With their increasingly sophisticated sensing capacity, mobile phones have been repeatedly used as pervasive, low-cost sensing devices in everyday life [5, 6]. In our line of work we attempt to employ mobile phones as a platform for sensing primary school students’ social participation in the school community.Over the past decade, schools have increasingly shifted towards inclusive education . Rather than dividing students across regular and special education schools, inclusive education argues for ‘one school for all’. Next to its educational benefits, one of the main arguments for inclusive education is the increased social participation of children with learning and communication disabilities in regular, as opposed to special education schools . However, despite the attention given to inclusive education in recent years, researchers have criticized a lack of empirical evidence on how exclusion is manifested in actual students’ behaviours . Even when studies have tried to capture how exclusion is manifested in students’ social interactions within the class and during play time, their focus was limited to a number of school cases as well as particular dimensions of diversity, thus leading to an uncertainty of how such results may generalize to the larger population and how educational exclusion is manifested at large [7, 9, 15]. Existing methodological tools rely largely on selfreporting, either from teachers or students, and are thus susceptible to a number of biases. For instance, teachers have been found to overestimate the social participation of disadvantaged children . In our line of work we attempt to develop technology that senses children’s social interactions in the playground [see [2, 11, 12] for earlier studies]. More specifically, we have developed a Google Android application that infers children pair-wise proximity during free play in open spaces (such as the playground) through frequently sampling the Bluetooth RSSI of nearby devices. Data are then modelled as a social network, thus enabling us to understand both how given students participate in the school community but also how inclusive the community is overall, and respective changes over time. In the remainder of this paper we describe how Proximy senses students’ social interaction, and present our early insights from the deployment of Proximy with 137 students over a period of 3 weeks.
Sensing social interactions with Proximy
Physical proximity has been repeatedly found to be a relevant indicator of social interaction in a number of settings [6, 13]. To sense students’ pair-wise proximity we employ a cost-effective solution using smartphones, strapped around students’ waists (see Figure 2). Proximy samples for nearby devices and their respective RSSI values every 10 seconds.
An empirical evaluation sought to test Proximy’s effectiveness in reliably discriminating between different ranges of proximity. Ten subjects wearing smartphones with Proximy were asked to stand in relative proximity of 1-10 meters, and with relative angle of 0 (facing each other), 90, 180 and 270 degrees. Figure 1 highlights the mean RSSI values for each condition (along with 95% Confidence Intervals). One may quickly note that the relative orientation does play a significant role. However, we find that we may reliably infer a pair-wise proximity of 4 meters or less when the RSSI value is -71 or higher, irrespectively of the relative angle of the two subjects, with the of 180° (which most likely reflects the absence of social interaction).
Deploying Proximy in the wild
We have deployed Proximy over a three-week-long study with a total of 137 students of age seven to ten years old (six classrooms in total), sensing students’ social interactions during free play in outdoor space (see Figure 2). Students were given their personal pouch with the smartphone as they exited the classroom and headed towards the open space. Over the course of the three weeks, and during two breaks of length of 45 minutes each, proximy was sampling for nearby devices and logging their signal strength (RSSI) every 12 sec. Data were cleaned to remove initial and final samples, were all students would get together to pick up and deliver their pouches. RSSI values of -71 or higher would indicate physical proximity. Data were then modelled as a weighted network (see Figure 3 for an example), with the total number of samples between any two students being reflected in the weight of the link between the two nodes in the network.
In order to analyse the validity of proximy as a tool measuring social interactions in the school community we employed established questionnaires measuring social participation and wellbeing in school communities, to act as ground truth data. These included: the Peer Nomination Inventory  which measures the acceptance and rejection of primary school children by her peers, a 24-item scale measuring loneliness experienced in school , the Interpersonal Competence Scale  with which teachers may assess the interpersonal competence of primary school children along 18 dimensions, and a checklist of diversity factors, such as the existence of learning or communication disabilities as well as their ethnic and socio-economic background. Our preliminary insights suggest that centrality metrics such as betweenness centrality, i.e., the number of times a student acts as a bridge along the shortest path between two of her peers, are moderately correlated to peer acceptance as measured through the Peer Nomination Inventory and only weekly related to experienced loneliness in school.
