Relational Event Model
Also known as: REM, relational event framework, dynamic network event model, event-history network model
The relational event model (REM), introduced by Carter Butts in 2008, analyzes streams of time-stamped interactions — emails, radio calls, messages, citations — as a continuous-time event-history process. Rather than treating a network as a static set of ties, it models the instantaneous rate at which any sender directs an action at any receiver as a function of the history of past events, letting researchers test how prior interaction shapes future interaction.
Key highlights
- Exploits the full temporal ordering of interactions rather than discarding it by aggregating into static ties.
- Tests fine-grained dynamic mechanisms — reciprocity, turn-taking, recency, brokerage — directly on the event stream.
- Builds on well-understood event-history (Cox) machinery, giving interpretable rate-multiplier coefficients.
- Naturally accommodates time-varying actor and dyad covariates alongside endogenous network history.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use a relational event model when your data are a time-ordered stream of discrete interactions and you want to learn how the history of interaction drives its continuation — reciprocation, turn-taking, brokerage, popularity dynamics. It is ideal for communication logs, animal-interaction sequences, citation streams, and emergency-response radio traffic. It is not appropriate when only cross-sectional or panel tie data are available (use ERGM or SAOM), when events lack a meaningful order, or when the risk set is so large that computing event statistics over all candidate pairs at every step is infeasible without sampling. The model assumes events are conditionally independent given the history and the chosen statistics.
Strengths & limitations
- Exploits the full temporal ordering of interactions rather than discarding it by aggregating into static ties.
- Tests fine-grained dynamic mechanisms — reciprocity, turn-taking, recency, brokerage — directly on the event stream.
- Builds on well-understood event-history (Cox) machinery, giving interpretable rate-multiplier coefficients.
- Naturally accommodates time-varying actor and dyad covariates alongside endogenous network history.
- Requires fine-grained, correctly ordered event data; aggregated or coarsely timed data lose the information REM exploits.
- Computing the risk-set normalization at every event is expensive for large actor sets, often necessitating case-control sampling.
- Results depend on the chosen window/decay for history statistics (how far back reciprocity or recency counts), a non-trivial specification choice.
- Conditional-independence assumptions can be violated by burstiness and external shocks not captured by the statistics.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does a relational event model differ from a stochastic actor-oriented model?
Both model network dynamics in continuous time, but SAOM is designed for panel observations of a tie network and reconstructs the unobserved change between waves through actor-driven tie toggles. REM is designed for an observed stream of discrete, time-stamped events and models the rate of each next event given the full history. When you actually observe the timing and order of interactions, REM uses that information directly; SAOM is for when you only see periodic network snapshots.
Do I need exact event times, or is the order enough?
REM can be fitted with an ordinal likelihood that uses only the rank order of events when exact timestamps are unavailable, or with an interval likelihood that uses the actual inter-event times when they are known. Exact times add information about rates and the baseline hazard, but the ordinal version captures the sequence of mechanisms when only order is observed.
How are very large risk sets handled?
When the number of possible sender–receiver pairs is huge, evaluating the full denominator at each event is costly. Case-control sampling of non-events from the risk set yields consistent estimates at a fraction of the cost, and is the standard approach for large-scale event streams.
Sources
- 1.Butts, C. T. (2008). A relational event framework for social action. Sociological Methodology, 38(1), 155–200.
- 2.Snijders, T. A. B., van de Bunt, G. G., & Steglich, C. E. G. (2010). Introduction to stochastic actor-based models for network dynamics. Social Networks, 32(1), 44–60.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 22). Relational Event Model. ScholarGate. https://scholargate.app/sociology/relational-event-model