Dynamic Causal Graphs: Time-Varying Chain Graphs for Social Influence and Contagion

Ogburn et al. (2018) introduced chain graphs as a parsimonious way to model static causal relationships under interference and contagion. However, real-world networks are highly dynamic: ties form and dissolve, influence pathways shift, and the structure of contagion changes over time (see also Clark & Handcock, 2024). This research proposes to develop “dynamic chain graphs”—a formalism and inference procedure for time-indexed chain graphs that evolve in response to both endogenous (e.g., user churn, group formation) and exogenous (e.g., platform changes, world events) factors. By using sequential data and possibly leveraging advances in lifted causal inference (Luttermann et al., 2024), the model could infer when and how causal influence pathways reconfigure, offering both better fit and richer insights (e.g., identifying critical “tipping points” for interventions). This would not only improve the realism of causal inference models but would also support adaptive policy design on platforms where network structure is in constant flux.

References:

1. Causal inference, social networks and chain graphs. Elizabeth L. Ogburn, I. Shpitser, Youjin Lee (2018). Journal of the Royal Statistical Society. Series A,.
2. Causal inference over stochastic networks.. Duncan A Clark, M. Handcock (2024). Journal of the Royal Statistical Society. Series A,.
3. Lifted Causal Inference in Relational Domains. Malte Luttermann, Mattis Hartwig, Tanya Braun, Ralf Moller, M. Gehrke (2024). CLEaR.