How many more points could an NBA player score by shifting shots toward their strongest areas?

This project estimates how a player’s scoring could change by moving shots from weaker locations toward areas where their past shots show stronger ability. You can explore how shot location affects scoring efficiency and where a different shot mix might help. This is a theoretical basketball scenario, not a real-game prediction or coaching instruction.

How to use it

Choose a season and player. Move the slider to compare their past shots with a possible new shot mix.

Moving some shots to different court areas while keeping the same total attempts. The model chooses locations without picking individual shots because they were misses. means changing where some shots happen while keeping the same total number of attempts.

Season

Search player

Open the list or type a name. Accents and punctuation are optional.

Loading seasons and players…

3 Read the court

Player chart

Choose a player

2025–26

Recorded shots at 0%: green circles show makes; rust crosses show misses.

Below 30%
30–40%
40–50%
50–60%
60% and above
The area has at least 10 recorded attempts and at least 90% model probability of beating the player’s past points per shot. An outlined diamond marks it; the 50% limit may leave no room to add shots.

Choose a player to see their shots.

These estimates assume the player can get the new shots and keep the same shooting ability. They do not show that moving shots would cause extra points in a real game. The model leaves out defense, fatigue, passing, shot creation, shot-clock pressure, and game situation.

Fig 1Choose a player.Choose a player.

Important limitations

These estimates neither predict real games nor guarantee that moving shots would cause scoring gains.
Players’ roles, coaches, teammates, plays, and available opportunities limit their control over shot locations.
The model excludes defense, contests, opponent quality, and game theory, including defensive reactions to repeated shot patterns.
It excludes game situation, score, clock, quarter, pressure, and playoff context.
It excludes shot creation, pass quality, dribbling difficulty, balance, technique, and part of free-throw value.
It excludes fatigue, injuries, travel, conditioning, and changes in health.
It holds each area’s ability fixed as locations and volume change, although historical success may decline at higher volume.
Four-foot cells simplify shot locations; the 25% slider and 50% receiving-area cap are project assumptions.
The 90% range covers uncertainty within the model, leaving out other real-world uncertainty.
The score compares a player with their own shot mix, without ranking players or grading overall talent.
One-area estimates need extra caution, and cyan diamonds mark hypothetical attempts rather than guaranteed makes.
Separate historical seasons do not predict development, aging, trades, changing roles, or team changes.
Court boundaries, player eligibility, evidence requirements, and receiving capacity prevent some player-seasons from getting estimates.
Do not use this explorer for coaching recommendations or betting.

This is a simplified look at future shot selection in an idealized basketball setting.

Definitions

Shot relocation

Moving some attempts from weaker estimated scoring areas to supported stronger areas while keeping total attempts unchanged. A shot’s made-or-missed result does not decide whether it moves.

Shots to Move

Your slider setting is the most the model may move: 25% means up to 25 of every 100 attempts, with less movement if source shots or receiving space run out. The moved count can include part of a shot because the calculation uses attempt shares.

Extra Points in the Selected Season

The estimated difference in points over the selected season’s included attempts: LeBron’s 2025–26 example at 25% shows +157 points across 919 shots. This compares modeled shot mixes without adding points to his official total or forecasting next season.

Extra Points per 100 Shots

The same estimated scoring difference expressed per 100 attempts, regardless of season volume; LeBron’s 2025–26 example at 25% shows +17.0 points. A negative value means fewer estimated points.

Shot Selection Score

This fixed 0–100 score compares the player’s past mix with feasible relocation at the 25% request: 90 means about 90% of that scenario’s expected points per shot, with higher scores indicating less estimated room to improve. It reports the median of the model’s scores and does not rank players or rate their overall offense.

Estimated Range (90%)

The middle 90% of the model’s estimates, such as 129 to 183 points around LeBron’s +157-point example; a wider range means more uncertainty within the model. It does not cover all real-game uncertainty or promise that future results will fall inside it.

Estimate from One Area

One area passes the evidence rules, so it receives moved shots until the 50% cap or available source shots stop movement. An area already holding at least half the attempts has no room for more, so no estimate appears; this label does not describe the player’s other strengths.

Make-chance map

The colors show estimated make probability across court areas, with nearby areas helping where attempts are sparse. Darker colors do not guarantee makes or qualify an area for relocation, which also requires shot-value and evidence checks.

