What Advanced Metrics Actually Tell Us
Dr. Maya Chen
2026-03-01 ·
The Strange Authority of a Number
In the 2015–16 season, Stephen Curry averaged 30.1 points per game while posting a true shooting percentage of 66.9. The number circulated everywhere that year. Analysts used it to explain the efficiency of his scoring, commentators repeated it during broadcasts, and fans cited it as proof that they were witnessing one of the greatest offensive seasons the league had ever seen.
Yet the number itself is an abstraction. True shooting percentage does not exist on the court in the same way that a three‑pointer does. No player feels it in their hands, and no referee signals it with a whistle. It is a constructed measure, a mathematical way of compressing several types of scoring into a single expression of efficiency.
This raises a surprisingly philosophical question: when we rely on numbers like this, what exactly do we think they represent?
Are advanced metrics discovering real features of basketball, or are they simply useful tools for prediction and comparison?
The debate echoes a long-standing disagreement in philosophy about scientific models. Some thinkers believe good models reveal real structures in the world—a position philosophers call realismRealism is the view that the entities or structures described by our theories and models genuinely exist in the world, independent of whether we observe or measure them. . Others argue that models earn their value simply by helping us organize and predict experience, regardless of whether their categories correspond to reality itself—an approach known as instrumentalismInstrumentalism is the view that theories and models need not describe reality accurately; they are valuable insofar as they serve as useful tools for prediction and practical decision-making. .
Advanced basketball analytics sits right inside this tension.
Models and the Court
A modelIn philosophy of science, a model is a simplified representation of some aspect of reality, built to highlight certain relationships while deliberately ignoring others in order to make complex systems understandable. , in the philosophical sense, is not a copy of the world. It is a simplified representation built to highlight certain relationships while ignoring others.
A subway map is a familiar example. The map does not show every street or building in the city. It bends geography, compresses distance, and exaggerates connections between stations. Yet it still represents something real about the transportation system because its simplifications are disciplined rather than arbitrary.
Advanced metrics function in much the same way. They compress the chaos of a basketball game into patterns that can be compared and interpreted.
True shooting percentage simplifies scoring into a single efficiency measure. Box Plus-Minus estimates a player’s overall contribution using box‑score statistics. Offensive ratings translate team possessions into points-per-possession logic. Each of these numbers is built from assumptions about how basketball works.
The philosophical question is whether those assumptions are merely convenient, or whether they actually track real structures within the game.
Efficiency and the Real Structure of Scoring
Some metrics feel surprisingly realist when you examine what they capture.
True shooting percentage is a good example because it reflects a simple but fundamental structure of basketball: not all points are created the same way.
A three‑pointer is worth more than a midrange jumper. Free throws convert possessions into points differently than field goals. When Curry’s scoring season produced a true shooting mark near 67 percent, the number was not inventing an imaginary concept. It was expressing a real relationship between shot selection, accuracy, and point value.
In that sense, the statistic behaves like a model that has latched onto a genuine feature of offensive basketball. It does not capture everything about scoring—shot difficulty, defensive pressure, or stylistic beauty—but it represents an underlying structure that genuinely shapes how points accumulate.
The model leaves out much of the game, yet what it keeps corresponds to something real.
When the Model Becomes the Illusion
Other metrics expose the limits of treating numbers too literally.
Box Plus-Minus, for instance, estimates player impact using patterns in the box score. The logic behind the metric is that rebounds, assists, steals, scoring efficiency, and similar statistics correlate with team success, so the model translates those signals into a single value.
But the translation is not neutral. The model must decide how much weight each event deserves and how those events combine into “impact.” Those decisions create blind spots.
Russell Westbrook’s 2016–17 season provides a vivid illustration. Averaging a triple‑double across the year produced extraordinary box‑score output, and early versions of the metric rated the season historically high. Yet that same moment exposed weaknesses in the model itself, because certain statistical patterns were being overvalued relative to their actual effect on winning.
The lesson was not that Westbrook’s performance lacked value. It was that the model’s interpretation of the box score had mistaken its own assumptions for reality.
When that happens, the number begins to function less like a representation and more like an illusion of precision.
