A relational event approach for jointly modeling event rates and event duration

Relational event models are used to study what drives actors in a social network to interact with each other and when. A key feature of these models is that they allow researchers to take the event history into account, resulting in a time-sensitive analysis. The central question is then how the event history can be summarized to explain social interaction behavior and predict when the next event is likely to occur and who will be involved. This chapter contributes to this central question by proposing a methodology that allows researchers to study how the duration of past events affects social interaction behavior. An estimation procedure is proposed to learn the non-linear impact of event duration on interaction rates. Additionally, an extension of the relational event model is proposed that can be used to study how the event history and other sources of information (e.g., individual characteristics) affect the duration of new events. The performance of the approach is evaluated using numerical simulations. Two case studies reveal that if we take the impact of past events on future interaction behavior into account, we can better predict who interacts when.