Modelling Interaction Duration in Relational Event Models

Rumana Lakdawala, Roger Leenders, Peter Ejbye-Ernst, Joris Mulder · 2026

Read Modelling Interaction Duration in Relational Event Models by Rumana Lakdawala, Roger Leenders, Peter Ejbye-Ernst, Joris Mulder (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-21000.txt.

Introduction

Relational events are interactions between actors over time, from team messages to violent encounters. Rumana Lakdawala, Roger Leenders, Peter Ejbye-Ernst, and Joris Mulder note that standard models track when events happen but ignore how long they last. Their Duration Relational Event Model, presented in this 2026 arXiv paper, brings duration into the analysis.

Past event durations enter the endogenous statistics, so long past events can shape the rate of future ones, and the model also predicts when events end from history and covariates. The risk set widens to cover ongoing events that may end and idle pairs that may start. Case studies on team dynamics and interpersonal violence illustrate the method, which ships in a new R package.

Catalog record

CREATORRumana Lakdawala, Roger Leenders, Peter Ejbye-Ernst, Joris Mulder
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE80 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-21000

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Modelling Interaction Duration in Relational Event Models Rumana Lakdawala, Roger Leenders, Peter Ejbye-Ernst, Joris Mulder Abstract The study of relational events, which are interactions occurring between actors over time, has gained significant traction recently. Traditional relational event models typically focus on modelling the occurrence and sequence of events without considering their duration even though duration information is frequently available in empirical relational event data. We introduce a novel Duration Relational Event Model (DuREM) that incorporates the temporal duration of events into the analysis. The proposed model extends the existing framework by (i) allowing the inclusion of past event durations in the endogenous statistics to account for how the duration of past events affects the rate of future interactions, and (ii) extending the traditional relational event model by also modelling when events will end based on past event history and covariates. This is achieved by extending the risk set to include both ongoing events at risk of ending and idle dyads at risk of starting new events. The methodology is implemented in a new R package ‘durem’. Two case studies concerning team dynamics and inter-personal violence are presented to illustrate the applicability of the model. 1 Introduction The analysis of social interaction processes has helped to address significant empirical questions in areas such as development of conflicts ( 7 ; 18 ; 3 ; 12 , e.g.,) , collaborative or organizational dynamics ( 19 ; 20 ) and the coordination of emergency response efforts ( 5 ) . To study these processes, several statistical methods have been developed, including dynamic models for social networks ( 15 ; 29 ) , discrete temporal models ( 13 ) , separable dynamic network models ( 17 ) , and point process modelling approaches ( 28 ) . Among these, relational event models (REMs), introduced by 5 and further developed by 31 ; 30 ; 28 , have become increasingly prominent. REMs are particularly valuable because they incorporate the history of past social interactions to examine how social behaviours evolve over time, offering a detailed understanding of temporal social processes. REMs differ from other social network models by focusing on the temporal dynamics of interactions, rather than on static or aggregated representations of relationships. Traditional social network models often capture relationships as enduring ties or connections between actors and analyse patterns like density, clustering, or centrality. These models are useful for understanding the broader architecture of network structures but typically do not account for the timing or evolutionary nature of the interactions. In REMs, the time to the next relational event between two actors is directly modelled, thereby allowing researchers to study the impact of key predictors on the social interaction rates across actors over time. By incorporating both temporal and structural dependencies, REMs offer a more dynamic view of how ties evolve and change. This temporal focus makes REMs particularly useful for studying phenomena where the order and timing of events are critical. For instance, understanding how and how fast information spreads through a network, how conflicts escalate, or how collaboration patterns emerge requires a model that captures not just who interacts with whom, but when and how often these interactions occur. REMs also allow for the inclusion of endogenous network factors such as reciprocity, transitivity, or inertia providing insights into how prior interactions shape future behaviours. These capabilities make REMs a powerful and flexible tool for examining a wide range of social interaction mechanisms in understanding network dynamics. Standard REMs were originally developed for analysing instantaneous events between individuals observed at a single point in time, such as radio transmissions ( 5 , e.g.,) . However, in practice, many interactions have a duration attached to it. For instance the length of a phone call, the amount of time spent in conflict, the duration of a collaboration, and the length of face-to-face communication. As REMs aim to understand interaction dynamics between actors over time, modelling the duration of relational events is as important as modelling the time between events. If a dyad has frequent interaction or its interactions tend to have a very long duration, both imply that the actors in the dyad are ‘strongly connected’ or have a strong relation with each other. Longer durations may indicate deeper engagement, greater investment, or more complications in a conflict. Alternatively, brief interactions may reflect fleeting exchanges, information exchange, or procedural communication. By incorporating duration, we capture a richer, more nuanced description of relational event data, which can reveal richer patterns of behaviour, relational dynamics, and the underlying structure of social networks. This perspective is especially relevant in contexts where the timing and length of interactions shape outcomes, such as negotiations, collaborations, or emergency crisis management. The influence of event duration has not been explored much however. One study explored how people allocate time to various activities within a day ( 24 ) , and another examined group entry and exit dynamics ( 14 ) . However, little attention has been given to how the duration of past events themselves influences interaction rates or affect relational dynamics. This gap leaves open questions about the relationship between event duration and the dynamics of social interactions. In this article we propose a duration relational event modelling framework (DuREM) for jointly modelling the duration of events and event occurrence. The proposed model extends the existing REM framework (i) by incorporating the duration of past events in the endogenous statistics allowing one to study how the duration of past events affects the interaction rates between dyads and (ii) by jointly modelling both the start and end of events based on the past event history and covariates. The first extension is achieved by modelling the weight of past events according to their duration using an unknown parameter which is learned from the data. The second extension introduces a joint model for the rates at which events start and end. This is achieved by extending the risk set to include ongoing events that are at risk of ending, as well as idle dyads that are at risk of starting new events. The inclusion of duration-weighted endogenous statistics is comparable with REM extensions regarding memory effects where the weight of past events depends on the elapsed time since a past event was observed ( 4 ; 2 ; 1 , e.g.,) . Using a weighting scheme, analogous to the memory effect, the model can quantify the degree to which the duration of past events impacts interaction patterns. The remainder of the paper is organized as follows. In Section 2 we first introduce the standard Relational Event Model (REM) followed by introduction to the proposed Duration Relational Event Model (DuREM). We present two empirical applications of DuREM in Section 3 : a case study on research team dynamics in Section 3.1 followed by a case study on the dynamics of interpersonal violence in Section 3.2 . The article ends with a brief discussion in Section 4 on the model limitations, connection to related works and future opportunities for research. 2 Joint modelling the duration of events and event occurrence In this section, we first present the standard relational event model and notations (Section 2.1 ). Next, we present a novel modelling framework for incorporating

Read the complete text