Area with Evidence for Moving Shots

An outlined diamond marks an area with at least 10 attempts and at least 90% model probability of beating the player’s current mix in points per attempt. The 50% cap determines whether it has room to receive more shots.

No Clear Area for Moving Shots

No area passes both evidence rules, so the past shots and Make-chance map remain available without a relocation estimate. This does not prove the player’s shot choices were ideal or that no change could help.

No Room to Move More Shots

The areas that pass the evidence rules have no remaining room under the 50% limit. The chart stays available, but the model supplies no score or extra-points estimate.

Estimated Points per Shot

The modeled chance of a make multiplied by the player’s observed mix of two- and three-point values within an area. This measures expected scoring per attempt, not make percentage alone.

How the model works

Data and seasons

This project analyzes regular-season NBA field-goal attempts from 2021–22 through 2025–26. Each season has its own analysis. The model does not combine a player’s shots across seasons. Players need at least 250 included attempts across 20 games in that season. The analysis includes shots inside its half-court boundary and excludes shots beyond it. The five seasons contain 1,507 player-season analyses. A player appearing in two seasons counts as two analyses.

Estimating shooting chances

The court contains 156 areas, each about four feet across, with smaller areas at the edges. The model uses made and missed shots to estimate each player’s shooting chances across these areas. Nearby areas help estimate chances where the player took few shots.

The model connects court areas that share an edge and keeps a range of plausible shooting chances. It combines shot evidence with assumptions about nearby areas. Its full name is a Bayesian conditional autoregressive model.

Each player has their own shooting map. Players share how strongly the model smooths neighboring areas, not one common shooting pattern. Limited shot samples leave uncertainty. The Bayesian approach tracks that uncertainty instead of treating a small sample’s shooting percentage as exact. An area with no recorded attempts can have an estimated shooting chance, but cannot pass the relocation evidence rule.

Finding areas that can receive shots

An area needs at least 10 of the player’s recorded attempts in that season. It also needs at least 90% model probability of beating the player’s past mix in expected points per attempt. Expected points combine the chance of making a shot with its two-point or three-point value. The calculation uses the player’s observed mix of twos and threes within each area. It compares the player with their own shot mix, rather than league-average shooting.

Zero, one, or several areas can pass. With zero, the model gives no relocation score or points estimate. One area can support an estimate if it has room for more shots under the cap. No receiving area can finish with more than 50% of the player’s attempts after adding shots. This limit does not remove historical shots from an area already above 50%; that area cannot receive more shots.

Choosing which shots move

The calculation ranks occupied areas by their estimated points per attempt, from weakest to strongest. It moves shots from areas below the player’s past mix, starting with the weakest. Made and missed shots help estimate shooting ability. An individual shot’s outcome does not decide whether that shot moves.

The slider requests 0%, 5%, 10%, 15%, 20%, or 25% of attempts. Movement stops at the request, when suitable source shots run out, or when receiving areas have no room. The calculation permits part of a shot at the final boundary. A faded marker represents that part.

With several receiving areas, the initial split follows how often the player shot from those areas. If one reaches 50%, the remaining moved shots go to other supported areas with room, using their existing usage proportions. With one area, it receives the moved shots until its cap applies. At least 75% of attempt volume stays in its historical locations at the largest request. The same slider setting gives the same markers. Higher settings retain the shots already selected at lower settings.

Estimating points and uncertainty

The calculation freezes the movement plan before testing it across 4,000 possible shooting-ability estimates from the fitted model. It uses the same possible abilities for both the past and changed shot mixes in each comparison. This gives estimated differences in points over the season’s included shots and per 100 attempts.

The extra-points numbers are averages across those 4,000 comparisons. The score uses the middle of the 4,000 scores, after limiting each score to 0–100. Each displayed range runs from the fifth to the ninety-fifth percentile: the middle 90% of those values.

How the model was tested

The 2025–26 comparison used separate groups of games for fitting, choosing court size, and the final test. Shots from the same game stayed together. The final test used 39,212 shots from 246 games unseen during fitting or selection. The selected model beat a simpler smooth-curve model on the predeclared test of shooting predictions. The test found no clear disadvantage in how its predicted shooting chances matched observed make rates. It found no clear uncertainty advantage for players with fewer shots.

