Modelling Dynamic Networks

Satellite meeting at the NetSci 2025 conference in Maastricht (June 3rd, 2025). 

We are pleased to invite you to join us for a half-day satellite workshop on “Modelling Dynamic Networks”, part of the Netsci 2025 Conference held in Maastricht on June 2-6, 2025. This atellite workshops will be held on Tuesday June 3rd, 2025 from 2.30pm to 6.30pm (Room FSE C1.015 at PHS1). Details on the schedule will be made available.

This workshop centres around the latest advancements in dynamic network modelling. It aims to cover a broad range of topics including Relational Event Modelling, Temporal Networks, Stochastic Actor-Ariented and Dynamic Actor-Oriented Models. It focuses on discussing both theoretical advancements and practical applications in dynamic network analysis.

Abstract Submission and Registration

Our workshop invites submissions for poster or contributed talks. There is also the possibility to simply attend without submitting any work. Register your attendance and submit your work using this submission/registration link by Feb. 14th, 2025. Response on your submission will be sent out before 20 February to facilitate early-bird registration at the main conference (Feb. 24th). We look forward to your participation in this exciting workshop!

Invited Speakers

(Confirmed so far)

Viviana Amati (University of Milano-Bicocca)

Random coefficient models for relational event data

Traditionally, the analysis of relational event data has focused on case studies that examine a single sequence of relational events. In this study, we shift our focus to the multilevel analysis of relational events, where the data consists of multiple sequences that are conceptually similar but collected from distinct groups of entities, with no interactions occurring between these groups. These sequences can be viewed as independent replications of the same event dynamics, reflecting the underlying social processes that generated the events. Such data often arises from relational event sequences collected across different geographical regions or from repeated studies involving separate groups of entities.

We propose procedures based on descriptive statistics and random coefficient models for the joint analysis of multiple independent relational event sequences. We also evaluate how well the observed event dynamics generalize across various contexts and groups. Our examination includes various methodological approaches, and we utilize simulations to identify situations where meta-analysis and random coefficient models are more effective. To demonstrate the application of the proposed procedures, we analyze multiple sequences of relational events collected in different social settings.

Rūta Juozaitienė (Vytautas Magnus University)

Relational state modelling

Researchers across diverse domains are increasingly considering various phenomena from a dynamic network perspective. This approach is based on the idea that nodes (e.g., individuals) are connected through ties (e.g., social relations) that can change over time. Once ties are established, they can flourish and evolve into close relationships over time or, alternatively, dissolve. The problem we want to tackle in this study is how we can model different types of dynamic ties at a mathematical process as how to infer such models from empirical data.

In this study, we conceptualize different levels of ties as states, emphasizing their relatively enduring nature rather than viewing them as brief events. We propose a collection of multivariate heterogeneous Poisson processes whereby underlying hazard rates describe the tendency for certain edges to transition from one state to another.

For the inferential aspects, we consider two types of data sampling procedures: (1) a complete data scenario, whereby for each edge the exact transition from one state to another is recorded and (2) data collected using a panel study approach, meaning that cross-sectional sequences of networks involving the same set of nodes are observed at different points in time. These repeated snapshots provide insights into the rate at which new ties form and how long they persist.

To analyze these data, we propose an estimation approach based on relational event models, which describe the waiting times between network changes. These models assume that within any given time frame, a specific set of relational events may occur, with each event’s rate depending on endogenous and exogenous factors. More commonly observed events tend to have a higher occurrence rate, while less frequent events have a lower rate. The rates of relational events can be modelled in terms of a hazard function. However, applying relation event models within a relational state framework presents challenges. Specifically, it requires appropriate definitions of endogenous network effects (e.g., reciprocity and triadicity) as well as suitable estimation techniques to account for the complexities of network evolution.

Nynke Niezink (Carnegie Mellon University)

Modeling the co-evolution of directed and undirected relational events: Cooperation and conflict in organized crime

The dynamics network actor model (DyNAM) was proposed for relational event analysis from an actor-oriented perspective. In this framework, network actors are assumed to decide who they send their ties to. For directed relational events, with a clear sender and receiver, this assumption decomposes the model likelihood in a term for actors’ activity rate (i.e., their tendency to initiate interactions) and a term for actor choice (i.e., whom actors choose to interact with). Parameters in the two model components can thus be estimated separately. For undirected relational events, no such decomposition is possible and, in applications so far, the activity rate in DyNAMs for undirected events is assumed to be constant. In the talk, we propose a methodology to estimate rate models for undirected relational events in the actor-oriented framework using ideas from multiple imputation.
 
We apply this method to study the co-evolution of cooperation and conflict among organized crime groups (OCGs), leveraging police record data (2004-2015) provided by Thames Valey Police – the largest non-metropolitan police force in England and Wales. The data consist of 25,977 organized crime-related events in which at least one OCG member was involved. In the study, we focus on 147 OCGs, which were involved in 79 cooperation and 141 conflict events. We find a strong interdependence between the two criminal networks among the OCGs. Yet, the event rate and partner choice of the groups are differentially affected by their network position (e.g., degree, embeddedness) and partner characteristics (e.g., primary criminal activity) for the two types of relational events. As the co-evolution of cooperation and conflict in the criminal context so far remained largely unexplored, this study fills a gap in our understanding of the dynamics underpinning organized crime operations.

Christoph Stadtfeld (ETH Zurich)

Available soon…

Veronica Vinciotti (University of Trento)

Causal discovery for dynamic network data

Identifying the mechanisms underlying the formation of a link between interconnected units is of primary interest in network science. Although this question is fundamentally of a causal nature, existing modelling and inferential procedures are exclusively descriptive and associative. In this talk, I will introduce causality in the identification of the underlying drivers of network processes. In the specific of dynamic network models, I will show how a causal relational event model satisfies a set of population conditions that uniquely identifies the causal drivers. The empirical analogue of these conditions provides a consistent causal discovery algorithm, which distinguishes itself from other inferential approaches. Crucially, a single observational dataset of the dynamic network process and of its potential drivers is sufficient for the proposed causal discovery approach. An analysis of recent bike sharing data in Washington D.C. finds that geographical, climate, temporal as well as endogenous variables are causal drivers of bike sharing.

Program

Stay tuned…

Organising Committee

Ernst C. Wit (Università della Svizzera Lugano, Switzerland), Gesine Reinert (University of Oxford, UK), Vladimir Batagelj (University of Ljubljana, Slowenia), Anuška Ferligoj (University of Ljubljana, Slowenia), Claire Gormley (School of Mathematics and Statistics, University College Dublin, Ireland), Sarika Jalan (Indian Institute of Technology Indore, India, Complex Systems Lab), Goeran Kauermann (Ludwig-Maximilians-Universität München, Germany), Eric D. Kolaczyk (Boston University, USA), Clelia Di Serio (Università della Svizzera Italiana, Lugano, and San Raffaele Vita-Salute University, Milano, Italy), Veronica Vinciotti (University of Trento, Italy).

For inquiries or further information, please contact Melania Lembo at melania.lembo@usi.ch.