Not All Bonds Are Created Equal: Dyadic Latent Class Models for Relational Event Data

I am pleased to invite you to the upcoming “Network Science” event, jointly organised by the Special Interest Group (SIG) on Network Science and the Centre for Complex Systems Studies (CCSS).

For this session, we are very happy to welcome Prof. dr. Joris Mulder (Tilburg University), who will give a talk entitled:
“Not All Bonds Are Created Equal: Dyadic Latent Class Models for Relational Event Data”. Please see the abstract below.

This event is open to everyone and offers a great opportunity for those interested in network science and complex systems to learn from an expert speaker, exchange ideas, and connect with colleagues.

The meeting will take place on 19 February 2026, from 15:15 to 17:00, in Room 4.16, Minnaert Building, Utrecht University. Refreshments will be available from 15:10.

Please save the date, time and find the programme below:

Programme

Abstract

Dynamic social networks can be represented as time-ordered sequences of interactions between pairs of individuals. The relational event model (REM) is a central tool for analysing such data, and existing approaches to unobserved heterogeneity typically rely on actor-level latent blocks. Under this framework, any latent variation in a REM coefficient has to be expressed through the latent classes assigned to the sender and receiver, because the model allows effects to vary only across those actor-level groupings. Yet many social processes generate pair-specific tendencies that cannot be decomposed into individual characteristics. Examples include persistent cooperation or conflict between particular country pairs, longstanding friendships or hostilities between individuals, or dyads with idiosyncratic histories. To capture this genuinely dyadic structure, we introduce a dyadic latent class relational event model (DLC-REM) in which effects depend on a dyad’s latent class. This more flexible parameterization allows unobserved heterogeneity to be represented where it actually occurs at the level of the dyad. Through simulations, we show that the DLC-REM can capture a broader range of data-generating processes than actor-based latent models—often with considerably fewer parameters. We illustrate the methodology using both simulated examples and an empirical application to relational event data.

We look forward to seeing many of you there.