Loading field map
How do you decide whether a given shot is a good one? For decades, basketball coaches and analysts relied on intuition, tradition, and raw field goal percentage. But the rise of shot profile analytics over the past twenty years has transformed that question into a rigorous, data-driven investigation. The central tension driving this subfield is the gap between the coarse metrics that dominated early basketball statistics and the need to evaluate every shot in its full context: location, defender pressure, game situation, and the player's skill. Six major frameworks have emerged to close that gap, each building on or reacting against its predecessors.
The Efficiency Framework, introduced around 2004 and often associated with the work of Dean Oliver, shifted the fundamental currency of basketball analysis from points per game to points per possession. Its Four Factors—effective field goal percentage (eFG%), turnover rate, offensive rebounding rate, and free throw rate—provided a systematic way to account for pace and efficiency. By replacing raw field goal percentage with eFG% (which gives extra weight to three-pointers), the framework established that not all makes are equal. This was a foundational move, but it treated every shot of a given type (e.g., all three-pointers) as having the same value, ignoring where on the court the shot was taken.
Shortly afterward, the Spatial Shot Chart Paradigm (2005) added location to the conversation. Coaches and analysts began plotting every shot on a diagram of the court, creating heat maps of shooting accuracy. This framework was primarily descriptive: it showed where a player or team shot from and how often they made those shots. It did not, by itself, say whether a shot was a good decision. The spatial chart coexisted with the Efficiency Framework as a visualization tool, but it lacked a mechanism to predict the value of future shots from specific zones.
The Expected Points Per Shot (EPPS) Models (2010) absorbed the spatial insight and added predictive power. By calculating the average points scored per shot from every location on the court, EPPS provided a baseline expectation for any shot attempted from that spot. This was a major step beyond the spatial chart: EPPS could compare the expected value of a contested mid-range jumper (say, 0.75 points per shot) against a corner three (around 1.2 points per shot), highlighting inefficiencies in real time. The framework treated each shot as an independent observation, assuming that the expected value was constant for a given location regardless of defender, game state, or player identity.
EPPS insights were quickly translated into a strategic doctrine by the Moreyball Philosophy (2012), named after Houston Rockets general manager Daryl Morey. Moreyball is not a statistical model but a methodological school that narrowed EPPS into a rigid decision rule: avoid mid-range shots entirely, focus on layups, dunks, and three-pointers, and never settle for the lowest-efficiency zone. This represented a living disagreement with traditional half-court offense, which often prized the mid-range as a fallback option. Moreyball took the EPPS conclusion—that the mid-range is the least efficient zone—and turned it into a team-wide identity, generating intense debate about whether such a rigid rule sacrifices adaptability.
The Bayesian Shot Models (2015) addressed a key limitation of EPPS: the assumption of fixed, known expected values. Bayesian methods treat a player's shooting ability as a probability distribution that can be updated with new data, incorporating prior information (e.g., college shooting percentages) and providing more reliable estimates for small sample sizes—for instance, a player who has taken only ten shots from a specific zone. These models did not replace EPPS but complemented it, offering a tool for handling uncertainty. They coexisted with EPPS, often feeding into the same analytical pipelines.
Also in 2015, Machine Learning Shot Analytics emerged as the current leading frontier. Machine learning models—such as random forests, gradient boosting, and neural networks—process high-dimensional data from player tracking systems (e.g., SportVU cameras). They can account for defender distance, shot clock time, dribble count, and even the proximity of teammates, capturing interactions that EPPS and Bayesian models cannot. This framework has transformed the subfield because it can evaluate shots in their full context rather than assuming fixed zone averages. However, this predictive power comes at a cost: machine learning models are often black boxes, making them harder to interpret than EPPS or Bayesian models. Coaches may see that a model labels a shot as poor but struggle to articulate why.
Today, the six frameworks coexist with a clear division of labor. The Machine Learning Shot Analytics framework leads the frontier because it can handle the massive, noisy data streams produced by modern player tracking. EPPS models remain widely used for quick, interpretable evaluations—they are the default metric on broadcast graphics and in post-game analysis. Bayesian Shot Models are integrated into advanced analytics workflows, especially for evaluating rookies or players with limited minutes. The Moreyball Philosophy continues to influence the design of entire offenses, though its rigidity is increasingly questioned as machine learning models reveal that some mid-range shots—those taken by elite shooters in specific contexts—can be valuable.
On what do the leading frameworks agree? Almost all of them affirm that shots near the rim and three-pointers generate higher expected value than mid-range jumpers, and that volume is critical: teams should not just take efficient shots but take many of them. The core disagreement, however, lies in how much context matters. The Moreyball camp treats the mid-range as always inferior, while machine learning models suggest that a wide-open mid-range shot by a skilled shooter in rhythm can be as efficient as a contested three. This debate has no settled resolution; it reflects the broader tension between analytical prescription and the irreducible human element of shot creation.
The history of shot profile analytics is not a story of simple replacement. Each new framework preserved or transformed insights from earlier ones. EPPS built on spatial charts; Moreyball narrowed EPPS into doctrine; Bayesian and machine learning models expanded the analytical toolkit. The field has become a layered ecosystem where different frameworks serve different purposes—from quick descriptive summaries to deep predictive models—and where the central question of what makes a good shot continues to evolve.