After testing, the selected model used all included season shots for the published estimates. The earlier four seasons reuse the same specification with separate fits. They do not repeat the model comparison. Checks covered missing results, shot counts, repeatable movement, uncertainty ranges, and the 50% cap.

Project details

Data: Recorded NBA field-goal attempts in five separate regular seasons.
Approach: Estimate shooting chances by location, then test a limited change in shot mix.
Seasons: 2021–22 through 2025–26.
Analysis code: View the analysis on GitHub.

Formulas

These equations describe one of the model’s 4,000 joint shooting-ability estimates. The calculation uses the same make probabilities for the historical and relocated shot mixes in each comparison.

Variable key

c, s: A court cell and the requested slider share (0%, 5%, 10%, 15%, 20%, or 25%).
A: Total included shot attempts in the selected season.
p_c: Modeled make probability in cell c for one joint simulation.
v_c: Cell point value: 2 plus the observed share of three-point attempts in that player-season cell.
w_c, w′_c(s): Historical and relocated shares of all included shots in cell c; the prime (′) marks the changed distribution at request s.
E_c, EPPS: Expected points per shot in a cell, and the abbreviation for expected points per shot.
current, relocated: Subscript labels for the historical and feasible changed shot mixes.
Season gain, Gain per 100: Extra points over A season attempts or over 100 attempts at request s.
Score: One simulation’s comparison of current and feasible 25% expected points per shot, before the 0–100 limit.
Σ_c: Add across court cells; a subscript c identifies the cell.
Q_0.05, Q_0.95: The fifth and ninety-fifth percentiles of the 4,000 calculated gains or limited scores.
Range_90%: The interval between those percentiles: the middle 90% of calculated values.
=, ×, −: Equals, multiply, and subtract; a fraction bar means divide.
( ), [ ], %: Parentheses and brackets group terms; (s) means evaluated at request s. In a range, brackets enclose its lower and upper endpoints. Percent (%) means a share out of 100.

Equations

1. Expected points in a cell

Multiply the chance of a make by the cell’s two-point and three-point value.

2. Current expected points per shot

Weight each cell’s expected points by how often the player shot there.

3. Expected points after relocation

Use the changed shot shares while keeping the same estimated shooting ability.

4. Extra points in the selected season

Apply the difference per shot to the player’s included season attempts.

5. Extra points per 100 shots

Express the same difference over 100 attempts, regardless of season shot volume.

6. Shot Selection Score

Compare the past mix with the feasible mix at the largest slider request.

The calculation limits each of the 4,000 scores to 0–100, then displays their median. Dividing two displayed averages does not reproduce that median of ratios. The 25% request uses the actual feasible movement when source shots, evidence, or the 50% destination cap prevent moving all 25%.

7. Estimated Range (90%)

Keep the middle 90% of the model’s 4,000 calculated values.

Gains display the average across those values. Scores display the median after the 0–100 limit. The ranges reflect uncertainty in modeled ability, not a simulation of future makes and misses.

Where the values come from

Make probability (p_c): The player’s season-specific spatial shooting model supplies 4,000 joint estimates across court cells.
Point value (v_c): For each player and season, divide three-point attempts in a cell by all attempts in that cell, then add 2. A cell with only twos has value 2; one with only threes has value 3. A cell crossing the arc uses the player’s observed mix, not its center’s location. Empty cells have no observed point value and receive no shot share; the calculation sets their zero-weight contribution to the sums to zero.
Current shares (w_c): Divide attempts in each cell by the player’s total included season attempts.
Relocated shares (w′_c(s)): Remove shots weakest first from below-baseline areas. Add them to supported areas in proportion to existing usage, redistributing when an area reaches the 50% cap. Freeze this plan before evaluating the 4,000 estimates.
Feasible movement: Use the smallest of the requested share, eligible weak-source share, and supported receiving capacity. Each supported cell’s added capacity is 50% minus its historical share, or zero if that share is already 50% or more; add those capacities across supported cells.
Uncertainty: Apply the fixed shot shares and point values to each joint model estimate, then summarize the resulting gains and scores with the ranges above.

Unavailable relocation keeps scores and gains missing, including at 0%. None of these equations turns missing evidence into a zero estimate.