A Poissonian explanation for heavy-tails in e-mail communication

Patterns of deliberate human activity and behavior are of utmost importance in areas as diverse as disease spread, resource allocation, and emergency response. Because of its widespread availability and use, e-mail correspondence provides an attractive proxy for studying human activity. Recently, it was reported that the probability density for the inter-event time $\tau$ between consecutively sent e-mails decays asymptotically as $\tau^{-\alpha}$, with $\alpha \approx 1$. The slower than exponential decay of the inter-event time distribution suggests that deliberate human activity is inherently non-Poissonian. Here, we demonstrate that the approximate power-law scaling of the inter-event time distribution is a consequence of circadian and weekly cycles of human activity. We propose a cascading non-homogeneous Poisson process which explicitly integrates these periodic patterns in activity with an individual’s tendency to continue participating in an activity. Using standard statistical techniques, we show that our model is consistent with the empirical data. Our findings may also provide insight into the origins of heavy-tailed distributions in other complex systems.

💡 Research Summary

The paper revisits the widely reported finding that the distribution of inter‑event times τ between consecutive e‑mails follows an apparent power‑law decay P(τ) ∼ τ⁻¹, a result that has been interpreted as evidence that deliberate human activity is fundamentally non‑Poissonian. The authors argue that this apparent heavy‑tail emerges naturally from two well‑known regularities of human behavior: (1) strong circadian (daily) and weekly cycles that modulate the baseline rate at which a person sends e‑mails, and (2) a tendency for individuals to continue an activity once it has been started, producing short bursts of rapid successive messages.

To capture these effects they introduce a “cascading non‑homogeneous Poisson process.” The first layer is a time‑dependent baseline intensity λ₀(t) that varies deterministically with the hour of day and day of week, reflecting the fact that people are far more active during working hours on weekdays than during night or weekends. The second layer models an “activity session”: each time an e‑mail is sent, with probability p₁ a high‑intensity session is initiated during which additional e‑mails are generated at a much larger rate λ₁. The session ends probabilistically, with the termination probability increasing with elapsed session time. This construction yields a mixture of a low‑intensity background Poisson process and high‑intensity bursts, both modulated by the same periodic λ₀(t).

Using a large corpus of real e‑mail logs (thousands of users, hundreds of thousands of messages), the authors estimate the model parameters by maximum likelihood. They find that λ₀(t) varies by roughly an order of magnitude between peak working hours and late night, and that the weekly pattern reduces the average rate on weekends to about 60 % of the weekday level. The session‑continuation probability p₁ averages 0.3–0.5 across users, implying that a single e‑mail typically triggers two to three additional messages in the same session. The burst intensity λ₁ is about five to ten times larger than the baseline λ₀(t).

Model validation is performed with several standard statistical tools. Kolmogorov–Smirnov tests between the empirical inter‑event time distribution and synthetic data generated by the model yield p‑values well above the conventional 0.05 threshold, indicating that the model cannot be statistically distinguished from the data. Likelihood‑ratio tests show that the cascading model fits significantly better than a simple non‑homogeneous Poisson process that only incorporates λ₀(t) without bursts. Moreover, the model reproduces the empirical distribution over the entire range of τ, from minutes to several hours, where the pure power‑law description fails to capture the curvature observed in the data.

The authors conclude that the heavy‑tailed inter‑event times do not necessarily imply intrinsically non‑Poissonian decision processes. Instead, the combination of predictable periodic variations in activity and a simple burst mechanism is sufficient to generate the observed τ⁻¹ scaling. This insight challenges earlier interpretations that attribute heavy tails to complex cognitive or memory effects, and suggests that similar mechanisms may underlie heavy‑tailed statistics in other domains such as phone calls, social‑media posts, or even natural phenomena like earthquakes.

Finally, the paper discusses limitations and future directions: (i) incorporating individual‑specific parameter heterogeneity more explicitly, (ii) extending the framework to other communication channels and to contexts where external shocks (e.g., emergencies) produce atypical spikes, and (iii) exploring analytical approximations that link the model parameters to the emergent power‑law exponent. Overall, the work provides a parsimonious yet powerful explanation for heavy‑tailed human activity patterns, emphasizing that periodicity and burstiness—rather than fundamentally non‑Poissonian dynamics—can be the primary drivers of the observed statistical regularities.