In basketball, passes are means to an end — tools that teams can use to increase their chances of scoring. In many cases, they can lead to efficient offense. Catch-and-shoot jumpers, for example, tend to be converted at a higher clip than comparable pull-up attempts. In pick-and-roll situations, possessions that are used by roll men generate more points on average than those that are used by ball-handlers. Shot location and numerous other factors obviously have something to do with it, as well, but the facilitation helps create or exploit offensive advantages.
But, as previously noted in this space, passes are hardly an absolute prerequisite to efficiency. Although the Golden State Warriors and Denver Nuggets distinguish their high-powered scoring with ball movement, the Houston Rockets, Milwaukee Bucks and Portland Trail Blazers manage to have top-five offenses while ranking among the bottom 10 in passes per game. We saw the same thing from the Cleveland Cavaliers in the two seasons prior to LeBron James’s departure. As a member of the Thunder, Kevin Durant famously proclaimed that “we’re not the San Antonio Spurs,” by which he meant that Oklahoma City was unlikely to “make 30 passes in a possession.” If offensive style charts teach us anything, it’s that there is “more than one way to skin the efficiency cat.”
Indeed, while passes have their benefits, they are not exactly unqualified goods. As with anything, they come with tradeoffs, and the opportunity costs can be substantial in certain cases. One commonality between the low-passing teams above is that they feature lead ball-handlers (James Harden, Chris Paul, Giannis Antetokounmpo, Damian Lillard, Russell Westbrook and James) who facilitate efficient offenses best when the ball is largely in their hands. A shift towards greater ball movement incurs the potential downside that the primary offensive option will be out of the main action when it’s time to get a shot. There is also the notion that increased passing might expose the team to additional turnover risk. Perhaps it’s ultimately desirable to institute a dynamic, egalitarian system in which every player touches the ball, but such a strategic choice is more complicated than we typically assume.
These considerations serve as a useful segue back to the question of how much influence the defense can exert on ball movement. If teams can achieve offensive efficiency whether they pass extensively or selectively, does it even matter that, as described in last week’s article, the offense has roughly 90 percent control over this aspect of the game?
Perhaps the issue ought to be reframed. Heretofore, we’ve defined ball movement in terms of sheer activity and quantified it as passes per possession. This metric includes inbound and other types of passes that have little to do with core offensive strategies. It might be more valuable to focus on productive manifestations of ball movement and revisit the offense-defense split from there.
Along these lines, one simple measure is the percentage of made shots that are assisted. Here’s a breakdown of team assist rates on both offense and defense over the past six regular seasons:
While the plot suggests that assist rates tend to vary more on offense than on defense, let’s take a deeper look. Employing the same linear regression model outlined in the previous article, we can attempt to predict a team’s assist rate for a particular game based on its season-long offensive assist rate and its opponent’s season-long defensive assist rate. All three variables are scaled to league average. The season-long numbers exclude data from the game in question. Altogether, there are 14,436 observations from regular-season games between 2013-14, when tracking data were initially released, and 2018-19 (through March 23).
This model explains 19 percent of the variance in assist rate. There is, to be sure, quite a bit that’s left unexplained by season-long averages on both sides of the ball. One factor might be game location. Assists are subjective, with scorekeepers, who are hired by the home team, given latitude on how to track the data. In 2016, Matthew van Bommel and Luke Bornn documented the extent of scorekeeper impact in the NBA. A year beforehand, when Ken Pomeroy examined offensive and defensive influences on NCAA assist rates, he discovered “a substantial home-court effect.”
Setting aside such concerns for another time, let’s examine the offense-defense split within the proportion of the variance in assist rate that’s actually explained by the model:
As the table suggests, approximately two-thirds is due to offense and one-third to defense. This result is comparable to Pomeroy’s finding for men’s college basketball, as well as to Christopher Long’s conclusion regarding pace. But, while defense has less control than offense, its influence on assist rate is greater than on passes per possession, which strikes me as intuitive. Consider, for example, a situation where the offense swings the ball around the court until it reaches a poor shooter above the break. If the defense sags and stays home on his teammates, it could induce him to dribble inside the arc and pull up from the elbow. Altogether, the type of unassisted shot that terminates the possession seems to be within the defensive sphere of influence to a larger degree than the sheer number of times the ball is distributed from one offensive player to another.
Apart from its subjective nature, one critical issue with assist rate is how it overlooks missed shots. We can address it at least partially with tracking data. NBA.com has “potential assists,” which are defined as “any pass to a teammate who shoots within one dribble of receiving the ball.” While this stat doesn’t capture the quality of the scoring opportunity (shots within one dribble of a pass can still vary widely in expected value), it has the advantage of being divorced from scorekeeper judgment. It also casts a wider net by ignoring the outcome of the shot — i.e., both makes and misses are tallied.
If we divide potential assists by field goal attempts and follow the same methodology that’s described above, we get these results:
The offense-defense split is 60-40. In other words, defensive influence on potential assist rate appears to be higher than on assist rate. My sense is that the more process-oriented conception of the former enables the defense to exert greater control. After all, with potential assists, the focus is on limiting the number of catch-and-shoot and one-dribble scoring opportunities. Assist rate, by contrast, is predicated on made baskets. As we know from previous research on opponent 3-point percentage, once the emphasis tilts toward whether a shot is converted or missed, defensive control starts to diminish.
This topic warrants further study, but there already seems to be some confirmation of what we intuitively know: the defense can do something about whether passes lead to advantageous shot attempts.