Summary
SIENA (Simulation Investigation for Empirical Network Analysis) models longitudinal network change as a stochastic process where tie formation and dissolution depend on network structure, actor attributes, and covariates. Unlike static ERGM snapshots, SIENA estimates how networks evolve over multiple observation waves.
The framework separates structural effects (reciprocity, transitivity, popularity), covariate effects (actor-level and dyadic), and behavioral co-evolution when actors' outcomes change alongside ties.
Key Points
- Networks are observed at discrete time points; change between waves is decomposed into micro-steps
- Estimation uses Method of Moments with Robins' simulation algorithm
- Can jointly model network dynamics and behavior dynamics (co-evolution)
- Structural effects mirror ERGM intuition but apply to change rather than static equilibrium
- Implemented in R via
RSiena— the standard tool for longitudinal network panel data
This is foundational for my master's work on venture capital networks. When the question is how networks form and change over time rather than what a single network snapshot looks like, SIENA is the appropriate starting point. I keep ERGM for cross-sectional analyses and SIENA when panel waves are available.
My SIENA & ERGM project repository collects implementation examples.