Contact Network in Primary School

Networks, Visualization


Primary school is an important stage in child development. The brain develops rapidly in the early years of childhood, and primary school plays a crucial role of nurturing the mind. Primary school is a place where young children not only learn to read and write, but also interact and get along with one another. Social skills teach children how to behave appropriately and adaptively in our cultural environments. Although social skills are not graded by teachers, they are constantly graded by peers. If a child scores well on a “social test,” he is likely to play well with others. Otherwise, he is apt to feel disconnected and left out.


PLOS One collected quantitative information in effort to trace mixing patterns of children in school environments. The study observed face-to-face contact patterns in a French primary school, among children in the range of 6-12 years old. Researchers measured the time-resolved face-to-face proximity of children and teachers using a “proximity-sensing infrastructure based on radio frequency identification devices.”

Data on face-to-face interactions was collected on two adjoining days, October 1st and 2nd of 2009. A total of 77,602 events of contact were collected amond 232 children and 10 teachers. Results indicate that each child made an average of 323 contacts each day with about 47 children. The average time of daily interaction came out to 176 minutes, with mostly brief contact within a matter of seconds.

The original purpose of conducting this study was to understand how contact between children at school effects the transmission of respiratory infections. According to PLOS One, the study was meant to reveal “important properties of the contact patterns between school children that are relevant for modeling the propagation of diseases and for evaluating control measures.” With the data collected, control and preventative measures could be implemented to reduce the disruption to education during education.


Although the data was primarily collected to address public health concerns, it also provides important insight on social interactions among primary school students. The individuals, or nodes, in the study contain two attributes: a class grade and gender. Measuring face-to-face proximity based on gender can provide some interesting insight on the gender dynamics among young students. Analysing the weight of the edges can reveal how long face-to-face proximity lasts relative to the student’s sex.

Originally, PLOS One used the data to reveal the the isolated network structure that occurs among children based on their class and grade. They created the following network data to show how contacts occur mostly within each class, and that each child spends on average three times more time in contact with fellow classmates than with children from other classes. I would like to interpret the network based on gender dynamics rather than class.  


SocioPatterns Dataset

SocioPatterns provides two GEXF files of the two-day study by PLOS One that can be loaded directly into Gephi. The file contains a nodes table, consisting of student ids, class name, and gender. The edges table outlines the source id and target id, the duration of the contact in seconds, and the number of contacts per day.


Gephi is a software used for network visualization and manipulation. Gephi provides algorithms that layout nodes inside a graphic space.


When I initially imported my graph into Gephi, the nodes were positioned randomly on the graph. I scaled the visualization in order to see the nodes more clearly and adjusted the edge thickness so that the longer duration of contact was perceivable by line weight. I ranked the node’s size based on degree, making sure the largest node was not too large. Then I reformatted the labels so that the class name would be displayed.

I experimented with a lot of the layouts before I found the appropriate one. First I tried “Force Atlas” which uses force-based algorithms to attract nodes that are linked to each other. After researching the layouts, I realized that ranking layouts, such as Circular and radial Axis, and geographic repartition layouts like GeoLayout, were not appropriate for my data since there is no central actor or geographic connection. Complementary layouts, like Force Atlas, Yifan Hu, and Frushterman-Reingold were most useful for depicting the face-to-face connection between different individuals.  

I created two visualizations, one where I colored the nodes based on class name (fig 1) and the other based on gender (fig 2). Similar to PLOS One’s findings, it is apparent that students have more face-to-face interactions with classmates rather than those outside of their class. Regardless of class or age, there also appears to be clustering by gender. Those with the highest degree of face-to-face contact are mostly male.

fig 1 : interaction by class name
fig 2 : interaction by gender


Using Gephi was a challenge, because I had a hard time finding out what each layout was doing with the data. I also had difficulty achieving the right effect on preview mode. For instance, when I changed the labeling from id to class name, the preview mode still showed id instead of class name (fig 3). This is probably an issue with the interface design, as it feels a bit antiquated and difficult to navigate. A lot of the features and settings are incomprehensible, and your actions cannot be undone.

fig 3

If I had more time with this project, I would try to visualize gender interactions across a more diverse class range, like elementary school to high school. I wonder how the network visualization would shift with a larger dataset, and if a pattern in gender interaction emerges relative to age. Do students mingle more with the opposite sex as they grow, or do they start to form a distance?

Also, as a graduate of Barnard, a women’s college, I came across the common opinion that female students are more engaged and comfortable in all-female classrooms than mixed-gender classrooms. I think this would be very interesting to visualize on Gephi if I had the data. I wonder how the interaction patterns of male and female professors and students are altered by the gender composition of the class.