Giuseppe Arena

University of Tilburg
Department of Methodology and Statistics


The Time is Now: Understanding Social Network Dynamics Using Relational Event Histories

How do social relations in classrooms change across different settings such as lectures and group work, and how do these relations affect the students’ motivation to keep working on a task and avoid rebellious acts? How do colleagues in working projects communicate with each other, and how is this affected by physical distance and means of communication? What triggers violent interactions between gangs and what can be done to slow down violent interactions or make them stop entirely? How do emergency responders (e.g., police, fire fighters, medical personnel) share information and coordinate among each other when emergencies occur?

These are just a few pending questions researchers have been facing in the last few decades (Monge, 1991; Mitchell & James, 2001; Cronin et al., 2011; Kozlowski et al., 2013, in press; Leenders et al., 2016). The difficulty in answering such questions lies in the fact that social interaction – whether it is the interaction among teachers and students, military in the field, doctors in a surgery room, or company employees working together on breakthrough innovations – is inherently dynamic in nature. Students in classes go through various stages of trust and development, move from one performance episode to the next, and constantly adjust their internal and external interactions. Soldiers in the field need to respond to hostile attacks and constantly adapt their mode of operation to their environment and their own well-being. It is safe to say that most, if not all, interaction among individuals, teams, organizations, and even countries, is dynamic in nature. Although most social scientists acknowledge the dynamic nature of human interaction, there is virtually no social science theory that describes exactly how long it takes to develop trust, how much faster integration occurs among classrooms made
up of Western cultures as opposed to classrooms combining Western and non-Western cultures, and exactly how long a particular interaction (e.g., a student insulting the teacher) will affect future relations.

In order to truly understand network dynamics and predict future events between individuals,
groups, or countries, we need (1) empirical data that captures networks dynamics accurately and with high resolution, and (2) we need statistical models that can extract the information contained in these data to answer practical and prevalent research questions. The first requirement is fulfilled due to the increasing availability of so-called relational event history data. This relatively new source of data, which is rarely used in social science research, consists of sequences of interactions, called relational events, between a sender and a receiver at a specific point in time. In a working team, a relational event could be one team member providing another member with information. In the classroom a relational event could be the teacher hushing a student. In the case of criminal organizations, a relational event
could be a gang member killing the member of another gang.

Due to new technical developments, sequences of these relational events can be collected relatively easily. Interaction occurs through communication technology (e.g., email) leaving digital traces about senders, receivers, and timing. Employees in companies wear sociometric badges that store the interactions between colleagues. Classrooms are being recorded on video to observe interactions between teacher and students. Police stores databases of criminal interactions between gangs. Because these data contain information about relational events in continuous time, these data can tell us how fast/slow teams operate, why and when it speeds up or slows down, how the past affects the future, and how (quickly) social order evolves.

The second requirement – the availability of statistical models to extract the treasure of information contained in relational event histories – has not yet been fulfilled. A key characteristic of relational event data is the time and order that events took place in the past. Obviously it makes a great difference whether an insult of a student towards the teacher is followed by the teacher hushing the student, or when the order of these two events is switched when predicting what will happen next. Currently available statistical models of social network interaction (such as the Exponential Random Graph Model (ERGM; Lusher et al., 2012) and the Stochastic Actor-Oriented Model (SAOM; Snijders et al., 2010) are unable to capture the time and order of events in an appropriate way. On the other hand, the recently proposed Relational Event Model (REM; Butts, 2008) yields a promising new approach for modeling the timing and ordering of events. At this stage, however, the REM is still in a very preliminary stage of development and therefore it can only be used for a limited set of research questions.

Due to the recent availability of relational event histories and the great potential of the REM,
the time is now to develop a statistical framework for building and testing dynamic theories to better understand temporal dynamic social processes. One of the several goals of this projects will be to develop a general and flexible statistical framework for modeling relational event data. The new framework will be referred to as the Bayesian relational event model (BREM) which combines the relational event model (Butts, 2008; Leenders et al., 2016) with novel Bayesian methods (Mulder, 2014a, 2016).


Prof. dr. Roger Leenders, dr. ir. Joris Mulder

Financed by



1 October 2018 – 1 October 2022