A Simpler, Better Way to Model Network Data: AME Beats ERGM and LSM

🔍 What This Paper Does

Introduces a Bayesian approach for inferential analysis of dyadic data that accounts for interdependencies through a set of additive and multiplicative effects (AME). The AME model is embedded in a generalized linear modeling framework, making it flexible for a variety of outcome types and substantive contexts.

🧩 How the AME Model Is Structured

• Uses a Bayesian estimation approach for dyadic outcomes.
• Models interdependence with additive and multiplicative latent effects (AME).
• Operates within a generalized linear model framework, so covariate effects remain interpretable in familiar GLM terms.

🧪 How AME Was Compared to Other Network Models

Contrasts the AME approach with two prominent alternatives: the latent space model (LSM) and the exponential random graph model (ERGM). Relative to these approaches, AME is shown to be:

• (a) Easy to implement
• (b) Interpretable within a general linear model framework
• (c) Computationally straightforward
• (d) Not prone to degeneracy
• (e) Able to capture first-, second-, and third-order network dependencies
• (f) Notably superior to ERGMs and LSMs on a variety of metrics and in out-of-sample contexts

📈 Key Findings

• AME combines flexibility and interpretability by marrying latent-factor interdependence with the GLM structure.
• Across multiple evaluation metrics and out-of-sample tests, AME outperforms both ERGM and LSM alternatives.

🌍 Why It Matters

AME offers a straightforward, principled route for nuanced inferential network analysis, suitable for a wide range of social science questions where dyadic interdependence must be modeled without sacrificing interpretability or computational tractability.