12  Introduction

This module addresses the practical application of statistical models for static and dynamic networks. It begins with data requirements, moves on to estimation and model assessment, and spends some time on the issues that can arise with model specifications that use dyad-dependent terms. It ends with a final lab – demonstrating how these tools can be used to conduct a principled analysis of a static network, and use the fitted model to simulate networks that reliably reproduce the observed structure. Simulations like these provide a solid foundation for epidemic modeling on networks.

12.1 Module Learning Objectives

• Understand the data needed to estimate ERGMs and STERGMs, and in particular how egocentric network study designs can be used in this context.
• Understand how the MCMC algorithm is re-purposed here to serve multiple goals across the workflow: maximum likelihood estimation, goodness-of-fit assessments and network simulation from fitted models.
• Learn the basics of diagnosing problems with MCMC convergence
• Understand how goodness of fit assessments against excluded network features can be used to validate models
• Understand how model degeneracy arises with dyad-dependent specifications, why it is a form of model misspecification, and how to avoid it
• Develop confidence in model specification, assessment and simulation with a hands-on lab assignment.

Bottom line, take-away:

• EpiModel relies on a uniquely powerful and general framework for representing the networks on which epidemics spread.
• The framework leverages statistical principles of sufficiency, Markov dependence and sampling to dramatically reduce data requirements.
• And it provides modelers with a solid foundation for data-driven simulation of networks that reliably reproduce the wide range of patterns observed in real data.