FYI.

This story is over 5 years old.

Sports

MoreyBall, Goodhart's Law, and the Limits of Analytics

Math tells NBA teams to shoot more three-pointers and shots at the rim, but that formula isn't working as well as expected for Houston and Philadelphia. What gives?
Troy Taormina-USA TODAY Sports

By any objective measure, the midrange shot is the worst in basketball. Math tells us that the best NBA offenses—that is, the most efficient—take lots of shots at the rim and from behind the arc. By those metrics, the Houston Rockets and the Philadelphia Sixers are playing the game the right way. Meanwhile, the San Antonio Spurs are one of the league's leading connoisseurs of the dreaded in-between game, jacking jumpers from an area of the floor that only the backward-looking Minnesota Timberwolves or the decidedly old-school Memphis Grizzlies seem to love.

Advertisement

And yet: neither the Sixers nor Rockets are as offensively productive as numbers suggest they should be. By contrast, the Spurs are matching their historically elite defense with a top-five offense. Theory isn't translating into practice.

So what gives—and what does this tell us about the limits of sports analytics?

Read More: The Myth Of Brilliant NBA Timeout Playcalling

Here's a thought: we might be seeing the basketball version of an economic concept known as Goodhart's Law, which holds that when a measure becomes a target, it ceases to be a good measure. (In sports terms, violating Goodhart's Law is roughly akin to "playing to the drill.") In the curious cases of the Sixers and the Rockets, it's quite possible that both teams have forgotten that three-pointers and shots at the rim are indicators of good offense, and not necessarily good offense in and of themselves—less a cause than a result.

Of course, if either club has got caught up in analytics for analytics' sake—hoisting very long and very short shots simply because that's what math dictates—well, it's hard to blame them. Basketball's data era is here to stay. From optical tracking and on-demand video libraries for on-court play to wearable tech and machine-learning-based training and maintenance regimens off the floor, there has never been more information available to teams, coaches, and players. And in this environment, there is a genuine advantage to be had in being first—except when it leads to erroneous implementation of overblown or premature conclusions.

Advertisement

TFW you do what the math says, but still fall off the bike. Photo by Bill Streicher-USA TODAY Sports

One of the bedrock principles of basketball analysis is the importance of getting shots from the right locations: at the basket and from three-point range (as well as from the free-throw line). The rationale there is that even with defensive pressure, shots from those areas are more efficient means of scoring than all but the most open of midrange shots. Some have taken to describing the desire to minimize the use of midrange shots as "MoreyBall," a mash-up of Michael Lewis's seminal Moneyball and Houston general manager and NBA analytic pioneer Daryl Morey.

(Important note: Morey himself does not claim to have developed this insight, and has seldom, if ever, publicly used the term MoreyBall himself).

Anti-modern, midrange-shot-heavy teams in Minnesota—and prior to this season, in New York and Washington—have felt the negative effects of swimming against the league-wide MoreyBall tide. However, the extremes of the approach in Philadelphia and Houston may be reaching a point of diminishing returns.

Since SportVU cameras were installed in every NBA arena prior to the 2013-14 season, Houston has led the league in the proportion of field goal attempts taken from either the basket area or beyond the arc. Philadelphia is third this year, after finishing second in 2014-15 and third (0.1 percent behind No. 2 Miami) in 2013-14.

In five of the six team-seasons compiled by the Sixers and the Rockets over the same time span, however, both squads have underperformed offensively, at least when measuring the Expected Effective Field Goal percentage (XeFG%) of their shots against the actual efficiency achieved. Take a look at the 2014-15 season:

Advertisement

In Philadelphia's case, an intentional lack of talent has a great deal to do with their struggles. One can't help but wonder if a slightly more balanced approach—forcing opposing defenses to defend more of the court while de-emphasizing the offensive weakness of the team's young, sometimes D-League-level roster—might result in moderately better performance.

