Source-linked AI summary
Contact patterns in a high school: a comparison between data collected using wearable sensors, contact diaries and friendship surveys
Rossana Mastrandrea, Julie Fournet, Alain Barrat
TL;DR
The paper asks how diaries, surveys, wearable sensors, friendships, and Facebook links differently capture interaction patterns, a question important for understanding social and transmission networks. Using concurrent data from a French high school, it finds that self-reported networks miss many short contacts but still recover the contact network’s class-level structure, while sensors provide broader sampling and temporal coverage.
Problem
Different interaction-collection methods are rarely compared in the same setting, and overlap among contact, friendship, and online networks remains an open question.
Method
The study compares temporally resolved wearable-sensor contacts with diaries, friendship surveys, and Facebook links collected concurrently from more than 300 high-school students.
Results
Diaries miss most short contacts and overestimate durations, but diary and friendship networks recover the contact network’s overall class-mixing structure despite lower sampling and fewer links.
Takeaways & Limitations
Sensor-based collection was more readily accepted and sampled the population better, whereas diaries and surveys captured a backbone of stronger, longer links.
Abstract
from arXiv · showhide
Given their importance in shaping social networks and determining how information or diseases propagate in a population, human interactions are the subject of many data collection efforts. To this aim, different methods are commonly used, from diaries and surveys to wearable sensors. These methods show advantages and limitations but are rarely compared in a given setting. As surveys targeting friendship relations might suffer less from memory biases than contact diaries, it is also interesting to explore how daily contact patterns compare with friendship relations and with online social links. Here we make progresses in these directions by leveraging data from a French high school: face-to-face contacts measured by two concurrent methods, sensors and diaries; self-reported friendship surveys; Facebook links. We compare the data sets and find that most short contacts are not reported in diaries while long contacts have larger reporting probability, with a general tendency to overestimate durations. Measured contacts corresponding to reported friendship can have durations of any length but all long contacts correspond to reported friendships. Online links not associated to reported friendships correspond to short face-to-face contacts, highlighting the different nature of reported friendships and online links. Diaries and surveys suffer from a low sampling rate, showing the higher acceptability of sensor-based platform. Despite the biases, we found that the overall structure of the contact network, i.e., the mixing patterns between classes, is correctly captured by both self-reported contacts and friendships networks. Overall, diaries and surveys tend to yield a correct picture of the structural organization of the contact network, albeit with much less links, and give access to a sort of backbone of the contact network corresponding to the strongest links in terms of cumulative durations.
I. INTRODUCTION
The study addresses how different methods capture face-to-face interactions and how contact, friendship, and online networks overlap in a high school. It compares concurrent sensor and diary measurements with friendship surveys and Facebook links.
- Face-to-face interactions shape social networks and channels of information, opinions, and infectious-disease transmission.
- Surveys and diaries provide contact context and duration estimates but are costly and vulnerable to recall and duration-estimation biases.
- Wearable sensors objectively detect close-range proximity, including short encounters, and support temporally resolved longitudinal contact data.
- Previous high-school evidence found that many sensor-recorded contacts, especially short ones, were omitted from surveys, whereas long contacts were better reported.
- The overlap among behavioral contacts, reported friendships, and online social links remains largely open, motivating their comparison.
- This study compares these interaction data sets collected concurrently from more than 300 high-school students during one week.
A. Study design, data collection and description
The study combined sensor, diary, friendship-survey, and Facebook-network data from students in nine French preparatory classes. Participation and sampling differed substantially across collection methods.
- The study covered nine second-year preparatory classes spanning MP, PC, PSI, and BIO specializations.
- Sensors detected close proximity through unobtrusive wearable badges tuned to exchange radio packets within 1–1.5 meters.
- 327 students participated out of 379 in the nine classes, corresponding to an 86.3% participation rate.
- Contact diaries collected one-day lists of close face-to-face contacts and approximate aggregate durations in four categories; 120 students returned diaries.
- Friendship surveys were obtained from 135 students, while 17 students provided local Facebook friendship networks.
