Which NBA scorers look most matchup-dependent
Across defenders he faced for at least 25 possessions, Jalen Brunson's points per matchup possession ranges from 0.065 to 0.842, and Shai Gilgeous-Alexander's from 0.072 to 0.852. Those are the widest spreads in the data. They look like evidence that some scorers live or die by the matchup, but the reliability of these splits is only about 0.61, so a substantial part of every spread below is noise rather than matchup dependence.
6 min read · Data as of 7 June 2026 · Analysis 27 July 2026
The widest spreads in the data
For every scorer, take each defender he faced for at least 25 matchup possessions, compute his rate against each, and measure how much those rates vary. Restricting to scorers with at least six such defenders leaves a small group, listed below by the spread of their rates.
| Scorer | Defenders | Overall rate | Spread | Lowest | Highest |
|---|---|---|---|---|---|
| Jalen Brunson | 9 | 0.328 | 0.261 | 0.065 | 0.842 |
| Shai Gilgeous-Alexander | 9 | 0.392 | 0.245 | 0.072 | 0.852 |
| Donovan Mitchell | 8 | 0.356 | 0.216 | 0.100 | 0.669 |
| Jakob Poeltl | 7 | 0.155 | 0.213 | 0.029 | 0.623 |
| Dillon Brooks | 9 | 0.301 | 0.207 | 0.068 | 0.637 |
| Tyler Herro | 7 | 0.205 | 0.191 | 0.060 | 0.566 |
| Scottie Barnes | 12 | 0.296 | 0.182 | 0.000 | 0.579 |
| Devin Booker | 6 | 0.273 | 0.178 | 0.153 | 0.645 |
The ranges are dramatic. Brunson's best matchup rate is roughly thirteen times his worst. Read quickly, the table says these scorers are enormously sensitive to who guards them.
Now apply the reliability number
A split-half test on this data puts the reliability of a matchup split near 0.61. That figure has to be carried into this table, because spread and noise are not separable by looking. A scorer with genuinely stable matchup effects and a scorer who happened to have one cold night against a defender he met three times will both appear here.
The lowest values are where this bites hardest. A 0.000 against a defender does not mean that defender erased Scottie Barnes; it means that across those particular possessions no points were scored, which the small-samples data shows is a common outcome rather than a rare one.
Why publish it at all
Because the spread is real as a description, and description has value as long as it is not dressed as explanation. Knowing that Brunson's observed rates ranged this widely tells you where to look on film, which possessions to pull, and which matchups are worth a closer question.
The alternative, publishing nothing until causation is established, would mean publishing nothing at all. The workable standard is to state what was measured, state what it cannot support, and let the reader carry both.
One note on the names
Jakob Poeltl on a list of variable scorers is a good reminder of what the unit does. He is a low-usage centre whose rate is small in absolute terms, so modest swings in points produce a large proportional spread. Ranking by spread will always surface low-usage players alongside high-usage ones, and that is a property of the measure rather than a finding about them.
How these figures were produced
- Brunson 0.065 to 0.842 across nine defenders; widest spreads as tabled
- Aggregate to scorer and defender pairs, keep pairs with 25 or more possessions, then per scorer take the count, the pooled rate, and the standard deviation, minimum, and maximum of per-defender rates. Restricted to scorers with six or more qualifying defenders.
Source is NBA player-matchup tracking, 24,974 aggregated rows, last refreshed 7 June 2026. Figures describe possessions already played. Nothing here predicts a future result, and no odds, lines, or selections are published anywhere in this section.
Related research
Do NBA matchup splits repeat? A split-half test
Splitting each NBA pairing's games in half and correlating the two halves gives r = 0.44. Corrected for length, reliability is about 0.61: real signal, and roughly 40% that will not repeat.
ReadWhat NBA matchup tracking actually records
NBA matchup tracking assigns the nearest defender, not the scheme. It records proximity and outcome, and leaves out help, coverage, and intent, which is where most misreadings begin.
Read