Source-linked AI summary
Contact patterns among high school students
Julie Fournet, Alain Barrat
TL;DR
The paper addresses the need for reliable, detailed descriptions of human contacts, particularly in strongly mixing settings relevant to social science and epidemiology. It analyzes two high-resolution wearable-sensor data sets from a French high school using individual and class-level temporal-network measures. Contact patterns show strong class structure and high stability across days and years, while individual contacts renew substantially but less randomly than in null models.
Problem
Reliable, detailed contact data are needed to understand human interaction and inform infectious-disease models, but longitudinal measurements in specific populations remain limited.
Method
The study analyzes two high-resolution wearable-sensor contact data sets from French high-school classes using temporal networks, contact matrices, and neighborhood comparisons.
Results
Contact matrices show strong class structure, while contact-pattern properties remain highly stable across days and years; individual neighborhoods vary substantially but less than under random renewal.
Takeaways & Limitations
Class-level description is an adequate modeling scale, and the data provide fine-grained information useful for realistic contact and spreading-process models.
Takeaways & Limitations
Wearable-sensor deployments are small-scale and context-specific, omit contacts with non-participating individuals, and do not identify physical contact types.
Abstract
from arXiv · showhide
Face-to-face contacts between individuals contribute to shape social networks and play an important role in determining how infectious diseases can spread within a population. It is thus important to obtain accurate and reliable descriptions of human contact patterns occurring in various day-to-day life contexts. Recent technological advances and the development of wearable sensors able to sense proximity patterns have made it possible to gather data giving access to time-varying contact networks of individuals in specific environments. Here we present and analyze two such data sets describing with high temporal resolution the contact patterns of students in a high school. We define contact matrices describing the contact patterns between students of different classes and show the importance of the class structure. We take advantage of the fact that the two data sets were collected in the same setting during several days in two successive years to perform a longitudinal analysis on two very different timescales. We show the high stability of the contact patterns across days and across years: the statistical distributions of numbers and durations of contacts are the same in different periods, and we observe a very high similarity of the contact matrices measured in different days or different years. The rate of change of the contacts of each individual from one day to the next is also similar in different years. We discuss the interest of the present analysis and data sets for various fields, including in social sciences in order to better understand and model human behavior and interactions in different contexts, and in epidemiology in order to inform models describing the spread of infectious diseases and design targeted containment strategies.
Introduction
Accurate, high-resolution contact data are important for studying social interaction and infectious-disease spread, especially in strongly mixing settings such as schools. Wearable sensors address limitations of self-reports, while longitudinal contact patterns remain under-studied; this paper presents two high-school data sets to investigate them.
- Detailed contact patterns matter in social sciences and infectious-disease epidemiology because human interactions shape social networks and disease spread.
- Schools are important settings for contact research because strong mixing among students can facilitate transmission within schools and into households.
- Wearable sensors enable objective, high-resolution measurement of close-proximity and face-to-face contacts while avoiding self-reporting biases.
- Longitudinal contact patterns have barely been studied, motivating two high-resolution data sets collected among high-school students in France in 2011 and 2012.
Methods
The study combines wearable-sensor measurements from participating high-school classes with temporal and aggregated network analyses. It characterizes individual contacts, class mixing, contact durations, and stability across time windows using network and matrix-based measures.
- Study design, high school context, and data collection: Data were collected from 118 students in three classes over four days in 2011 and 180 students in five classes over seven days in 2012.
- Study design, high school context, and data collection: Participation was close to 100% because students were incentivized through projects using the data, although only selected second-year classes participated.
- Study design, high school context, and data collection: The study analyzes whole classes rather than student subsets and assumes limited contact between participating second-year students and first-year students.
- Study design, high school context, and data collection: Wearable badges detect close proximity through radio packets exchanged within approximately 1–1.5 meters, with participants asked to wear them at all times in school.
- Data analysis: The infrastructure records a temporal contact network at 20-second resolution, with individuals represented by sensor identifiers and classes by class labels.
- Data analysis: Aggregated networks quantify adjacency, degree, contact-event counts, cumulative contact duration, and individual strength across selected time windows.
- Data analysis: The analysis compares degree and weight distributions using averages and squared coefficients of variation, and examines how node properties and neighborhoods change across aggregation periods.
- Data analysis: Class-level mixing is summarized with matrices based on total contact time, edge counts, and edge density, while matrix similarity evaluates temporal stability.
Results
The high-school contact network is strongly organized by class, with broad and heterogeneous contact durations. Despite daily fluctuations and changing individual neighborhoods, aggregate contact statistics and class-level structures are robust across days and null-model comparisons.
- Contact dynamics: Contact durations are heavy-tailed: 88% last less than 1 minute, but more than 1% last at least 5 minutes.The average duration is 44 seconds, with squared coefficient of variation CV 2 = 5.1; inter-contact durations are similarly broad and bursty.
- Class structure: 91.5% of all contacts involve students from the same class, while more contacts occur between classes sharing similar study topics.The five classes also separate broadly into two groups with stronger within-group than between-group mixing.
- Network structure: The aggregated network has 180 nodes and 2220 edges, with average shortest path length 2.15, clustering coefficient 0.48, and strong modular structure.A random network with the same numbers of nodes and edges has clustering coefficient ≈0.17.
- Gender structure: 57.7% of edges join two male students, 7.9% join two female students, and 34.4% connect students of different genders.The corresponding same-gender preference patterns differ between male and female students, but gender subdivision is not needed to resolve the contact structure at the analyzed level.
- Longitudinal analysis: Contact counts fluctuate strongly during each day, peaking around lecture breaks, yet the structure between classes is robust across days.The authors conclude that a single-day collection captures the class-level contact structure well, despite within-day variation.
- Longitudinal analysis: Daily contact networks have similar degree and duration distributions, while students’ total contact time changes less than their individual contact neighborhoods.Students meet more distinct individuals within their class than in other classes, and both neighborhood counts continue growing throughout the study.
- Longitudinal analysis: Empirical cosine similarities exceed those from all tested null models, including one preserving class-level topology and cumulative contact-duration statistics.Day-to-day neighborhood changes are substantial but smaller than expected under random-contact alternatives.
Discussion
The study finds that high-school contact patterns are strongly structured by classes and remain highly robust across days and years, while individual contacts renew substantially. These findings support class-level and temporally robust modeling, while wearable-sensor data remain complementary to surveys because of differences in scale and contact information.
- Class structure: Contact matrices show strong within-class structure and an additional grouping of two and three classes.The diagonal structure reflects school organization, while off-diagonal patterns are linked more to spatial arrangement and schedules.
- Class structure: Gender-based subdivision does not appear necessary for modeling spreading processes when students are already divided by class.The authors identify class-level description as adequate for models evaluating spread in this population.
- Temporal robustness: Daily contact patterns are robust in temporal activity, statistical distributions, and neighborhood renewal, although individual contacts are not identical across days.Neighborhood changes are substantial but lower than under random-contact null models, and renewal rates are similar across the two years.
- Temporal robustness: Contact-pattern similarity remains very strong across years despite different student populations.The comparison uses data from the same setting collected in successive years and supports cross-year stability of the observed patterns.
- Implications and scope: Wearable sensors provide fine-scale, high-temporal-resolution contact information that complements surveys but typically covers smaller, context-specific populations.Sensors reveal mixing within and between classes and contact-duration heterogeneity, whereas surveys provide large-scale age-based matrices and information about physical contact types.