some of the most contentious arguments in the field concern the proper allocation of credit for things we know have value on the team level. Dean Oliver’s Four Factors tell us that rebounding is pretty important at the team level. But how much of the value gained by the team is properly assignable to the player who so happens to secure that rebound? While forcing a change of possession with zero points scored is a great outcome for the defense, it doesn’t meet even passing scrutiny to give the rebounder full credit. What of the guy who challenged the initial shot? Or the player who boxed his man out to allow a teammate to snatch the board? It’s an interesting conundrum, but arguing the final points rapidly reaches diminishing returns in the absence of better input data to study.[1. Some of the Second Spectrum research discussed here could be a strong start in that direction.]
Some aspects of basketball are a little easier to parse — after all, it’s easy to identify the shooter as the one “using” an offensive possession. Thus, it has customarily been the practice to ascribe all credit for an offensive possession to the shooter. Intuitively, this is understood to be an extremely broad generalization. In what meaningful way is it really proper to assign DeAndre Jordan full credit for these two points?
Or Boris Diaw with these three?
Herein lies a key point of differentiation with baseball[2. To a large extent, the static confrontation between batter and pitcher makes assigning individual credit far easier in baseball. The analytic differences between the two sports is a subject for a future post]: neither of those baskets were made by the scorer in isolation from the actions of their teammates. Other players were involved with creating those opportunities. By subtly shifting the frame of analysis from examining which one player “used” a possession, we can instead look at all the players who were “involved” with the conclusion of a possession, whether via shot, foul or turnover.
Prior to last season, there was a data problem. In terms of publicly available stats, all we really had were shots and assists. Thankfully, the SportVU system has added some meaningful inputs to the tool box, primarily through tracking assist “chances” — that is passes that lead to a shot that if it went in, would lead to an assist for the passer. In the above plays, Blake Griffin would be credited even if Jordan missed the lay in, as would Duncan had Diaw not knocked in the corner 3[3. Public SportVU data also tracks secondary or “hockey” assists when a pass immediately leads to another pass which leads to a basket and free throw assists which are passes leading to shooting fouls.].
By measuring a player’s playmaking contribution through this potential assist data, it becomes possible to measure a player’s “playmaking” contribution. This “assist usage” can be expressed on a scale similar to the more familiar (shooting) Usage stat[4. USG% is an estimate of the portion of possessions a player uses on offense while on the court. It’s simply a player’s turnovers + shot attempts + .44 * FTA (the .44 coefficient is an empirical estimate of the percentage of free throw attempts that truly “end” a possession not already captured as an FGA) / offensive possessions played]. However, unlike Usage, each “assist chance” play can be credited to between zero and two players[5. There’s no theoretical reason why it could not be more players other than there is no systematic tracking for plays beyond the hockey assist. For example Ginobili would receive no credit for that Spurs tic-tac-toe passing sequence as three players handled the ball after he passed.].
This “assist usage” describes an important part of a team’s offense in terms of creating shots for others. A benefit to looking at playmaking in this way is it allows us to construct a similar “Usage/Efficiency” model for this aspect of the game. Much like shooting, it turns out that the more shots one creates for teammates, the harder it is to complete passes, as shown by this graph mapping pass completion percentage[6. 1 – (Bad Pass Turnovers / passes)] against the percentage of of total passes which lead to assist chances[7. Thanks to Matt Femrite for the visualization.].
This is to some degree intuitive. The “easy” pass probably doesn’t often lead to a layup, else what’s the defense there for?
By measuring turnovers against assist opportunities, we can get a better idea of who is actually turnover prone. Point guards tend to be high turnover players, both in terms of per minute and under traditional TOV%. In 2013-14, point guards overall had a TOV% of around 15.6%, compared with around 12.3% for wings and 13.0% for bigs. This is a result of the TOV% formula which is just turnovers / (turnovers + shots + trips to the free throw line). In other words, one of the best ways to “improve” your TOV% is just to shoot more.
By comparing shots, assist chances and turnovers with total possessions played, we can look at a player’s “True Usage.” This number represents a player’s total offensive involvement. Unlike traditional usage, which would theoretically sum to 100%[6. Offensive rebounds are typically omitted as possessions from usage calculations so some possessions might be counted twice or more. on the team level], there is no set total, though of course at least 100% of possessions are going to be accounted for in this way. But, to my thinking, it’s more intuitive that in many possessions, multiple players are credited with involvement — as the plays above show, those baskets arose from the contributions of multiple players, without all of which, the basket doesn’t happen.
Additionally, it better reflects how an NBA offense really works. Even though point guards often don’t carry the biggest load in terms of shooting the basketball, the point guard (or other wing “taking on primary ball-handling responsibility”) has a large overall role to play in the creation of shots. Looking at the NBA positional averages for True Usage, this is clear:
Applying this example to specific teams, let’s look at how big a hole there is to fill in Miami’s offense with LeBron leaving:
So not only was LeBron responsible for the largest portion of Miami’s shots taken, he also created the most shots for others. All told LeBron was the highest non-point guard True Usage player in the league (with the rest of the top five being Tyreke Evans, Kevin Durant, Manu Ginobili and Monta Ellis, all of whom fit the “primary ball-handler” role when on the floor).
Speaking of Ginobili, how does San Antonio’s roster measure up?
Finally, a hot topic for much of last season and especially the playoffs was Oklahoma City’s offense. Looking at the balance (or lack thereof) in their roster, and it’s somewhat clear why it ran the way it ran — only a few players were really capable of creating shots, so the ball tended to stick in their hands:
Aside from Westbrook, Durant and Reggie Jackson to a degree (though his numbers are inflated some by his time with Westbrook out of the lineup), the rest of OKC’s parts simply couldn’t create shots for themselves or others. This area more than anything is where the replacements for James Harden are most lacking (Harden posted a 16.8% Assist Usage in 2013-14).
A note about using these numbers to evaluate players: these stats are measuring the size of a players’ role and with the exception of turnovers (more turnovers are always bad, everything else being equal) are not a reflection of effectiveness within those roles. TS% is still the gold standard for shooting efficiency, but a similar measure of passing efficiency is difficult because of that old problem of assigning credit — simply measuring the points scored off of players passes might tell us a lot more about his teammates than about his own intrinsic playmaking ability. But that’s another topic for another time.
In the next week or so, we’ll publish sortable True Usage stats for the entire league for 2013-14 here on Nylon Calculus as a resource, so stay tuned for that.