Conclusion and future work
In this paper we presented Proximy, a mobile app that samples Bluetooth’s signal strength (RSSI) as a way to infer children’s social interactions during free play. We found that proximity may reliably infer whether two children are closer than 4 meters or not, irrespectively of their relative orientation. Our preliminary insights suggest some validity in proximy, as reflected in moderate correlations with established psychometric measurement tools. Our current work seeks to analyse the data in more depth, through exploring different ways of modelling the data as a social network.
1. Asher, S., Hymel, S. and Renshaw, P. (1984). Loneliness in children. Child development. 1456– 1464.
2. Belim, V., Lyra, O., Teixeira, P., Caraban, A., Ferreira, M.J., Gouveia, R., Lucero, A. and Karapanos, E. (2014). Beyond gamification: sociometric technologies that encourage reflection before behavior change. Proceedings of the 11th Conference on Advances in Computer Entertainment Technology – ACE ’14 1–6.
3. Booth, T., Ainscow, M., Black-Hawkins, K., Vaughan, M. and Shaw, L. (2000). Index for inclusion: Developing learning and participation in schools. CSIE, 1 Redland Close, Elm Lane, Redland, Bristol BS6 6UE, United Kingdom (24.50 British pounds, including UK postage and handling).
4. Cairns, R., Leung, M., Gest, S. and Cairns, B. (1995). A brief method for assessing social development: Structure, reliability, stability, and developmental validity of the Interpersonal Competence Scale. Behaviour Research and Therapy. 33, 6, 725–736.
5. Doerflinger, J., Gross, T., Lyra, O., Karapanos, E., Kostakos, V., Cuadrado-Cordero, I., Soria-Morillo, L.M., Gonzalez-Abril, L., Ortega-Ramirez, J.A., Pau de la Cruz, I., Valero Duboy, M.A. and Kanjo, E. (2012). ICTD Work, Plus mFeel. IEEE Pervasive Computing. 11, 3, 43–45.
6. Eagle, N. and (Sandy) Pentland, A. (2006). Reality mining: sensing complex social systems. Personal and Ubiquitous Computing. 10, 4, 255–268.
7. Fryxell, D. and Kennedy, C. (1995). Placement along the continuum of services and its impact on students’ social relationships. Research and Practice for Persons with Severe Disabilities. 20, 4, 259–269. 8. Koster, M., Nakken, H. and Pijl, S. (2008). Assessing social participation of pupils with special needs in inclusive education: The construction of a teacher questionnaire. Educational Research and Evaluation. 14, 5, 395–409.
9. Lareau, A. and Horvat, E. (1999). Moments of social inclusion and exclusion race, class, and cultural capital in family-school relationships. Sociology of education. 37–53.
10. Liasidou, A. (2012). Inclusive Education, Politics and Policymaking. A&C Black.
11. Lyra, O., Karapanos, E., Gouveia, R., Barreto, M., Nisi, V., Nunes, N.J., Zimmerman, J. and Forlizzi, J. (2013). Towards persuasive sociometric technologies for inclusive educational settings. Proceedings of the Biannual Conference of the Italian Chapter of SIGCHI on – CHItaly ’13 1–9.
12. Lyra, O., Karapanos, E., Gouveia, R., Nisi, V. and Nunes, N.J. (2013). Engaging children in longitudinal behavioral studies through playful technologies. Proceedings of the 12th International Conference on Interaction Design and Children 396–399.
13. Paradiso, J.A., Gips, J., Laibowitz, M., Sadi, S., Merrill, D., Aylward, R., Maes, P. and Pentland, A. (2009). Identifying and facilitating social interaction with a wearable wireless sensor network. Personal and Ubiquitous Computing. 14, 2, 137–152.
14. Simpson, R. (2004). Inclusion of students with behavior disorders in general education settings: Research and measurement issues. Behavioral Disorders. 30, 1, 19. 15. Vaughn, S., Elbaum, B. and Schumm, J. (1996). The effects of inclusion on the social functioning of students with learning disabilities. Journal of Learning Disabilities. 29, 6, 599–608.
16. Wiggins, J. and Winder, C. (1961). The Peer Nomination Inventory: An empirically derived sociometric measure of adjustment in preadolescent boys. Psychological Reports. 9, 3, 643–677.