High-Resolution Measurements of Face-to-Face Contact Patterns in a Primary School

A child in a French schoolyard is wearing a small badge clipped to a lanyard. The badge doesn't record words or location. It just senses whether two faces are close enough — within about one and a half meters — to breathe on each other, and for how long. That simple act of sensing, multiplied across two hundred forty-two people over two school days, produces something researchers had never had before: a real map of who contacts whom in a school, at what time, and for how long. What that map reveals rewrites what epidemic models have assumed about children for decades. The assumption those models relied on is called homogeneous mixing — the idea that a child is equally likely to encounter any other child in the school at any given moment. It's a convenient mathematical fiction. It makes the equations tractable. According to Stehlé and colleagues, it is wrong in ways that matter enormously for decisions about school closures, class isolation, and pandemic response. To get the real picture, the team deployed the SocioPatterns proximity-sensing infrastructure in a primary school in Lyon, France. Children and teachers wore active radio-frequency identification badges on their chests. The hardware was tuned so that packets could only be exchanged when two wearers were genuinely facing each other — chest to chest, conversation distance — at roughly one to one and a half meters. That range was chosen deliberately as a proxy for the kind of encounter that could transmit a respiratory infection. The time resolution was precise: every twenty seconds, the system checked whether each pair of badges had exchanged a packet. If yes, the contact was live. If a full twenty-second window passed with nothing, the contact was broken. The probability of detecting a genuine proximity over any given twenty-second interval exceeded ninety-nine percent. Data collection ran across two consecutive school days — Thursday and Friday, October first and second, two thousand nine. Readers covered classrooms, the canteen, stairways, and the playground. Of the two hundred forty-one enrolled children, two hundred thirty-two participated — a participation rate of about ninety-six percent — along with all ten teachers, for a total of two hundred forty-two individuals. Badges were not worn during sports activities. Everything else was captured. What came out of those two days was seventy-seven thousand six hundred two contact events. That number alone starts to tell the story, but the per-child averages make it visceral. Each child had on average three hundred twenty-three contact events per day with forty-seven distinct other children, accumulating roughly one hundred seventy-six minutes of face-to-face time. That's nearly three hours of proximity contact in a single school day. The distribution of those contacts is skewed hard. The mean contact duration is thirty-three seconds, and eighty-eight percent of all contacts last less than one minute. But the tail is long: more than zero point two percent of contacts exceed five minutes. At the pairwise level — looking at how much time any two specific individuals spent together — sixty-four percent of pairs accumulated less than two minutes across a whole day, but nine percent spent more than ten minutes together, and zero point thirty-eight percent spent more than an hour. There's no single typical contact. The variation is enormous, and that matters for transmission: brief encounters carry little risk, but the minority of prolonged ones carry a disproportionate share of total exposure time. There's also a clear rhythm to the school day. When Stehlé and colleagues aggregated contacts in twenty-minute windows, degree — the number of active contact partners — spiked at moments corresponding to morning break, lunchtime, and afternoon break. Between those pulses, contacts were mostly within classrooms. During the pulses, the network briefly opened up across class lines. That word — briefly — is important. The single most striking finding in this study is how confined children's contact worlds really are. Each child spent, on average, three times more time in face-to-face proximity with classmates than with children from other classes. Stehlé and colleagues visualized this through an exposure matrix — a grid whose rows and columns represent classes or grade levels, and whose entries record the average daily contact time between individuals from each pairing. That matrix is strongly block-diagonal: the brightest entries cluster on the diagonal, where same-class pairs sit. Same-grade pairs light up next. Cross-grade pairs are dim. The numbers make it concrete. A first-grader spent an average of three hundred twenty-two minutes per day in contact with other first-graders — but only twenty-four minutes with second-graders and fewer than fourteen minutes with third-graders. A fourth-grader averaged one hundred ninety minutes with fellow fourth-graders and just three minutes with second-graders. The structure is sharp, consistent, and repeatable. To confirm this wasn't just day-to-day noise, the researchers measured the cosine similarity of each child's contact neighborhood between the two days — essentially asking how much overlap there was between the set of people a child touched on Thursday and the set they touched on Friday. The overall average was zero point six seven. Within a child's own class, it rose to zero point seven four. Across different classes, it fell to zero point two. The within-class social world is stable. The between-class social world largely resets each day. On the second day, the average child had twenty-six repeated contacts with people they had met on the first day — nineteen of those within the same class, just seven across classes. They also made about twenty new contacts, but only one point four of those were within the same class; the other eighteen or so were children from different classes, encountered mostly during breaks and lunch. This is the transmission topology in plain terms: a tight, repetitive core of within-class ties and sparse, partially renewed between-class connections concentrated in a few daily chokepoints. Those chokepoints matter for public health. When the researchers traced the trajectories of children through the physical space of the school — reconstructed from which readers detected which badges — the routes through canteens, corridors, and playgrounds produced predictable moments of inter-class encounter. Capacity constraints in the canteen and playground meant that only two or three classes shared any given break, limiting, but not eliminating, cross-class mixing. A disease introduced into one classroom does not instantly flood the school. It tunnels first within the class, and then, at these narrow temporal and spatial windows, may spill outward. This has direct implications for how we model and manage school-based outbreaks. The exposure matrix that Stehlé and colleagues derived can be plugged directly into transmission models to replace the homogeneous-mixing assumption — that convenient fiction — with empirically grounded structure. Instead of a model where every child is equally reachable, you have a model where a virus has to navigate actual social architecture: dense within-class clusters, grade-level blocks, and a few daily windows of wider mixing. That changes the predicted speed of spread. It changes the value of class closure versus full school closure. It alters the calculus of targeted interventions like isolating a single classroom or shifting break schedules to reduce overlap. The researchers are candid about what the study can't tell us. Two days in one school in one country is a limited sample. The badges measure proximity, not physical contact — a child sitting at a desk near a classmate registers the same as one whispering in a friend's ear. Sporting activities weren't captured. The study doesn't include out-of-school contacts, which shape transmission in ways that school data alone can't address. But the method itself is scalable. Wearable proximity sensors with high time resolution and near-complete participation can be deployed in any school. The exposure matrices they generate are concrete, usable tools — not theoretical constructs but empirically measured grids that modelers can work with directly. The deeper point Stehlé and colleagues make is about what happens when you stop assuming and start measuring. The homogeneous mixing model isn't wrong because modelers were lazy — it was wrong because nobody had data good enough to replace it. Now they do. A child wearing a radio badge on a lanyard, walking toward a classmate in a Lyon schoolyard in October two thousand nine, helped build the first real map of how infection actually moves through a school. That map is specific, structured, and far more useful than the assumption it replaces. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.