Primetime Risk: Do Favourites Wobble Under the Lights?

Everyone has a primetime team they do not trust. The lights, the short week, the national audience, the coach who cannot handle the moment — it is one of the most durable stories in football, and survivor advice repeats it every September: fade the big favourite on Thursday or Sunday night. We tested it against real prices across twenty seasons. It is not there.

What we tested

The hypothesis: favourites in standalone primetime windows win less often than their de-vigged market price implies. Not "primetime games are weird" — specifically, that the price fails to account for whatever primetime does to a team.

How we tested it

Primetime is classified from the real kickoff fields on every game — nflverse's weekday and local gametime, with zero blanks across 2006 to 2025. A game is primetime if it is played on any day other than Sunday, or if it is a Sunday game kicking off at 18:15 local or later, which is Sunday Night Football. That gives 1,230 real primetime games against 4,051 non-primetime games, every one of them with a de-vigged bookmaker moneyline and a real final score.

For each game we take the favourite's de-vigged win probability and its actual result, sum the probabilities to get expected wins, and compare that to actual wins as a Poisson-binomial z-score. Ties are excluded. The same procedure runs on the non-primetime games as a control, and the two are then differenced.

The first check: is primetime even priced differently?

Before testing anything predictive, we ran the collinearity check that killed the defensive-line finding. If primetime games were systematically closer or more lopsided than Sunday afternoon games, a naive test would pick that up and we would be measuring the schedule, not the lights.

They are not. The mean de-vigged favourite win probability is 0.6673 in primetime and 0.6681 outside it — a gap of eight ten-thousandths. The market does not treat a Thursday night favourite as a different animal from a one o'clock favourite. That is a real result in its own right: there is nothing to strip out, so the residual test is just the raw test.

What the numbers said

Primetime games

1,230

Control games

4,051

Pricing gap

0.6673 vs 0.6681

Difference

z = +0.92

Primetime favourites finished +0.44 sigma above their implied win rate. Non-primetime favourites finished -1.11 sigma below theirs. The difference between the two arms is z = +0.92 — comfortably inside noise, and worth pausing on because of its sign: if anything, favourites are marginally better under the lights than in the afternoon, which is the exact opposite of the premise being tested. We are reporting that honestly rather than burying it, but a +0.92 is not evidence of anything. It is what a coin flip looks like when you squint.

No pattern inside the buckets either

A real effect that is diluted overall usually shows up somewhere in the distribution — big favourites cracking, or small favourites holding. Sorting primetime games by the favourite's implied win probability produces nothing of the sort.

• 50-60% favourites (351 games): 53.9% actual vs 55.6% implied, z = -0.67.
• 60-70% favourites (437 games): 65.9% actual vs 64.6% implied, z = +0.58.
• 70-80% favourites (307 games): 77.5% actual vs 74.7% implied, z = +1.13.
• 80%+ favourites (135 games): 83.7% actual vs 84.4% implied, z = -0.23.

The signs alternate: down, up, up, down. That is not a dose-response curve, it is four independent noisy samples. The largest single bucket z is +1.13 on 307 games, which is exactly the size of wobble you should expect when you cut a null result four ways. The non-primetime control behaves the same way, including a -1.98 in its 60-70% bucket that is likewise not survivable after multiple comparisons.

What it means for your picks

Do not downgrade a survivor pick because the game is at night, and do not upgrade one either on the strength of the mildly positive number above. If a primetime game is a good pick on win probability and pick popularity, it is a good pick. The schedule slot is not information the market has left on the table.

One practical note that survives this finding: primetime games kick off after most of the Sunday slate, which matters for pool timing and for late-week injury news, and that is a real edge we have measured elsewhere. It has nothing to do with the lights.

Limits

The classification threshold is 18:15 local, which is a judgement call at the boundary — a handful of late-afternoon Sunday games sit just below it. Playoff games in standalone windows are included. And 1,230 games sounds large but only supports detecting a fairly sizeable effect; a genuine one-percentage-point primetime penalty would not be visible here. What we can say is that the big, actionable version of the story that survivor advice actually tells does not exist.