Agent-based models increasingly rely on networked synthetic populations derived from multisource empirical data. Yet the process by which empirical summaries (e.g., mixing structures, degree distributions, geocoded location data) are translated into simulated networks is often ad hoc. The opacity and heuristic nature of the underlying workflows can be prohibitively intimidating for new users and difficult to reproduce even for experienced practitioners. This workshop focuses on workflow design, diagnostics, and judgment in generating networked populations using exponential random graph models (ERGMs) for social simulation. The underlying example will focus on modelling syringe-sharing networks to simulate vaccine interventions in an agent-based modelling framework. Using ERGMs, participants will work through key stages of a reproducible pipeline: defining network targets from empirical data, including geographic attributes; stepwise ERGM fitting; diagnosing failure modes (e.g., model convergence); and assessing alignment between simulated and target networks. The workshop emphasizes reproducibility, transparency, and explicit modelling choices, aiming to develop shared standards of practice that integrate both the art and technical science of network modelling.

The workshop comprises 6 modules of about 30 minutes each.

1. Aggregate summaries → network parameters
2. Geocoded attributes → social mixing
3. Sequential ERGM specification
4. Assessing Failure Modes
5. Simulation as diagnostic
6. Networks → ABMs