All scores →
Loading…
Loading…
Loading…

Expected points and shot quality: tracking-era basketball

By 4Basketball Staff · · 11 min read

Tracking cameras let analysts estimate how likely each shot was to go in. This guide explains expected points, shot quality versus shot making, and the limits of the models behind the numbers.

For most of basketball's history, a shot was a shot. The scorer's table recorded whether it was a make or a miss, whether it was worth two points or three, and, in later decades, roughly where on the floor it came from. That was enough to build field-goal percentage, effective field-goal percentage and a long list of familiar statistics. What it could not do was say whether a particular miss was a bad decision or good process with bad luck. A contested heave at the end of the shot clock and a wide-open corner three were both just misses.

That changed when cameras began tracking every player and the ball many times per second. With positions recorded continuously, analysts could attach to each shot a rich set of facts: where the shooter was, where the nearest defender stood, how many seconds were left on the shot clock, how many dribbles preceded the attempt and how fast the shooter was moving. From those facts they could estimate the probability that a typical shooter would make that shot. Multiply the probability by the points at stake and you have expected points, a measure of the quality of the shot itself, separate from whether it went in.

This article explains how expected points work, why separating shot quality from shot making is so useful, how the numbers are used to judge offenses and defenses, and where the models fall short. If you are new to the geography of shots, begin with how to read a shot chart, and for the underlying principles of good shot selection see shot selection: what makes a good shot good.

The tracking revolution in brief

Optical tracking arrived in the NBA in the early 2010s. A camera-based system was installed in every arena by the 2013-14 season, and the league later moved to a different provider and a different system. The details of vendors matter less than the result: every player's position on the floor, and the ball's position, were captured many times a second, and every event in a game could be located and described in far more detail than a scorer could record by hand.

The effects reached well beyond shot analysis. Defensive matchups could be measured, passes could be counted by type, and movement such as cutting and screening could be quantified. But shot quality was one of the earliest and clearest applications, because it addressed a question analysts had long wanted to answer: how much of a shooter's results come from the shots he takes, and how much from how he makes them?

Before tracking, shot quality could be approximated only by location. A shot at the rim is easier than a shot from 20 feet, so a shooter who took many rim attempts had an advantage. Tracking made it possible to add context: a rim attempt with a rim protector standing under the basket is harder than an uncontested layup in transition, though both are "at the rim."

What goes into a shot-quality model

A shot-quality model is a statistical formula that takes the circumstances of a shot and returns a probability that it goes in. The general method is the same as any predictive model: collect a large number of past shots with known outcomes, find the relationship between the circumstances and the results, and apply it to new shots.

Typical inputs include:

  • Distance and angle from the basket. This is the foundation, and it alone explains a great deal.
  • Defender distance. The proximity of the nearest defender, and sometimes how many defenders were close.
  • Shot type. A catch-and-shoot attempt, a pull-up jumper, a dunk, a layup, a floater or a post shot.
  • Shot clock and game context. Late-clock shots tend to be harder because the offense has run out of options.
  • Dribbles and touch time. Shots taken after several dribbles are usually harder than those taken in rhythm off a pass.
  • Shooter identity. Some versions account for the individual shooter's ability; others deliberately treat all shooters as average to isolate the quality of the look.

The league's own tracking statistics group defender distance into categories that fans now see regularly. A shot with the closest defender six or more feet away is generally labeled wide open, four to six feet open, two to four feet tight, and under two feet very tight. These labels are a coarse version of what the full models do with continuous distance.

From probability to expected points

Once you have a probability for a shot, the arithmetic is simple. Expected points equal the probability of making the shot times its point value. If a three-pointer has a 40 percent chance of going in, its expected value is 1.2 points. If a layup has a 62 percent chance, its expected value is about 1.24 points. A contested mid-range jumper with a 42 percent chance is worth 0.84 points. These figures are invented for illustration, but they reflect the kind of ordering analysts regularly find: rim attempts and open threes at the top, contested twos at the bottom.

You can add refinements. A shot that draws a foul has value beyond the field-goal attempt, and free throws can be built in. A missed shot might be rebounded by the offense, which adds value for shots that tend to produce offensive rebounds. Some models therefore estimate the expected points of the entire possession that follows the shot. At that point you are close to the idea of points per possession, applied to a shot rather than a possession as a whole.

Summing a team's shots

If you add up the expected points of all the shots a team took in a season, you get a measure of the quality of the team's offense that does not depend on whether the shots went in. Compare it with the team's actual points from those shots and you have a picture of its shot making. An offense that creates high-quality looks but shoots poorly is a candidate for improvement. An offense that creates mediocre looks but shoots extraordinarily well may be living on borrowed time, or may have an exceptional shooter whose talent really does lift the average.

Shot quality versus shot making

This distinction is the most important idea in the whole topic, so it is worth stating plainly.

Shot quality is how good the opportunity was, on average. It is a property of the situation: where, how contested, in what flow of the offense.

Shot making is how well the shooter performed relative to that average. It is a property of the player, plus luck.

The two combine to produce results. A player can score efficiently because he gets great shots (high quality), because he makes difficult shots (high shot making), or both. They are not the same skill. A role player in a system built around spacing may enjoy a high shot-quality number because teammates create open looks for him; his own shot making may be average. A volume scorer who creates his own offense against pressure may have a modest shot-quality number and a strong shot-making number, since he takes hard shots and makes more of them than a typical shooter would.

The distinction helps with fairer comparisons. If two players both shoot 38 percent on threes, you can ask whether one had more open looks than the other. If so, the one who shot 38 percent on tougher looks is the better shooter, even though the raw percentages match. This idea is a close relative of what the true shooting percentage and effective field-goal percentage do for the value of threes and free throws: they adjust the raw field-goal percentage so that you compare like with like.

