pith. sign in

arxiv: 2604.07063 · v1 · submitted 2026-04-08 · 📊 stat.ME

Introduction to Relational Event Modelling

Martina Boschi , Ernst C. Wit This is my paper

Pith reviewed 2026-05-10 18:10 UTC · model grok-4.3

classification 📊 stat.ME
keywords relational event modelsevent-history analysisnetwork modellinggeneralized additive modelsdata simulationempirical applicationstime-dependent interactions
0
0 comments X

The pith

A hands-on tutorial shows how to model timed relational events by expressing them as generalized additive models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper seeks to lower the barrier for using relational event models by supplying theory alongside concrete steps for simulating synthetic data and fitting models to real examples such as emails or citations. It fills the noted absence of practical guides by walking through model fitting, inference choices, and comparisons across strategies. A reader would care if this lets more analysts study when and why sender-receiver interactions occur without requiring deep prior expertise in either network methods or event histories.

Core claim

Relational event models treat interactions between a sender and a receiver at a precise time as the basic unit of analysis; recent advances allow these models to be written as generalized additive models that capture complex non-linear effects, and the tutorial supplies the code and examples needed to simulate data under this framework and apply it to empirical cases while comparing modelling options.

What carries the argument

Relational event models expressed as generalized additive models, which encode time-dependent effects and sender-receiver dependencies to predict event occurrence.

If this is right

  • Analysts gain the ability to generate synthetic relational-event data sets that match chosen dependence structures.
  • Empirical studies can directly compare different modelling and inference strategies on the same interaction data.
  • Complex non-linear time patterns become identifiable in sender-receiver processes such as citations or transfers.
  • Event-history and network perspectives merge into a single usable framework for dynamic interaction data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Wider availability of working code could increase the number of studies that treat interaction timing as the primary outcome rather than an afterthought.
  • The same simulation approach might be adapted to test robustness of findings when event rates change over long observation windows.
  • Researchers studying very large networks could build on the tutorial's structure to develop scalable approximations.

Load-bearing premise

The presented simulation code and application examples correctly implement current best practices and that readers can follow them to apply the models on their own data.

What would settle it

Running the tutorial's simulation procedure on a small test data set and obtaining event counts or parameter estimates that deviate substantially from the expected distributions shown in the paper.

Figures

Figures reproduced from arXiv: 2604.07063 by Ernst C. Wit, Martina Boschi.

Figure 1
Figure 1. Figure 1: Synthetic relational case-control dataset showing academic collaborations between universities in different [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Network statistics computation in practice. The [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Top Left. Covariate columns of the case-control dataset, including values of the explanatory variable evaluated for both the event (rec_ev, dist_ev) and the corresponding non-event (rec_non_ev, dist_non_ev). For covariates assumed to have a linear effect, the dataset includes their difference (delta.rec). In cases where a covariate has a non-linear effect, two additional columns are included, with values s… view at source ↗
Figure 4
Figure 4. Figure 4: Top. Snapshot of relational event data structured for fitting a Poisson regression with one covariate x. For each event, ti is recorded in column stop and the preceding event ti−1 in start. The event column encodes whether an event occurred (∆Nsr(ti) = 1) or not (∆Nsr(ti) = 0), distinguishing events from non-events. The column log_exposure contains the logarithm of the interarrival time, log(ti − ti−1), an… view at source ↗
Figure 5
Figure 5. Figure 5: Top. Snapshot of relational event data structured for fitting a Cox regression with one covariate x. For each event, ti is recorded in column stop and the preceding event time ti−1 in start. If the covariate x remains constant within the interval, no additional adjustment is needed. However, if x changes within the interval, these changes must be tracked by inserting additional records that reflect the upd… view at source ↗
Figure 6
Figure 6. Figure 6: Top. Snapshot of relational event data formatted for fitting a conditional logistic regression with one covariate x. For each event, ti is recorded in column stop and the preceding event ti−1 in start. As in [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Top. Snapshot of case-control relational event data formatted for fitting a logistic regression with one covariate x. For each observed event, the corresponding time ti as well as the associated sender_ev and receiver_ev are recorded. sender_nv and receiver_nv are also stored for related non-events. For both events and non-events, the value of the covariate x is evaluated in x_ev and x_non_ev and their dif… view at source ↗
Figure 8
Figure 8. Figure 8: Top: Snapshot of a shifted case-control relational event dataset, formatted for fitting a logistic regression with a single covariate, weekday, which indicates whether the day is Monday to Friday. For each observed event, the corresponding event time (time_ev) and associated sender and receiver (sender_ev, receiver_ev) are recorded. The same structure is used for non-events. The time of the non-event (time… view at source ↗
Figure 9
Figure 9. Figure 9: Top. Snapshot of case-control relational event data formatted for fitting a logistic regression with random effects for sender activity and receiver popularity. For each observed event, the corresponding time ti as well as the associated sender_ev and receiver_ev are recorded. sender_nv and receiver_nv are also stored for related non-events. Two additional columns, id_ev and id_non_ev, are added for weight… view at source ↗
Figure 10
Figure 10. Figure 10: Computation of corrected conditional AIC [Wood et al., 2016] in [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Synthetic shifted case-control dataset: interstate academic collaborations. The first three columns report [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Left. The estimated coefficients for fixed-effect covariates are unbiased both by fitting the model using unshifted and shifted non-events. Right. Estimated curve for the effect of distance appropriately represents the true underlying trend of the contribution of this covariate to the log-hazard. The estimated curve differs from the true one by a constant. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Summary of the relational event model fitted to the WTC radio communication data using [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Estimated time-varying and non-linear effects. The time-varying effect, estimated via a spline function of [PITH_FULL_IMAGE:figures/full_fig_p027_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Estimated non-linear effects of Reciprocity and Time of the Day. The reciprocity effect exhibits a general downward trend, with occasional peaks that may correspond to delayed replies to emails needing more attention. The time-of-day effect indicates that email activity varies over the day, showing a strong increase from 8 AM until 3 PM. Certainly, REMs are not without challenges. Chief among them is the … view at source ↗
Figure 16
Figure 16. Figure 16: Fitted effects without shift (Modelling a unimodal network of emails in a company). [PITH_FULL_IMAGE:figures/full_fig_p037_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Fitted effects with shift (Modelling a unimodal network of emails in a company). [PITH_FULL_IMAGE:figures/full_fig_p038_17.png] view at source ↗
read the original abstract

