Statistical analysis of time-stamped multi-actor events in social networks

• Applicant: Dr. Jürgen Lerner
• Subject Area: Empirical Social Research
• Term: since 2016
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 321869138

Relational event models (REM) can model the propensity of actors to engage in dyadic time-stamped interaction events and are becoming an established framework for networks of communication, collaboration, or social interaction in general. REM are appropriate for longitudinal social network data in which interaction is given by relational events with fine-grained time information, rather than stable relational ties. Network data of this kind is becoming more and more frequent due to the rise of computer-mediated communication and collaboration, automated data collection and coding strategies, and the availability of public high-quality data resources. A major shortcoming of most current models, however, is that they are designed for dyadic interaction - relating two actors, or nodes, in one event - while social interaction events can often comprise any number of actors. For instance, meetings can be attended by any number of persons, scientific papers can be written by any number of authors, emails can be send to any number of recipients, or project-related tasks can be tackled by teams of any size. Treating multi-actor interaction as collections of dyadic events is invalid in general. The lack of a general statistical model for networks of time-stamped multi-actor events is a severe limitation for social network analysis, given that network data of that type naturally arises in a large variety of application settings. Building on work achieved in the first funding period, this methodologically oriented project will generalize REM to relational hyperevent models (RHEM) for time-stamped multi-actor events in social networks. RHEM will be developed for directed and undirected interaction, for multi-variate and multi-mode networks, for events with an associated outcome, and for events given in various time granularity. Newly developed models will be implemented and disseminated in an extension of the open-source software "eventnet" - developed within the first funding period - and introduced to practitioners in hands-on workshops at conferences and summerschools. Method development will be guided by and applied in a variety of application settings - building on and extending collaborations with social scientists from different areas that have been started during the first funding period and in which the need to extend REM to multi-actor interaction has become pressing. We argue that relational hyperevent models will make a big impact in many application areas of social network analysis for the simple reason that social actors often communicate, collaborate, meet, decide, agree, or quarrel in groups beyond size two.

• DFG Programme: Research Grants