Long-term evolution of email networks: Statistical regularities, predictability and stability of social behaviors

In social networks, individuals constantly drop ties and replace them by new ones in a highly unpredictable fashion. This highly dynamical nature of social ties has important implications for processes such as the spread of information or of epidemics. Several studies have demonstrated the influence of a number of factors on the intricate microscopic process of tie replacement, but the macroscopic long-term effects of such changes remain largely unexplored. Here we investigate whether, despite the inherent randomness at the microscopic level, there are macroscopic statistical regularities in the long-term evolution of social networks. In particular, we analyze the email network of a large organization with over 1,000 individuals throughout four consecutive years. We find that, although the evolution of individual ties is highly unpredictable, the macro-evolution of social communication networks follows well-defined statistical patterns, characterized by exponentially decaying log-variations of the weight of social ties and of individuals’ social strength. At the same time, we find that individuals have social signatures and communication strategies that are remarkably stable over the scale of several years.

💡 Research Summary

This paper investigates whether long‑term macro‑level regularities emerge in social communication networks despite the inherent randomness of tie turnover at the micro‑level. Using a four‑year longitudinal dataset of email exchanges within a large organization (over 1,000 employees), the authors construct monthly weighted networks where nodes represent individuals and edge weights correspond to the number of emails exchanged. They quantify each individual’s “social strength” as the sum of incident edge weights and define a multidimensional “social signature” for each person based on eight behavioral metrics (e.g., proportion of communication with top‑5 % partners, average daily volume, weekly periodicity, diversity of contacts, etc.).

At the micro‑scale, the study finds that roughly 30 % of ties are either dropped or newly created each year. Logistic regression and random‑forest models using predictors such as past interaction frequency, hierarchical position, and departmental proximity achieve only modest predictive performance (AUC ≈ 0.55), confirming that the fate of any specific tie is largely unpredictable.

In contrast, macro‑scale dynamics display clear statistical patterns. The logarithmic variations of edge weights (Δlog w) and of individual social strength (Δlog s) follow distributions with means near zero and decreasing standard deviations over time (from ≈0.12 in the first year to ≈0.08 in the fourth). This exponential decay indicates that the network rapidly converges toward a statistical equilibrium where large fluctuations become rare. Moreover, the overall weight distribution shifts from an initial power‑law‑like tail toward a log‑normal shape as the system matures, suggesting a homogenization of interaction intensities.

A striking result concerns the stability of individual social signatures. Pairwise cosine similarity of a person’s signature vectors across the four years averages 0.87 (SD = 0.05), demonstrating that each employee maintains a remarkably consistent communication strategy over several years. Cluster analysis reveals that these signatures align with departmental and hierarchical structures, implying that organizational role strongly shapes long‑term behavior.

To validate that the observed regularities are not artifacts of random processes, the authors compare the empirical data with two null models: (1) a time‑shuffled randomization that destroys temporal correlations, and (2) a first‑order Markov model that preserves transition probabilities but not higher‑order dependencies. Both null models exhibit slower decay of log‑variations and substantially lower signature stability (p < 0.001), confirming that the real network’s dynamics are governed by mechanisms beyond simple randomness.

The discussion extrapolates these findings to practical domains. In epidemic or information‑diffusion modeling, the emergence of a predictable macro‑state implies that forecasting spread becomes more reliable once the network has settled into its equilibrium regime. For organizational design, the persistence of individual communication signatures suggests that reshuffling teams without accounting for entrenched interaction patterns may reduce efficiency, whereas leveraging stable ties can enhance coordination.

Limitations include the exclusive reliance on email data, which omits instant messaging, face‑to‑face meetings, and other informal channels, and the focus on a single large‑enterprise context, which may limit generalizability. Future work is proposed to integrate multimodal communication streams, examine smaller or more heterogeneous organizations, and explore how external shocks (e.g., organizational restructuring, remote‑work transitions) perturb the identified statistical regularities.

Overall, the study demonstrates a dual nature of social networks: while individual tie changes are essentially stochastic, the aggregate system obeys well‑defined statistical laws, and individuals exhibit long‑lasting behavioral “signatures.” This insight bridges micro‑level unpredictability with macro‑level predictability, offering a nuanced framework for both theoretical modeling and practical management of dynamic social systems.