Which one persists?

Analysts who study how these two components behave from season to season find an interesting asymmetry. Shot quality is largely a product of team context and the structure of the offense, and it tends to persist when a team's personnel and system stay the same. Shot making, for most players, regresses toward the average in a way that matches the pattern of regression to the mean, with elite shooters as the notable exception: a small number of players have repeatedly shot well above what the average shooter would have made from the same looks, and those players are the reason that shot making is not purely luck. For everyone else, a season of surprising accuracy is more likely to partially fade than to continue.

Shot quality on defense

The same models can be run in reverse to judge defenses. Add up the expected points of the shots a team's opponents took and you get the quality of the shots the defense allowed. A defense that forces contested, late-clock attempts, pushes opponents into low-value areas, and denies the corners and the rim will show a low expected-points-allowed number. A defense that gives up open threes will show a high one, however well its players guard the ball.

This view cuts through a major source of noise. Opponent shooting percentages on defense depend a good deal on luck, particularly on three-pointers. Over a season, a defense can hold opponents to a poor three-point percentage because the opponents missed shots they would normally make. The quality of the shots allowed is a steadier measure of what the defense controls, and a better predictor of future defensive results than the percentages are. For more on this problem, see defensive metrics: why defense is hard to measure.

The modern league's most noticeable defensive pattern, a drive to deny corner threes and rim attempts while leaving the long two-pointer open, is in part a product of this analysis. Coaches discovered that the shots defenders can afford to give up are those with the lowest expected value, and the numbers pushed schemes in that direction. The offensive counterpart appears in our look at the three-point revolution in the numbers.

A worked comparison

Consider two hypothetical players, each of whom takes 500 three-point attempts in a season and makes 38 percent of them.

Player A takes mostly catch-and-shoot threes with no defender within six feet. The model estimates that an average shooter would make about 38 percent of those same looks. Player A's shot making is therefore average: he is getting good looks and converting them at the expected rate.

Player B takes a lot of pull-up threes with a defender close by, many of them late in the shot clock. The model estimates that an average shooter would make about 33 percent of those looks. Player B made 38 percent, five points better than expected. That difference, spread over 500 attempts, is about 25 extra makes, or 75 extra points beyond what the average shooter would have produced.

The raw percentages are identical; the stories are not. Player B is likely the better shooter, and he is carrying a harder offensive burden. Player A is the beneficiary of a system that gets him clean looks, which has real value but is a team accomplishment as much as an individual one. The right conclusion is not that Player A is bad, but that the percentages alone cannot separate the two cases.

The limits of the models

Shot-quality numbers are powerful, and they come with limitations that every careful reader should know.

  1. The model is only as good as its inputs. Some important features of a shot are not captured: the quality of the pass, the effect of a screen, how much a defender's hand disrupts a shot even when he is not close, and fatigue.
  2. Players are not average. Models that treat every shooter the same will undervalue elite shooters' open looks and overvalue weak shooters' open looks. Models that include shooter identity avoid this problem but blur the line between quality and making, since a good shooter's shot quality is partly a result of his own ability.
  3. Selection effects. Shots that are taken are not a random sample of the shots that could have been taken. Players choose their attempts, and defenders choose how to guard them. A model trained on actual shots cannot see the shots that never happened, such as the open pass not thrown.
  4. Measurement error. Tracking systems are impressive but imperfect. The recorded distance to a defender, the instant of release and the position of the ball all carry small errors, and those errors can matter on borderline classifications.
  5. Proprietary methods. Many models are built by companies and teams, and their details are not public. Two sources can produce different expected-points numbers for the same shots, and neither is wrong, since each rests on different assumptions.
  6. Small samples. A single player's shot-making score over 300 attempts has a wide margin of error. One cold month can wipe out an apparent skill, and one hot month can create one. The cautions in small samples in basketball apply with full force.

How to use the numbers as a fan

You do not need access to proprietary data to benefit from this way of thinking. A few habits carry over.

  • When you see a shooting percentage, ask what kind of shots produced it. Open catch-and-shoot threes and contested pull-ups should not be treated as the same task.
  • When a team's offense changes, look at the types of shots it takes. A team that gets more rim attempts and corner threes has made a structural improvement that does not depend on shooting luck.
  • When a defense improves, ask whether it is allowing better or worse shots, or whether the opponent's makes and misses simply changed.
  • Treat a player's one-season shot making as a hint, not a verdict, unless he has a long record of outperforming the model.
  • Remember that all of this is about the probability of a shot, not its certainty. A 40 percent three is missed 60 percent of the time, and a good process can still produce a bad night.

The public leaders and stats pages on this site, such as leaders, show many of the familiar efficiency measures, and our page on our data explains how often they are refreshed and what they include.

What to take away

Expected points and shot-quality models answer a question that box scores could not: was that a good shot? By estimating the odds of each attempt from location, defender distance, shot type and context, they separate the quality of the opportunity from the skill of the shooter, and they reveal how much of an offense's or a defense's results came from process rather than chance. The method has real limits, from measurement error to the difficulty of modeling individual shooters, but its core lesson is robust: a team that consistently creates good shots and denies them to its opponents is built on a durable foundation, while a team that merely makes or misses at an unusual rate is standing on sand.

Search more on this

Keep reading

Small samples in basketball: when a hot streak is noise

A hot week or a cold month often says less than it seems. Learn how small samples fool us, how big a sample you need for common stats, and a simple way to shrink extreme numbers toward reality.

4Basketball Staff · Jun 15, 2026 · 11 min read