Bayesian Exponential Random Graph Model
Bayesian Exponential Random Graph Model (Bayesian ERGM) · Also known as: Bayesian ERGM, Bayesian p-star model, Bayesian p* model, BERGM
The Bayesian Exponential Random Graph Model (Bayesian ERGM or BERGM) extends the classical ERGM framework by placing prior distributions over the model parameters and using Markov chain Monte Carlo methods to obtain full posterior distributions. Introduced by Caimo and Friel (2011), it allows researchers to quantify parameter uncertainty and incorporate prior knowledge when modelling the structural features of social and other complex networks.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Bayesian ERGM when you want to model the structural determinants of a single observed network — reciprocity, closure, degree heterogeneity — and need full posterior uncertainty rather than point estimates. It is especially appropriate when the network is small (fewer than 200–300 nodes), prior information is available, or you need to compare nested models via Bayes factors. Avoid it for very large networks (thousands of nodes) where MCMC mixing becomes prohibitive, or when the research question calls for predicting link formation over time rather than characterising a static network snapshot.
Strengths & limitations
- Provides full posterior distributions over parameters, enabling principled uncertainty quantification beyond standard errors.
- Naturally incorporates prior knowledge from theory or previous studies through the prior distribution.
- Supports formal Bayesian model comparison via Bayes factors or posterior predictive checks.
- Handles small networks more reliably than maximum-likelihood ERGM, which can suffer from near-degenerate likelihoods.
- Retains all the structural flexibility of the classical ERGM family: reciprocity, transitivity, k-stars, and covariate effects.
- Computationally intensive: the exchange MCMC algorithm requires simulating auxiliary networks at every iteration, making it slow for networks with more than a few hundred nodes.
- Requires careful prior specification; vague priors can still be informative in the ERGM parameterisation due to the non-linear relationship between parameters and network density.
- Mixing and convergence of the MCMC chain can be difficult to diagnose and achieve, especially for complex model specifications.
- Software support is narrower than for classical ERGM; the R package Bergm is the primary implementation.
Frequently asked
How does Bayesian ERGM differ from classical ERGM?
Classical ERGM uses maximum likelihood estimation (via MCMCMLE) and produces point estimates with standard errors. Bayesian ERGM places priors on the parameters and uses MCMC to obtain full posterior distributions, providing credible intervals and supporting formal model comparison via Bayes factors.
What software implements Bayesian ERGM?
The primary implementation is the R package Bergm (Caimo & Friel), which builds on the ergm package in the statnet suite. It provides functions for posterior sampling, convergence diagnostics, and posterior predictive goodness-of-fit checks.
How large a network can Bayesian ERGM handle?
In practice, networks up to roughly 200–300 nodes are tractable with current MCMC implementations. Larger networks make auxiliary-network simulation at each MCMC step very slow, and mixing can deteriorate. For large networks, approximate methods or standard ERGM with bootstrap standard errors are typically preferred.
How should I choose priors?
A common default is a multivariate normal prior centred at zero with moderate variance (e.g., variance of 4 or 9 on each parameter). If you have results from a previous study or a related network, those estimates can inform a tighter prior. Always perform a prior sensitivity check by refitting with a more diffuse prior to assess how much the posterior depends on prior choice.
What goodness-of-fit diagnostics should I report?
Report posterior predictive distributions of key network statistics not included in the model — such as the geodesic distance distribution, triad census, or degree distribution — and compare them visually and numerically to the observed values. Also report MCMC trace plots and effective sample sizes for each parameter.
Sources
- Caimo, A., & Friel, N. (2011). Bayesian inference for exponential random graph models. Social Networks, 33(1), 41–55. DOI: 10.1016/j.socnet.2010.09.004 ↗
- Exponential random graph models. Wikipedia. link ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Exponential Random Graph Model (Bayesian ERGM). ScholarGate. https://scholargate.app/en/network-analysis/bayesian-exponential-random-graph-model
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bayesian Social Network AnalysisNetwork analysis↔ compare
- Bayesian Stochastic Block ModelNetwork analysis↔ compare
- Modularity AnalysisNetwork analysis↔ compare
- Stochastic Block ModelNetwork analysis↔ compare