Tom A.B. Snijders is Professor of Statistics and Methodology at the Dept. of Sociology, University of Groningen, Emeritus Fellow, Nuffield College, University of Oxford and an Associate Member at the Dept. of Statistics, University of Oxford.
Abstract: Homophily is a basic feature of social networks. For numerical actor variables, its specification in statistical network models is usually done by means of the absolute difference between ego and alter on the variable under consideration; sometimes, as an alternative, by the ego-alter interaction. It is argued that such specifications are incomplete for continuous actor variables and for ordinal numerical variables with three or more categories. The reason is that ego is not necessarily attracted mostly to others with the same value as ego; often the attraction is to some value between ego's value and the 'social norm'. (Attraction here is to be understood not necessarily as a preference, but rather as an empirical tendency.) Therefore, the usual representation will often amount to a misspecification. This is elaborated in an extension of the usual specification of effects of actor variables in stochastic actor-oriented models for network dynamics. This new specification may have consequences for results of studies of social selection. An example is given.
Dan McFarland is Professor of Sociology and Organizational Behavior at Stanford University. His research focuses on the social and organizational dynamics of educational systems like schools, classrooms and universities. In particular, Dan has performed a series of studies on classroom organization and interaction; on the formation of adolescent relationships, social structures, and identities; on interdisciplinary collaboration and intellectual innovation; and on relational sociology. His interdisciplinary collaborations with linguists and computer scientists are cutting-edge studies of big data and methodological advances in social networks and language modeling.
Abstract: This paper attempts to directly consider the nature of relationships and the role of interaction dynamics more deeply. To this end, relationships are reconceptualized as a story between persons that is perceived (labeled), agreed upon, and enacted in interaction. From this perspective, types of ties like friendship are relational frameworks that are mutually recognized and enacted via certain interactional footings. To identify the effect of interactional footings over and above previously identified network mechanisms, we rely on systematic social observations of hundreds of settings that extend across one hundred thousand turns of social interaction, as well as longitudinally collected sociometric surveys and institutional records. With these data, interactions are not only coded for a variety of qualities, but they are situated in various social contexts and institutional framing efforts. For example, a particular interactional event, like the act of agreement between i and j at time t, can be embedded in a particular setting, a task (or sequence), a role-relation, and a reported friendship relation. Since most interactions are guided by any one or more of these framing efforts machine learning is employed to identify the interactions associated with each one while taking into account their overlap. Ultimately, the goal is to identify the interactional signal of a perceived and agreed upon reports of friendship. In such a fashion, we identify the interactional footings or ³friendship script² that actors employ to signal the relational frame of ³friendship². This signal - as a latent dimension - is then tested for its predictive capacity on friendship formation to ascertain if it has an effect over and above previously held mechanisms of tie formation.