Fritz & Fuchs-Kreiss Unlock Practical Asymptotics for Relational Event Models

Cornelius Fritz and Alexander Fuchs-Kreiss present a counting process view of Relational Event Models (REMs), a framework for analyzing dyadic interactions observed in continuous time. By embedding REMs in counting process theory, they formalize the asymptotic properties of Maximum Likelihood Estimation (MLE) under three distinct regimes: when network size (n) approaches infinity, when the observational period (T) approaches infinity, or when both do. This is crucial because real-world network data often grows in different dimensions — a small chat group observed for years versus a massive social network sampled briefly. The authors focus on Cox-type multiplicative models and detail the core assumptions needed for asymptotic normality in each case.

Through simulation studies, the paper examines how structural modeling choices — particularly temporal windowing and logarithmic transformations — impact empirical coverage and estimator convergence. They derive guiding principles for specifying REMs in realistic contexts, showing that improper windowing can severely bias estimates of triadic closure and reciprocity. This work bridges a gap between theoretical guarantees and practical application, giving network scientists clear recommendations for model specification. The findings are directly applicable to fields like social network analysis, epidemiology, and organizational behavior, where dynamic relational data is common.