48 Introduction
This module collects several advanced EpiModel topics that a research-level modeler needs but that do not belong to any single earlier module. It has two parts.
The first is multi-layer networks: models with one shared node set but two or more edge sets, representing distinct kinds of contact (main versus casual partnerships, household versus community, sexual versus needle-sharing), each with its own formation and dissolution model and possibly interacting with the others. The tutorial builds a two-layer epidemic from independent layers up to cross-dependent ones.
The second is a set of advanced practical topics that recur in research work: running an epidemic over a fully observed dynamic network without fitting an ERGM; the model-running interfaces for partnership histories (contact tracing, partner notification), custom output statistics, and time-varying parameters and scenarios; how to choose a defensible number of simulations; and a guide to improving estimation and simulation when a network model will not converge or recover its targets, working through full TERGM estimation and the ERGM and TERGM fitting controls. Several are documented in full in the EpiModel package vignettes; here we give the concept, a worked pointer, and when to reach for each. The consolidated control.net guide now sits at the end of Module 9, alongside the extension API it wires together.
48.1 Slides
48.2 Module Learning Objectives
- Define a multi-layer network and implement one in EpiModel, from independent layers up to cross-dependent layers, and run and interpret an epidemic on it.
- Query partnership histories with the cumulative edgelist and
get_partners()for contact-tracing-style models. - Record the output statistics a research question needs with
epi.by,mutate_epi(), and custom trackers. - Represent interventions and counterfactuals with time-varying parameters and scenarios.
- Choose a defensible number of simulations using a Monte Carlo convergence check.
- Diagnose why a network model will not converge or recover its targets, and know which ERGM and TERGM fitting controls to reach for.