- The present collection followed two earlier sensor-only data collections in the same high school.
B. Ethics and privacy
The study obtained informed consent from adult participants and limited collected personal information to specified metadata.
- All participants were at least 18, gave signed informed consent, and received study information before deployment.
- Collected personal information was limited to the metadata specified by the study protocol.
C. Data analysis
The analysis represents sensor contacts as temporally resolved weighted networks and diary, friendship, and Facebook data as differently structured relational data. Class-level contact matrices summarize link density and cumulative contact duration.
- C. Data analysis: Sensor data form a temporal network with 20-second resolution, where link weights equal total contact time over an aggregation window.
- C. Data analysis: Grouping students by class produces contact matrices describing mixing patterns between classes.
- C. Data analysis: Link-density matrix entries normalize observed links between class pairs by the maximum possible number of links.
- C. Data analysis: Duration matrices report total contact time between class groups and can be normalized to average time per student.
- C. Data analysis: Diary data form a weighted directed network without temporal resolution, and reported links need not be reciprocated or have matching durations.
- C. Data analysis: Friendship surveys form directed unweighted nomination networks that can be symmetrized for comparison with aggregated contacts.
- C. Data analysis: Facebook data consist of known pairs and unknown pairs rather than a complete network because few students supplied local networks.
A. Contact patterns from sensor data
Sensor measurements reveal highly variable, bursty face-to-face contact patterns with strong within-class mixing and robust daily organization.
- 67,613 contact events among 327 students totaled 3,770,160 seconds, approximately 1,047 hours.
- 75% of contacts lasted less than 1 minute, while 2% lasted more than 5 minutes; the duration coefficient of variation was 2.7.
- The aggregated network contained 327 nodes and 5,818 weighted edges, with average degree approximately 35 and clustering coefficient approximately 0.5.
- 62,342 contacts, representing 92.2% of all contacts and 93% of contact time, occurred within classes.
- Daily contact matrices were highly similar, with cosine similarities ranging from 92.7% to 98%, indicating robust daily fluctuations and class mixing patterns.
B. Contact diaries
Contact diaries captured a directed, weighted subset of reported interactions, with incomplete participation and substantial disagreement between students’ reports.
- Only 120 students completed diaries, producing 502 directed weighted edges; restricting to students with sensor data yielded 109 students and 416 links.
- 158 contacts were reported by only one student, whereas reciprocated reports represented 129 student pairs.
- The overall probability of reporting a contact was approximately 62%, falling to 40% when the highest reported duration was assumed correct.
- Among 129 reciprocated pairs, both students reported the same duration category in 81 cases and differed by one category in 35 cases.
C. Comparing contact diaries and sensor data
Compared with sensor data, diaries recorded fewer links but preserved the contact network’s class-mixing structure, especially for longer interactions.
- Sensor-based contact density and average degree were almost twice those of the diary network, although the degree distributions had similar shapes.
- The sensor and diary contact-matrix link densities had 97% similarity despite low diary sampling.
- 70.4% of diary links matched sensor contacts, whereas only 41.4% of sensor contacts matched diary links.
- Links absent from diaries tended to have smaller sensor-measured cumulative durations, while all links above a threshold near 1 hour were reported.
- Diary-reported duration categories covered both shorter and longer sensor-measured durations, with broader distributions for larger reported categories.
- 32% of 1–5 minute contacts were reported, compared with 71% and 69% in the next categories and 100% for contacts exceeding 1 hour.
SENSORS
Diary reporting shows duration-dependent recall and uneven individual-level correspondence with sensor-measured contact degrees.
- 65.8% of matched links were overestimated in diaries, 29.7% had the same duration category, and 4.5% were underestimated.
- The correlation between sensor degree and diary in-degree was 0.4, whereas diary out-degree showed no significant correlation with sensor degree.
- For students with CV ≤1, diary out-degree correlated with sensor degree at close to 0.35; no such correlation appeared for CV >1.
- The Kendall’s τ rank-correlation between sensor-ranked and diary-ranked link durations was approximately 45%.
