College football rain and totals

This study first published on a 1,441-game sample and read as a null result. We have since completed the historical odds backfill — including the 2023–2025 seasons, which a second odds source turned out to cover after our first source came up empty — and filled every remaining gap in archived game-window weather. The sample is now 7,072 games across all ten seasons, with 960 wet. On the larger sample the answer changed: light rain is still nothing, but the heavy-rain band now clears both the pre-existing 40-game minimum and our significance threshold. The original null headline has been restated accordingly, and the earlier figures are superseded rather than quietly dropped.

Across 7,072 outdoor games from 2016 to 2025 that carry a closing total, rain matters only once there is enough of it. Any measurable rain moves scoring -1.2 points on a season-adjusted basis (t = -2.02, only marginally past our threshold and driven almost entirely by the heaviest games) with the Over rate at 46.4% against 49.9% for dry games. Split by amount, the 71 games with 2.5 mm or more score -5.1 points (t = -2.81) and finish -3.9 points under the closing total (t = -2.35). Rainfall amount carries a mild gradient of about -0.9 points per millimetre.

A summary card of the headline numbers, free to republish with a visible credit and a link back to this page.

Across more than 7,072 outdoor college football games that had a closing total, only genuinely heavy rain — 2.5 mm or more during the game window — reliably lowered scoring. Those 71 heavy-rain games averaged -5.1 points fewer than dry games and finished -3.9 points under the closing total (t = -2.35), passing our pre-set significance threshold. The simplest explanation is not that books ignore rain, but that they price light rain correctly and move too slowly once a reliable heavy-rain forecast is in place. We have not yet run the forecast-timing study ourselves, so we report the historical closing-line result and leave the mechanism as the leading hypothesis.

The smaller figure in each cell is the season-adjusted gap against the dry control, with its t statistic. Roughly 2 is our threshold: the trace and steady rows fall short of it and are not edges we would price, while the heavy row clears it on both points and margin against the line. Bands under 40 games are greyed out and flagged; the heavy row (n=71) clears that pre-existing minimum. The "any measurable rain" row pools trace, steady and heavy games, which is why it reads weaker than heavy rain alone — most wet college games are barely wet.

Scoring, closing totals and the share of wet games all drift year to year — season average totals fall from about 57 in 2016 to about 52 in 2025. That is why every comparison above measures a game against its own season rather than against the pooled sample.

Coverage is now even across the span: our archive holds results, game-window weather and a closing total for every season 2016–2025, with only scattered games missing a line because no source recorded one. Nothing is filtered out for what it showed.

• Heavy rain is the finding. The 2.5 mm-and-up band (71 games) scores -5.1 points against the dry control with t = -2.81, and lands -3.9 points under the closing total with t = -2.35. Both clear our threshold, on a sample that clears our 40-game minimum.
• A trace of rain is not information. The under-0.25 mm band (481 games) scores -0.5 against the control. A radar smudge at kickoff tells you nothing about the total.
• Steady rain is not enough either. The 0.25–2.5 mm band (408 games) comes in at -1.5 points with t = -1.61 — the right direction, short of significance.
• Pooling all wet games understates the effect. Any measurable rain reads -1.2 points (t = -2.02) because most wet college games are barely wet. Amount is what matters: about -0.9 points per millimetre across 960 wet games.
• The market prices light rain, not heavy rain. Across all wet games the margin against the line is -0.4 points (t = -0.66) — nothing. In heavy rain it is -3.9 points, which does clear significance. The most plausible reason is slow adjustment to reliable heavy-rain forecasts, not a wholesale miss on rain. We have not validated that timing claim independently, so it is framed as the working explanation rather than a proven fact.

• Sample: 7,072 outdoor college football games, 2016–2025, each with a final score, a closing total and game-window rainfall. 960 of them had measurable rain. Indoor venues and closed roofs are excluded entirely rather than used as a control.
• Weather: hourly reanalysis at the venue's own coordinates, read across the game window rather than off a single snapshot. Rainfall is archived model data, not a gauge reading at the stadium, so small amounts carry more uncertainty than large ones. As of September 2026 this weather coverage is complete for every game in the archive.
• Control group: no measurable rain. Every figure is season-adjusted — each game against its own season's average, for both points and margin against the closing total.
• Odds: closing lines come from a sports data feed rather than a single sportsbook, and for a completed game the recorded number is the last consensus line that feed held. Coverage now spans all ten seasons; a few hundred games league-wide have no line in any source and are simply absent. Live in-play prices are excluded.
• Forecast lag is a hypothesis, not a proven mechanism. The study finds that heavy-rain games closed too high relative to the final score. The working explanation is that the market adjusts more slowly than the weather forecast, not that it ignores rain. We have not yet run the open-to-close line-move study required to confirm that timing story, so the page states the historical result and treats forecast lag as the leading candidate.
• The heavy-rain band is 71 games — past our 40-game minimum and past our significance threshold, but still a modest sample for a result this size. We will re-check it every season and restate it if it fades.
• The 40-game minimum and the significance threshold are house rules applied to every study before results are seen, not thresholds chosen to admit this row.
• Not controlled: team quality, pace and pass-run balance, wind arriving alongside the rain, field surface, and snow, which reanalysis reports as precipitation rather than separately.
• This study reports Over/Under rates because closing totals are attached. Where they are not — our MLB and NFL wind work — we report points only.

Data generated 2026-10-29. Numbers refresh as the archive and odds backfill top up; corrections are dated and kept visible.

• Open-Meteo ERA5 historical reanalysis — archive-api.open-meteo.com (hourly game-window weather at each venue’s own coordinates)
• Open-Meteo geocoding API — open-meteo.com/en/docs/geocoding-api (venue coordinates and elevation)
• ESPN college football scoreboard and odds — espn.com/college-football (schedules, final scores and consensus closing totals)
• Internal college football game archive — venue coordinates, final scores and closing totals.
• This study has been corrected since first publication. Every dated change is listed on our corrections log.