The Contacts That Break Epidemic Models Are the Ones Nobody Can Measure

Think about the people you'll touch today. Most of them you could name in advance. The family at breakfast. The colleagues at the office, or the kids in the classroom. This is roughly how the people who forecast epidemics see the social world too, and it's a sensible way to see it. They sort our contacts into a few tidy boxes: the household, the school, the workplace, and one last box for everything else. If you know who shares your home, your classroom, your desk cluster, you know most of who you breathe near. And if you know who breathes near whom, you can trace how a virus moves. That logic sits underneath every outbreak forecast you have ever seen on the news.

A team in Germany decided to check whether the logic holds. They took a careful, large-scale record of how people actually mix, real people noting how many others they had spoken to that day and where, and laid it beside the contacts that a large epidemic simulation invents on its own. For the household, the school, the workplace, the two lined up reasonably well. Children clustered with children; working-age adults mixed mostly with each other. The model was not fooling itself there.

Then there was the fourth box. The bus. The market. The line at the coffee shop. The stranger on the next park bench. Here the simulation and reality came apart badly. The mismatch in that one category ran more than five times larger, on average, than in any of the orderly ones. The model could picture the classroom. It had no idea what happens on the sidewalk.

That gap is not academic. The researchers fed each version of the contacts into the same disease model and watched a single number move, the one every public-health official stares at during an outbreak. It says how many new people, on average, each infected person passes the virus to. Above one, the thing grows; the higher it climbs, the faster it spreads and the more people you must vaccinate to halt it. From the real-world contacts, that number came out at 3.82. From the model's invented ones, 2.79. A 37 percent gap. Not a rounding error. A different epidemic.

The model did fine on the boxes you can count. What sank it was the box you can't. Models are built from things that arrive in lists: addresses, class rosters, payrolls. The rest of life, the casual, drifting, unscheduled mixing that fills the hours between the labeled rooms, keeps no roster, and so it slips the net. A small error in the place no one can measure swamped all the accuracy in the places they could.

None of this is clean, and the researchers say so plainly. People underreport the strangers they brush past; the test model is a simple one that ignores how often you run into the same faces. Both instruments are rough. But the error runs one way — the model consistently comes in under the survey, not over it. The contacts that define the loose, unmapped middle of your life are exactly the ones the forecasts most need, and least have.