Weighted Exponential Random Graph Model
Weighted Exponential Random Graph Model (Valued-Edge ERGM) · Also known as: W-ERGM, valued ERGM, weighted p-star model, valued exponential random graph model
The Weighted Exponential Random Graph Model (W-ERGM) extends the classic binary ERGM framework to networks whose edges carry quantitative values — such as frequency of contact, trade volume, or collaboration intensity. It models the entire valued-edge network as a probability distribution defined over all possible weighted graphs, enabling researchers to test whether structural patterns such as reciprocity, transitivity, or degree distribution arise beyond what chance alone would produce.
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 W-ERGM when you have a single observed weighted network and want to test whether specific structural features — reciprocity, transitivity, degree heterogeneity — explain the observed pattern of edge weights beyond a null model, especially when the research question is explicitly about network formation or tie strength rather than node-level outcomes. It is appropriate for count-valued networks (number of interactions, citations, co-authorships) and can be extended to ordinal weights. Do not use W-ERGM when the network is very large (thousands of nodes) because MCMLE becomes computationally prohibitive; prefer weighted stochastic block models or spectral methods instead. It is also unsuitable when you have only aggregate node-level data rather than a full edge-weight matrix, or when the primary goal is prediction rather than inference about structural parameters.
Strengths & limitations
- Provides principled statistical inference — p-values and confidence intervals — for structural parameters in weighted networks.
- Incorporates theoretically motivated features (reciprocity, transitivity, degree distribution) as explicit model terms.
- Exponential-family framework connects to well-developed asymptotic theory and model comparison via AIC/BIC.
- Can jointly model the presence and weight of ties by combining binary and count ERGM components.
- Implemented in widely used open-source software (R ergm package with ergm.count extension).
- Computationally intensive: MCMLE for weighted networks is slower than for binary ERGMs and can fail to converge on large networks.
- Degeneracy remains a risk: poorly chosen statistics can yield a model that places near-all probability mass on empty or complete weight matrices.
- Inference applies to a single observed network; longitudinal or panel network data require separate frameworks such as TERGM.
- Continuous (non-count) edge weights lack mature off-the-shelf ERGM implementations and require custom samplers.
- Requires the full adjacency matrix, ruling it out when only aggregate node-level summaries are available.
Frequently asked
How does W-ERGM differ from the standard binary ERGM?
Standard ERGM models the presence or absence of each edge as a Bernoulli outcome. W-ERGM replaces that binary outcome with a count or ordered categorical weight, allowing the model to capture not just whether a tie exists but how strong it is. The exponential-family structure is the same; what changes is the support and the sufficient statistics.
Can W-ERGM handle continuous edge weights like correlation coefficients?
Current mainstream implementations (ergm.count) are designed for non-negative integer counts. Continuous weights require custom MCMC samplers or approximation via discretisation. This is an open research area; consult recent methodological literature before applying W-ERGM to continuous-weight data.
My MCMLE did not converge — what should I do?
Non-convergence usually means the model is degenerate or the statistics are poorly scaled. First simplify the model by removing high-order statistics; then try geometrically weighted terms (GWD, GWESP) with fixed decay parameters to reduce degeneracy. Increase the number of MCMC steps and inspect trace plots of each sufficient statistic.
How do I assess whether the model fits the observed network?
Use the goodness-of-fit (gof) function in the ergm package to simulate networks from the fitted model and compare simulated distributions of degree, edge-weight, and geodesic distance with the observed network. If observed statistics fall in the tails of the simulated distributions, the model is missing important structure.
Is W-ERGM appropriate for dynamic or longitudinal weighted networks?
No — standard W-ERGM is a cross-sectional model for a single network observation. For networks observed at multiple time points, use the Temporal ERGM (TERGM) or the Separable Temporal ERGM (STERGM) frameworks, which model tie formation and dissolution processes separately over time.
Sources
- Krivitsky, P. N. (2012). Exponential-family random graph models for valued networks. Electronic Journal of Statistics, 6, 1100–1128. DOI: 10.1214/12-EJS696 ↗
- Robins, G., Pattison, P., Kalish, Y., & Lusher, D. (2007). An introduction to exponential random graph (p*) models for social networks. Social Networks, 29(2), 173–191. DOI: 10.1016/j.socnet.2006.08.002 ↗
How to cite this page
ScholarGate. (2026, June 3). Weighted Exponential Random Graph Model (Valued-Edge ERGM). ScholarGate. https://scholargate.app/en/network-analysis/weighted-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.
- Weighted Degree CentralityNetwork analysis↔ compare
- Weighted Modularity AnalysisNetwork analysis↔ compare
- Weighted Social Network AnalysisNetwork analysis↔ compare
- Weighted Stochastic Block ModelNetwork analysis↔ compare