Mediation and Moderation Analysis in ERGM using ergMargins

Presenter: Scott Duxbury

Exponential random graph models (ERGM) are widely used in the social sciences to examine determinants of graph structure. This 3-hour workshop will introduce attendees to mediation and moderation analysis in ERGM using the ergMargins package for R. The workshop will describe why ERGM coefficients cannot be compared between models and why coefficients for interactions—including node matching, node mixing, and other common measures of homophily and heterophily—cannot be interpreted without adjustment. Topics covered will include (1) mediation analysis, (2) moderation analysis, (3) mediation analysis when the mediator is an interaction, and (4) mediation analysis when the main effect is an interaction. We will review a range of special cases, including interactions involving both continuous and discrete variables, necessary conditions for a causal interpretation, and mediation analysis involving endogenous graph statistics. Attendees will come away from the workshop with a deeper understanding of inferential difficulties in ERGM and with knowledge on how to address each issue using ergMargins.