Referee research archive

Observed wins and model expectation. Across the 913 playoff games played since 2015-16, Chris Paul’s teams won 1 of the 11 that Scott Foster worked. Expected wins come from a model fitted to playoff team-games using both teams’ regular-season strength and home court. This adjusts for those factors but does not fully account for how officials are assigned.

Opponents faced with Foster had a slightly lower mean regular-season win rate than those faced without him: 0.599 against 0.606 by win rate. Of the 689 official-and-player pairs with at least 10 shared playoff games, this one ranks #1 in this dataset.

four other players finished above their expected wins. Same official, same seasons, same method that produced the figure above.

OG Anunoby, Miles McBride, Josh Hart and Mitchell Robinson all cleared their expectation by four wins or more in games worked by the same official while Eric Gordon won 1 of 10. Comparing many pairs produces both positive and negative extremes.

Marcus Smart and Tony Brothers have a larger shortfall in wins than the featured pair. A larger raw shortfall does not necessarily mean a smaller p-value, because sample size and expected wins also differ.

Chance also produces extreme records. Across 689 pairs, the chance calculation's expected minimum p-value is 0.00145. Scott Foster and Chris Paul come in at p = 0.0016, less extreme than that reference. The expected minimum is a comparison scale, not a corrected significance threshold or a guarantee about what chance will produce.

In the full grid, 7 pairs clear p < 0.01 where chance predicts 6.9, and 27 clear p < 0.05 where chance predicts 34.5. These counts do not exceed their chance expectations. They provide no aggregate evidence of excess extreme pairs under this comparison.

A claim found by looking can only be confirmed on games nobody had seen when they found it. This one was in circulation by the end of 2019-20. Before then,Chris Paul was 0–6 with Scott Foster, and after it was famous, 1–4.

Both periods are too small for the declared test. Both halves fall under the 10 shared games this page requires before it will evaluate a pair. That threshold was written down before the analysis. The available later games therefore cannot provide the planned confirmation test.

Consecutive calls switch teams more often than after shuffling. Across 498,007 consecutive foul pairs, the next whistle goes against the other team 52.4% of the time, against 50.5% when you shuffle the same game’s calls into a random order, with t = 27.4.

Shuffling also removes possession order. After an offensive foul, the fouling team turns the ball over and defends. If possession drives the pattern, the next foul may be on that same team; a simple make-up-call explanation predicts a switch.

The reversal after offensive fouls is consistent with possession changes. It weakens the simple make-up-call explanation but cannot rule out individual make-up calls.

Two first-quarter fouls are associated with fewer minutes. They occur in 6.2% of 21,400 star-games, with 1.17 fewer minutes than the player's own average. Across 73 officials the spread is 1.01× what random assignment produces, at p = 0.4535. The rates are 6.2% on the road against 6.1% at home, a gap of 0.09 percentage points.

Three officials work every game and the play-by-play never records which one made a call, so every figure on this page credits all three and cannot isolate an individual effect. None of it separates a correct call from an incorrect one. The questions, the thresholds and the five named claims were all fixed in writing before the playoff data was fetched (ml/referee_player_preregistration.md). Results are published even when they do not meet their declared evidence thresholds.