Home Teams Stopped Winning in 2019. The Crowds Left in 2020.

NFL home teams won by 2.4 points a game from 2016 to 2018. In 2019, with every seat sold, the edge was minus 0.14 — and the empty-stadium season that followed looked exactly the same.

The Sports Page · The Professor · on the experiment that arrived a year late

The question: does a home crowd win football games?

Probably not the way you think. NFL home teams won by an average of 2.4 points from 2016 to 2018. In 2019, in front of full stadiums, that edge was −0.14. It was already gone.

Why that is surprising. 2020 was supposed to be the test. Empty stadiums, no noise, a natural experiment nobody could have run deliberately. But the effect had vanished a year early, with the crowds still there.

What you get: a worked example of the thing that ruins most natural experiments.

What the seasons look like

Home advantage is real and old. For decades a home team won about 57 games in 100. Then this happened.

Three seasons at roughly two and a half points. Then 2019 drops to nothing. Then 2020 — the empty one — sits at +0.05, which is the same as nothing.

Then it comes back, and stays back, through five seasons of full houses.

The comparison that matters

Here is the test everybody wanted to run. Compare the season with crowds to the season without.

2019 was −0.14. 2020 was +0.05. The difference is 0.2 points, with a standard error of 1.3. That is a t of 0.15; the two seasons are not separable from each other at all.

Now compare each to the three seasons before them. 2016–2018 beats 2019 by 2.6 points (t = 2.4) and beats 2020 by 2.4 (t = 2.3). Both of those are real.

So the ordering is not crowds, then no crowds, then crowds. It is edge, then no edge, then no edge, then edge — and the crowds only left for the third of those.

What we are not going to tell you

We do not know what happened in 2019, and we are not going to invent something.

Plenty of people have theories. Better travel and sleep science. Fewer genuinely hostile buildings. Rules that moved the game toward the quarterback, who is the player least affected by noise. Officiating reviewed more closely than before. Each is plausible; none of them is measured here.

What we can say is narrower and firmer: if crowd noise were the engine, 2020 should have been the worst season on this chart. It was not. It was indistinguishable from the season before it, when every seat was sold.

What to take home: check the year before the shock

This is the trap in every natural experiment, and it catches careful people.

You notice an event — a rule change, a shutdown, a new stadium — and you compare after to before. If the number moved, the event did it.

But that only works if the number was steady until the event. Here it was not. It had already fallen, a full season early, and anyone who compared 2020 to 2018 would have found a clean two-point drop and a tidy story about crowd noise.

So before you credit an event with a change, look at the period immediately before it. Ask whether the thing was already moving. If it was, the event is a suspect and not a cause — and you have a more interesting question than the one you started with.