Instrumental Success on the Court
Sometimes a model proves its worth in a more pragmatic way.
Consider the Houston Rockets during the 2017–18 season. Their offense was organized around a simple analytic idea: certain shots—three‑pointers, layups, and free throws—produce more points per attempt than long midrange jumpers.
The Rockets built an entire offensive system around that logic, emphasizing spacing and shot selection that maximized those high‑value attempts. The result was a 65–17 record and one of the league’s most efficient offenses.
From a philosophical perspective, this is the instrumental side of analytics. The model does not need to capture every aesthetic or tactical nuance of basketball. It only needs to guide decisions in ways that improve outcomes.
Whether the shot‑value framework perfectly describes the essence of good offense is almost beside the point. The model succeeded because it changed behavior in a way that produced wins.
In this sense, analytics can function less like a theory about reality and more like a strategy for navigating it.
The Problem of Invisible Value
Defense pushes the philosophical puzzle even further.
Take Rudy Gobert’s impact on the court. His value has often been recognized through team defensive performance and awards like Defensive Player of the Year, yet much of what makes him effective never appears clearly in the box score.
Rim deterrence, positional gravity, and the way a center reshapes opponents’ shot choices are not easily counted. They exist as causal forces within the game even when they are not directly recorded.
Defensive analytics therefore attempts something ambitious: it tries to infer real effects that are only indirectly observable.
This is where realism becomes tempting again. Analysts build models precisely because they believe those hidden contributions actually exist. The goal is not merely to predict defensive performance but to represent the underlying mechanisms that produce it.
The numbers remain imperfect, but they point toward realities the eye alone struggles to measure consistently.
When Models Converge
Occasionally, different kinds of evidence begin to agree.
Nikola Jokić has repeatedly ranked near the top of advanced metrics while also leading elite offenses and winning MVP awards. Box‑score models rate him highly. Efficiency statistics highlight his scoring and passing productivity. Team-level outcomes reflect the effectiveness of the offenses built around him.
When independent measurements convergeConvergence in philosophy of science occurs when multiple independent methods or lines of evidence point toward the same conclusion, strengthening confidence that the conclusion reflects something real rather than an artifact of any single method. in this way, confidence in the underlying conclusion grows. It does not mean any single metric is perfectly accurate. But the overlap between multiple simplified models suggests that they are detecting something real about Jokić’s influence on the game.
The models differ in design, yet they point toward the same structure of impact.
Seeing the Numbers Differently
All of this suggests that advanced metrics occupy a middle ground between truth and utility.
Treating them as literal measurements of basketball reality is naive, because every metric depends on assumptions about what counts and how value should be distributed. At the same time, dismissing them as arbitrary numbers misses why they work at all.
Metrics succeed because many of their simplifications track genuine patterns in the game—patterns in possession value, shot efficiency, spacing, and lineup interaction. They are abstractions, but disciplined ones.
The best way to read them is with a kind of cautious realism.
A metric is not the game itself. It is closer to a map drawn by someone who knows which details matter and which can safely be left out. The map may distort the terrain, yet it can still guide you across it.
And when the map fails, the failure usually tells us something as well—about the limits of the model, the complexity of basketball, or the parts of the game we have not yet learned how to measure.
In that sense, analytics does more than produce numbers. It continually forces us to ask what parts of basketball are truly visible, and what parts are still hiding behind the models we use to understand it.
Footnotes / Philosophy Terms
1. Realism ↩
Realism is the view that the entities or structures described by our theories and models genuinely exist in the world, independent of whether we observe or measure them.
2. Instrumentalism ↩
Instrumentalism is the view that theories and models need not describe reality accurately; they are valuable insofar as they serve as useful tools for prediction and practical decision-making.
3. Model ↩
In philosophy of science, a model is a simplified representation of some aspect of reality, built to highlight certain relationships while deliberately ignoring others in order to make complex systems understandable.
4. Convergence ↩
Convergence in philosophy of science occurs when multiple independent methods or lines of evidence point toward the same conclusion, strengthening confidence that the conclusion reflects something real rather than an artifact of any single method.