Hunting for Weather-Influenced Upsets

Our last post found that weather doesn't change who wins, on average. But "on average" can hide stuff. What if bad weather causes the occasional wild upset, just in both directions, so they cancel out when you zoom out? We went hunting for those games. What we found is a great lesson in not fooling yourself.

How do you even measure a "surprise"?

First we needed a guess for what should have happened. We trained a simple model to predict each game's final margin using only how good the teams were (their SP+ ratings), with nothing about weather at all. And we made it play fair: it only ever learned from games it had already "seen," the same way you'd have to predict a game before it's played. The gap between what actually happened and what the model expected is that game's surprise, in points.

Here's the humbling part. Even on a sunny day with zero weather drama, the typical game surprises the model by about 11 points. Football is chaos. That number matters because it's the bar a "weather upset" has to clear to mean anything. An 11-point surprise isn't a story. It's a Tuesday.

Does bad weather make games harder to predict at all?

Before picking out individual games, we asked the bigger question: are surprises just bigger when the weather's bad, even if they go both ways? If weather were a "great equalizer," this is where it would show up.

None of those differences are big enough to trust. (A statistician would say they're "not statistically significant," which basically means they could easily be random luck. We checked.)

So here's the honest answer: weather doesn't make games less predictable, at any level of "bad" we tried. Which means the list below is exactly what it looks like: the biggest upsets that happened to land on a rough-weather day, not proof that rough weather causes upsets. We're sharing it anyway because it's a fun list of games, not because we think the weather did it.

The list

The 30 biggest gaps between "what team strength said should happen" and "what actually happened," among games with measurable rain, any snow, or wind of 15+ mph. An Upset badge means the team the model favored didn't just win by less than expected. It lost.

What would change our minds

One list of upsets proves nothing on its own. That's exactly why we checked the big-picture question first. To turn this from "fun curiosity" into "real finding," we'd want to see the same pattern again and again in separate chunks of data: different seasons tested on their own, real pre-game weather forecasts instead of what the weather actually did, different levels of competition. We haven't seen that. If a future re-run with more data finds something real, you'll read about it here first.