D. Multiplex network of students’ relationships
The high school data are treated as a multiplex of face-to-face contacts, reported friendships, and Facebook links, whose overlap reveals distinct relationship layers. Reported friendships align with longer contacts and class mixing, whereas Facebook links without reported friendship tend to represent shorter contacts.
- Network layers: The multiplex combines contact, reported-friendship, and Facebook-link layers, but Facebook coverage is incomplete because many pairwise links are unknown.The friendship network is symmetrized for comparison, while Facebook does not provide a complete observed network.
- Network structure: Despite different network densities, reported friendships closely mirror the class mixing structure of the contact network.The contact and friendship link-density matrices have high structural similarity, including class blocks and dominant within-class patterns.
- Friendship and contact overlap: 86% of reported friendships correspond to contact links, whereas only 24% of contact links correspond to reported friendships.Restricting contacts to aggregate durations above 1 minute or 3 minutes raises the contact-to-friendship correspondence to 39% or 50%, respectively.
- Friendship and contact overlap: All contact links with aggregate duration above roughly 2.5 hours correspond to declared friendships, although reported-friendship contacts can also be short.Declared friends have larger-average and broader duration distributions, especially when friendship is reciprocated.
- Metadata and relationship layers: Facebook links are more similar to contact links than reported friendships in their relationship with shared student features.A substantial fraction of pairs sharing none or only one feature still have a Facebook link.
- Facebook and contact overlap: Facebook-only links have narrower, shorter contact-duration distributions than reported friendships, while links shared by both layers have broader durations.Contacts exceeding a threshold occur only between reported friends, so Facebook friendship is not equivalent to reported friendship in face-to-face behavior.
IV. DISCUSSION
The study compares concurrently collected sensor, diary, friendship-survey, and Facebook data from a high school, finding systematic sampling and reporting differences while recovering the contact network’s broad class structure. Diaries and surveys capture a sparse backbone of stronger links, whereas sensors provide more complete temporal contact data.
- Study design: Data from 327 students were collected concurrently for one week using wearable sensors, contact diaries, friendship surveys, and Facebook links.The analysis compares overlap and complementarity among these data sets in the same population and time period.
- Sensor–diary comparison: Many sensor-detected short contacts were absent from diaries, whereas contacts with long enough aggregate durations had high reporting probability.All contacts above a sufficiently long sensor-measured duration were reported in the diaries.
- Reporting biases: Students tended to overestimate contact durations relative to sensor measurements, consistent with documented response and social-desirability biases in self-reported surveys.The paper links duration overestimation to response-style and response-set biases, while noting that degree may also matter.
- Network structure: Despite lower sampling and fewer links, diary networks recovered the sensor network’s overall class-mixing structure; omitted short contacts made paths appear longer.The diary network was less dense because many short contacts were not reported.
- Friendship and online links: Facebook links not reciprocated as reported friendships were associated with shorter, narrower aggregate contact durations, giving Facebook ties a more casual character.The paper also reports incomplete Facebook-network sampling and a lower probability of observing contact conditional on a Facebook link than conditional on reported friendship.
- Implications: Wearable sensors were more readily accepted than diaries or surveys in this high-workload school context, yielding better population sampling.The authors suggest combining methods to compensate for sensor sampling bias and capture contacts with the external population.
B. Comparison between the networks built from sensor data and from contact diaries
Sensor and diary networks differ substantially in sampling and connectivity, but diary data preserve important aspects of the sensor-measured network. Node-level comparisons show that sensor degree aligns with incoming rather than outgoing diary reports.
- Network-level comparison: Sensor and diary data produce degree distributions and shortest-path structures that can be compared directly, with diary networks including isolated nodes represented as “No path.”The supplied figures compare degree distributions and shortest-path lengths between the two networks.
- Node-level comparison: Sensor degree correlates with diary-network degree overall, but not with diary out-degree; it correlates with in-degree instead.Out-degree counts contacts reported by the individual, whereas in-degree counts contacts reported by others about that individual.