Do Referee Crews Move Totals? 7,275 Games Say No

The Folklore This Page Tests

Every betting season, the same content resurfaces: "Crew X games go under." "Referee Y is a home-team whistle." The claim behind it is that officiating crews have measurable, persistent effects on totals and margins — effects big enough to bet on once the referee assignment is announced. It's a testable claim, so this page tests it against 7,275 played regular- and postseason games from 1999 through 2025 with a named referee in the nflverse game file.

The short answer: the raw leaderboards that fuel the folklore are real numbers producing a fake conclusion. Adjust for when each referee worked, and the crew effect on totals collapses to exactly what a random-number generator would hand you.

The Raw Numbers Look Convincing. That's the Trap.

Take every referee with at least 150 games and compute average combined points. You get a leaderboard with a genuinely wide spread — over five points from top to bottom, against a 44.3 league average:

Five points is enormous — half a standard total-line move is half a point. If Blakeman games really produced 5 more points than Carey games, you'd bet it blind. But look at the years column. Carey's career ended in 2013; Torbert's began in 2015. NFL scoring environments moved substantially across those windows — the league's combined average ranged from the low 40s in the early 2000s to roughly 47 in 2020. A referee's raw average mostly tells you which seasons he was employed in, not what his crew did to games.

The Fix: Compare Each Game to Its Own Season

The adjustment is one line of arithmetic: for every game, subtract that season's league-average total from the game's total, then average those differences per referee. A crew that genuinely inflates scoring should stay positive after the adjustment; a crew that merely worked in 2020 should fall back to zero.

Here is what survives, for all 21 referees with 150+ games. The noise math matters, so it's in the table: a single NFL total has a standard deviation of about 14.2 points, so even a 250-game career average carries a standard error near ±0.9 points. Anything inside roughly two standard errors of zero is indistinguishable from chance.

The verdict row: zero of the 21 crews clear two standard errors. With 21 independent tests at that threshold, pure chance would be expected to produce about one false positive — the data can't even manage that. Mike Carey, the rawest "under referee" in the sample, lands at 1.9 standard errors after spending his entire career in the lowest-scoring stretch of the data; the raw 5.1-point spread compresses to a distribution centered on zero that looks exactly like sampling noise around a true effect of nothing.

The Home-Whistle Version Fares Barely Better

The margin-side folklore — "crew Y favors home teams" — gets the same test: each home-site game scored 1 for a home win, 0 otherwise, minus that season's league home-win rate (the league baseline is 56.5% across the sample, and it has declined by era, which is why the season adjustment matters here too).

Two crews of twenty clear the two-standard-error bar: John Hussey's games ran about 11 points of home-win percentage high, Bill Leavy's about 9 low. Before anyone builds a betting model on two names: twenty tests at a 5% threshold are expected to produce one hit by chance, and getting two is the kind of thing that happens in roughly a quarter of random re-runs. It's weak evidence at best — and neither name repeats on the totals list, which is what a real officiating effect would be expected to do.

What This Data Can't See (Honest Scope)

This exhibit is built from final scores, not penalty logs — the nflverse game file used here carries no flag counts. It is well documented in public charting that crews differ in penalty tendencies: some throw more offensive-holding flags, some call defensive pass interference at higher rates. Nothing on this page contradicts that. The claim being tested — and rejected — is narrower and more actionable: that those tendencies aggregate into a points-level effect on totals or margins big enough to see in 150–350 game samples. They don't. Flags apparently reshuffle how games unfold more than how many points they end with.

One more scope note: referee assignments aren't random. Senior crews draw more primetime and playoff games, which skew toward better teams. That's another reason raw crew splits mislead — and another confounder the season-adjustment only partially removes.

How to Actually Use This

• Ignore raw referee totals stats. They are era artifacts. Any site quoting a crew's career over/under record without a season adjustment is selling noise.
• Treat referee-news line moves as sentiment. If a total moves on an assignment announcement, the move itself is information about the market, not about the game.
• Expect regression. A crew running hot on unders over 30 games is 30 draws from a 14-point-per-game noise distribution — a season's worth of games can't establish a crew effect even if one existed at the ±1 point scale.
• Watch the right level. If crew effects matter anywhere, it's in penalty-rate props and live in-game tendencies, not full-game totals — and testing that requires flag-level data, not this file.

For the general framework — why per-game noise swamps small true effects, and what sample sizes can and can't detect — the same machinery runs through the spread-accuracy exhibit (13.2-point average miss on an unbiased line) and turnover luck. For the era context that drives the raw leaderboard, see how NFL scoring has moved and the decline of home-field advantage; for what totals lines are made of, the over/under explainer.

Reproduce It

Everything above computes from one public file: games.csv from the nflverse nfldata repository (bundled at /data/games.csv with source notes). Filter to played games with a non-empty referee, group by referee, and average total; for the era adjustment, subtract each game's season-average total first. The home-win test uses result > 0 at home sites against each season's home-win rate. Standard errors are the per-crew standard deviation over √n (totals) and 0.5/√n (win proportions). All figures on this page were computed 2026-07-09 on the 1999–2025 file (7,275 games with a named referee); numbers will drift slightly as 2026 games append.

Data: nflverse/nfldata, public. Referee names as recorded in the game file; a "referee" here means the named crew chief, standing in for the whole crew.