The Rockets, by contrast, have much better talent. Yet outside of the 2013-14 season, when Chandler Parsons and Jeremy Lin were driving offensive creativity alongside James Harden, Houston also has failed to meet or exceed its XeFG%:

Watch the Rockets closely, and you'll notice that they frequently take shots that are "quality" in theory but seem not quite right: a three-pointer slightly out of rhythm or off-balance; a forced shot in heavy traffic near the basket. As the saying goes, a little bit of knowledge can be a dangerous thing. Misused and misinterpreted, new information can easily make a particular team worse off if the wrong lessons are drawn—and across basketball, new orthodoxies based on partially understood data could end up swapping old counterproductive biases for new ones.

The MoreyBall formula works great—if you're as good at shooting and driving as James Harden. Photo by Troy Taormina-USA TODAY Sports

Similar data misapplication already has happened in other sports. In English soccer, the "discovery" that goals were far more common after strings of three or fewer passes than after longer periods of sustained pressure led to a (supposedly) scientific long ball style. This discovery was rooted in a basic analytical mistake—more goals were scored after short strings of passes because shorter strings of passes vastly outnumber longer ones, and on a proportional bases more goals in fact followed from sustained possession—but no matter: the British game became enraptured with hoof-and-hope attacks, the lingering detrimental effects of which can still be seen in player development today.

Advertisement

Along the same lines, hockey analysts have determined that shot attempts are more predictive of future scoring margins than past scoring margins. Players whose time ice correlates with positive shot differential (as measured on metrics such as Corsi and Fenwick, named after their creators) thus have more value than traditional point totals might suggest. That is, at least until teams start playing to the metric, focusing on taking as many shots as possible rather than trying to score goals.

This may seem a distinction without a difference, but just as the development of the English "Route One" system degraded the quality of the possession in soccer, so too does indiscriminate long-range shooting in hockey. Precise calculation of "shot quality" is not needed to demonstrate that a shot from long distance and an acute angle is unlikely to beat a professional goalie, and is in many respects functionally similar to simply turning the puck over to the other team. Simply flinging the puck at net would thus break the assumptions of the underlying metric, and the team would "underperform" the model's prediction—which largely matches the experience of the Edmonton Oilers last season.

Not all shots are good ones. Photo by Kim Klement-USA TODAY Sports

Back to basketball. By and large, perhaps out of sheer conservatism, NBA teams have done a good job of staying on the right side of Goodhart's Law. Coaches understand that "early offense" has well-established benefits; indeed, the Steve Nash-Mike D'Antoni Phoenix Suns "Seven Seconds or Less" attack was based on early shots. That said, no team—not even the current Golden State Warriors, who have perfected the D'Antoni approach—assumes that shooting quickly is good in and of itself; to the contrary, they grasp that early shots tend to be good shots because the same shots would be good at any point during a possession. How so? Simple. Pushing the pace allows teams to get more open looks from higher-value spots on the floor because the defense has not yet recovered to fully guard those areas. This benefit would not accrue from simply dribbling down the floor, launching at the first semi-decent opportunity, and justifying it with "early offense is good." (See the discussion of "dynamic efficiency" here for a demonstration of how well NBA teams have incorporated this lesson into their play).

Likewise, NBA teams largely avoid a Hoosiers-style "four passes before we shoot" approach, even though every team in the league shoots significantly more efficiently on shots set up by teammates than not. Again, the lesson isn't "throw more passes." It's "throw passes to open players who can make shots from those spots." Case in point: Utah currently leads the league in passes per game, but is No. 24 in the proportion of its shot attempts that come via passing. The Jazz aren't trying to goose scoring just by throwing the ball around more; that would be stupid, and counterproductive. The team's lack of "productive" passing is more likely due to a dearth of perimeter playmakers.

Like assists and early shots, three-pointers and shots at the rim are indicators of good offense, but they're not good offense in a vacuum, and teams that use them as targets should be wary of putting the cart before the horse. Those are good shots in theory. In practice, the best shots are the ones the personnel on hand can make, a lesson the Spurs have put to use as an one of the NBA's annual leaders in measures of shot-making accounting for difficulty:

Ultimately, the goal is to score the most points on a per possession basis, not to hit benchmarks for the sake of doing so. It's good to play the right way, but better to play the best way for your own personnel. Striking that balance isn't easy, but when hit, it takes analytics from the realm of the spreadsheet and puts it back on the court, where it ultimately belongs.