A tutorial on Bayesian model averaging for exponential random graph models.

Area of Science:

• Social Psychology
• Network Science
• Computational Statistics

Background:

• Exponential Random Graph Models (ERGMs) are increasingly used in psychology to analyze network structures.
• Valid ERGM inference depends on accurate specification of network statistics, including endogenous and exogenous effects.
• Current practices often rely on single models, risking misspecification and ignoring model selection uncertainty.

Purpose of the Study:

• To introduce and guide the implementation of Bayesian model averaging (BMA) for ERGMs.
• To demonstrate how BMA enhances robustness to model misspecification in network analysis.
• To address the challenge of model selection uncertainty in ERGM applications.

Main Methods:

• Utilizing Bayesian model averaging (BMA) to evaluate multiple candidate ERGM models.
• Incorporating theoretical considerations in model specification for network statistics.
• Providing annotated R code for practical implementation and replication.

Main Results:

• BMA accounts for uncertainty in both parameter estimation and model selection.
• The approach is demonstrated to be more robust to model misspecification than single-model inference.
• Application examples using college friendship and Florentine marriage networks highlight averaging exogenous covariate effects.

Conclusions:

• BMA provides a more reliable framework for ERGM analysis in psychology and related fields.
• This method improves the validity of statistical inference by addressing model uncertainty.
• The tutorial facilitates the adoption of BMA for advanced network analysis.