Studying face-to-face interactions in groups: Models and applications

Face-to-face interactions are essential parts of our social lives. They mark the times and places when we come together, exchange knowledge, forge new norms, establish rituals, or influence each other’s beliefs. While face-to-face interactions are repeatedly mentioned in social science theories —- in sociology, psychology, management, educational science -- their compositions and dynamics are rarely analyzed empirically. Although this research gap can be traced to various reasons, an important one derives from the fact that most face-to-face interactions occur in groups. Because of this group nature, current statistical tools devised for social networks (in other words, dyadic relations) are ill suited for observations of face-to-face interactions. This thesis thus advocates for the extension of social network analysis tools to the study of group interactions in face-to-face contexts. In a first study, this thesis proposes a new statistical model (DyNAM-i) for observations of events in which individual join and leave non-over-lapping social groups, thus allowing to study the dynamics of face-to-face interaction groups. Extending previous actor-oriented frameworks, this model can be used to test hypotheses related to how individuals join and remain in interaction groups. Importantly, the model translates the dyadic logic of previous social network models to a group logic in order to examine the mechanisms driving interactions in groups. Two study cases on face-to-face interactions in a small party and an office setting are presented to demonstrate the model's applicability. The second study presents an empirical application of the DyNAM-i model focusing on the role of depressive symptoms on interaction behaviors. More particularly, the study aims at understanding the mechanisms leading to dyadic isolation -- i.e., the tendency to spend more time in dyadic interactions rather than group interactions -- of individuals experiencing depressive symptoms. The previously proposed model is used to test two classes of mechanisms, one pertaining to preferences of individuals with depressive symptoms and the other one to preferences of other individuals. The presence of such mechanisms is assessed on two datasets of face-to-face interactions among university students, collected using RFID sensors. The third study proposes another new modeling framework (ERPM), this time to represent observations of individuals' partitions. This type of model can be used, for example, to explain observed snapshots of interaction groups (i.e., the composition of groups interacting at a certain point in time). Inspired by modeling techniques specific to network and partition data, the proposed model defines an exponential family of distributions for partitions. These distributions can then be specified to test factors responsible for the composition of the groups forming an observed partition. Once more, a case study is used to explain how such models can be specified, estimated, and interpreted. In the same way as for the second study, the fourth study shows an empirical application of the ERPM developed in the previous study. This time, the study focuses on the mechanisms driving adolescents to interact and form friendships with others of similar socioeconomic backgrounds. Observations of meal sharing during a summer camp are analyzed using the previously proposed model, and the evolution of friendship ties is modeled with a longitudinal network model. All the methodological advancements proposed in this thesis contribute to the growing field of statistical analysis of social relations by offering new models tailored for social interactions in groups. The empirical studies presented here provide examples of the theoretical insights that such models can provide. Besides, they contribute to a better understanding of two topics highly relevant to the psychological and sociological fields: the role of depressive symptoms and social class inequalities in social relations. To conclude, this thesis aims to inspire future scholarly efforts in the analysis of social interactions and social groups