My Current View
The choice between SIENA and ERGM should be driven by temporal structure of data and question, not researcher habit or software familiarity.
- Multiple waves + change question → SIENA first
- Single network + structural association question → ERGM first
- Both may inform the same project at different analytical stages
Supporting Evidence
- SIENA models tie change between observations; ERGM models probability of a static graph
- Misapplied ERGM on pooled panels flattens dynamics; SIENA on one wave is ill-defined
- My VC imprinting work needed both: snapshots for critical periods, panels for evolution
Counterarguments
When only one cross-section exists, ERGM is appropriate and SIENA is not an option. Tool choice is constrained by design, not philosophy.
Related
Counterarguments
- Some research questions are inherently cross-sectional; forcing longitudinal framing misleads.
- ERGM degeneracy and SIENA convergence issues can make neither tool usable on sparse networks.