Expected Shot Value
The average value of a shot from each court location, measured across the league. Defender position and shooter skill are outside this model.
Keep in mind Two shots from the same location receive the same expected value, even when the shooters and defenders differ.
On this page
Expected value by court location
The half court is divided into a grid of one-foot cells. For each cell, the model estimates the probability a shot from there goes in, and converts that to an expected effective field goal percentage, so a cell behind the arc is credited at 1.5 times a cell inside it.
xeFG% = P(make) × (1.5 if the cell is behind the three-point line, else 1)
Expected eFG% is what an average shooter converts from a given spot. The gap between what a team actually shot and what the surface expected from those same spots is shots above expected. It compares shooting results with the value expected from those locations.
A three-pointer can have a lower make probability and still have a higher expected value than a two-pointer. The eFG% scale accounts for the extra point.
Comparing the two location models
Measured 2026-07-02
The page compares two models. The zone baseline assigns every cell the average of its official zone, so its colour changes in blocky steps at zone boundaries. The gradient-boosted model uses court coordinates, so its estimates can vary within a zone.
The location model improves log-loss and Brier score by roughly one percent over the zone baseline. Those metrics assess its probability estimates; the improvement does not mean it predicts one percent more makes and misses correctly.
Both surfaces are trained on prior seasons under an expanding window, so a season is never scored by a model that has seen it. Shot efficiency drifts upward over time, which means the most recent season’s expected values can run slightly low; shots above expected for the current season are therefore biased a little high.
Shot context the model cannot see
Defender distance, shot clock, touch time, and dribbles can help describe a shot. Those inputs are absent from the location data used here. A wide-open corner three and a contested one off the dribble receive the same expected value at the same location.
Shot value estimates what a shot from this location is worth on average. It cannot judge whether a particular attempt was a good shot.
Full limitations
- —Defender distance and contest level.
- —Shot clock, touch time, dribbles, and whether the shot was assisted.
- —Shooter skill. Players with different shooting records receive the same expected value from the same cell.
- —Game context: score, period, and whether the possession was a scramble or a set play.
- —Cells with few attempts are noisy by construction. The corners of the chart carry far less data than the paint.