Does the whistle
favor stars?

Backlog idea #49 ("Do refs favor stars? Foul calls vs player fame/salary — [real bias]"), previously untouched. Basketball-Reference's own 2025-26 NBA season per-game box score stats (the most recently completed season as of this run — 2026-27 is only a few weeks old, too thin for stable per-36-minute rates) joined to each player's current, already-signed 2026-27 salary from Basketball-Reference's own contracts page — the best available single-row, keyless proxy for how much the market values and knows a player. 318 players clear a reliable-rate floor (G≥20, at least 15 minutes per game); traded players' per-team splits were collapsed to Basketball-Reference's own combined-team row so nobody is double-counted.

Both ends of the claim hold, and the direction is the one the backlog's own tag predicted. Personal fouls committed per 36 minutes fall as salary rises: -0.234 fewer fouls per log-salary point, 95% CI [-0.314, -0.154] — excludes zero, R²=0.094, p=2.3e-08. Free throw attempts per 36 minutes rise: +1.002 more per log-salary point, CI [+0.807, +1.196] — excludes zero, R²=0.246, p=4.1e-21. Higher-salary players get fewer fouls called on them on defense and more fouls called for them on offense — both directions a "star treatment" story would predict.

Neither gap is just shot volume or position in disguise. Controlling for field goal attempts per 36 (how much offensive load a player carries) and whether the player is a traditional post position (PF/C, who foul more just by playing interior defense), both slopes shrink but hold: personal fouls -0.172, CI [-0.264, -0.079] — still excludes zero; free throws +0.433, CI [+0.222, +0.644] — still excludes zero, meaning even two players taking the identical number of shots per 36 minutes, the higher-paid one draws more fouls on those possessions.

Spec	n (players)	slope / log-salary pt	95% CI	R²	Verdict
Personal fouls/36, naive	318	-0.234	[-0.314, -0.154]	0.094	excludes zero, p=2.3e-08
Personal fouls/36, HC3	318	-0.234	[-0.310, -0.158]	0.094	excludes zero, p=1.5e-09
Personal fouls/36, + FGA/36 + position, HC3	318	-0.172	[-0.264, -0.079]	0.199	excludes zero, p=0.00029
Free throws/36, naive	318	+1.002	[+0.807, +1.196]	0.246	excludes zero, p=4.1e-21
Free throws/36, HC3	318	+1.002	[+0.788, +1.215]	0.246	excludes zero, p=3.4e-20
Free throws/36, + FGA/36, HC3	318	+0.433	[+0.222, +0.644]	0.425	excludes zero, p=5.8e-05
Personal fouls/36, G≥50 only, HC3	261	-0.233	[-0.320, -0.147]	0.089	excludes zero, p=1.4e-07
Free throws/36, G≥50 only, HC3	261	+0.936	[+0.698, +1.173]	0.211	excludes zero, p=1.2e-14

A second, independent measure of fame agrees. Salary is a continuous proxy; being named to the 318 players' 28 2025-26 All-Star Game is a discrete, reputational one that doesn't depend on a front office's contract math. The two measures tell the same story:

Rate	Non-All-Star (n=290)	All-Star (n=28)	Gap	Bootstrap 95% CI	Verdict
Personal fouls / 36 min	2.99	2.54	-0.45	[-0.69, -0.21]	excludes zero, Welch p=0.0011
Free throw attempts / 36 min	3.17	6.91	+3.74	[+3.03, +4.48]	excludes zero, Welch p=6e-11

All-Stars draw more than double the free throws per 36 minutes of everyone else (6.91 vs. 3.17) and commit noticeably fewer fouls (2.54 vs. 2.99).

Robustness. Restricting to the 261 players who appeared in at least 50 of their team's games — full-season regulars, not partial-season call-ups or players returning from injury — barely moves either naive estimate: personal fouls -0.233 (CI [-0.320, -0.147]), free throws +0.936 (CI [+0.698, +1.173]). Spearman rank correlations agree in sign and significance with every OLS spec above (ρ=-0.308 for fouls, +0.448 for free throws, both p<1e-7).

Read plainly: in this one season, higher-paid and more decorated NBA players get a measurably different whistle than everyone else, in both directions, and it is not fully explained by how much they shoot or where they play on the floor. What this design cannot do is separate a biased whistle from a real, legitimate mechanism pointing the same way — star players draw extra defensive attention by design (more contact initiated against them) and are more frequently used in isolation plays that are whistled more leniently on D for everyone involved. A single-season, no-play-by-play design can observe the gap honestly; it cannot adjudicate why it exists.