Exponential Random Graph Models
Statistical Modeling of Network Formation Processes
Network Statistics
Total Nodes
0
Total Edges
0
Edge Density
0.00
Triangle Count
0
Clustering Coefficient
0.00
2-Star Count
0
Average Degree
0.00
Understanding Exponential Random Graph Models (ERGMs)
Exponential Random Graph Models (ERGMs) are a family of statistical models for analyzing social networks as random graphs. They allow researchers to test hypotheses about network structure by modeling the probability of observing a particular network configuration based on various network statistics.
The ERGM Framework:
- Probability Function: P(Y = y) = (1/κ) × exp(Σ θ_k × g_k(y))
- Parameters (θ): Coefficients that indicate the importance of different network features
- Statistics (g): Structural features like edges, triangles, or degree distributions
- Normalization (κ): Ensures probabilities sum to 1 across all possible networks
Key ERGM Parameters:
- Edge Parameter (θ_edges): Controls overall network density (usually negative)
- Triangle Parameter (θ_triangles): Measures tendency toward transitive closure and clustering
- 2-Star Parameter (θ_2stars): Captures tendency for nodes to have similar degrees
- Positive Triangle Parameter: Increases clustering - "friends of friends become friends"
- Negative 2-Star Parameter: Prevents hubs and promotes degree homogeneity
Estimation and Simulation:
- MCMC Sampling: Markov Chain Monte Carlo methods simulate networks consistent with parameters
- Goodness of Fit: Compare observed network statistics with simulated distributions
- Degeneracy: Some parameter combinations lead to unrealistic all-or-nothing networks
- Convergence: MCMC chains must converge to stable distributions for valid inference
Applications in Organizational Research:
- Advice Networks: Test whether expertise or hierarchy predicts advice-seeking ties
- Collaboration Networks: Examine role of past collaboration, homophily, and structural balance
- Communication Networks: Analyze formation of informal communication channels
- Innovation Networks: Study knowledge transfer and boundary spanning behaviors
Interpreting Parameters:
- Positive coefficients indicate the feature makes networks more likely
- Negative coefficients indicate the feature makes networks less likely
- Magnitude indicates strength of effect on network formation probability
- Parameters are interpreted while controlling for other network features