Interactions and time shape many aspects of life. Everyday activities -- like conversations, emails, money transfers, citations, and even acts of violence -- are relational events: interactions between a sender and a receiver at a specific moment. At the intersection of event-history analysis and network modelling, relational event models (REMs) offer a powerful framework for studying when and why these events occur. Recent advances have made it possible to express REMs as generalized additive models, allowing researchers to capture complex, non-linear patterns over time. While an essay and a comprehensive review exist, a hands-on tutorial paper on REMs is still missing. This work fills that gap. It provides a practical introduction to REMs, incorporating the latest developments in the field. It demonstrates how to simulate synthetic relational-event data and walks through several empirical applications, comparing different modelling and inference strategies. By bringing together theory, simulation, and application, this tutorial lowers the barrier to entry and makes REMs a more accessible and practical tool.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The manuscript is a hands-on tutorial introducing relational event models (REMs) for analyzing time-stamped interactions between senders and receivers. It covers theoretical foundations at the intersection of event-history analysis and network modeling, demonstrates simulation of synthetic relational-event data, and walks through empirical applications that compare different modeling and inference strategies, including recent formulations of REMs as generalized additive models. The central claim is that integrating theory, simulation, and applications fills a gap and lowers the barrier to entry for using REMs.

Significance. If the tutorial's explanations and examples are accurate and reflect current best practices, the work would provide a valuable practical resource for researchers in statistics, sociology, and related fields studying dynamic relational data. The explicit comparison of modeling strategies is a strength that can help users make informed choices, and the focus on simulation offers a concrete way to build intuition before applying the methods to real data.

minor comments (2)
  1. The abstract states that 'an essay and a comprehensive review exist' but does not cite them; adding these references would strengthen the novelty claim and help readers locate prior work.
  2. To support the tutorial's goal of lowering the barrier to entry, the manuscript should explicitly indicate the availability of code and data for the simulation and empirical examples (e.g., via a GitHub repository or supplementary materials).

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their thorough and positive review of our manuscript. We are pleased that the tutorial is viewed as a valuable practical resource that integrates theory, simulation, and empirical applications to lower the barrier for using relational event models. The recommendation to accept is appreciated.

Circularity Check

0 steps flagged

No circularity: purely instructional tutorial with no derivations or predictions

full rationale

The paper is a hands-on tutorial introducing relational event models (REMs), covering theory, synthetic data simulation, and empirical applications. It makes no claims of deriving new predictions, first-principles results, or fitted quantities from within its own content. The central contribution is filling a gap for accessible instructional material by integrating existing concepts, without any self-definitional loops, fitted inputs renamed as predictions, or load-bearing self-citations that reduce the argument to its own inputs. No equations or modeling steps are presented as novel derivations that could exhibit circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

This is an introductory tutorial paper on an established statistical framework. It relies on standard assumptions from event-history analysis and network modeling plus the recent domain claim that REMs can be cast as generalized additive models; no new free parameters, axioms, or entities are introduced by the authors.

axioms (1)
  • domain assumption Relational event models can be expressed as generalized additive models to capture complex non-linear patterns over time
    Stated in the abstract as a recent advance that the tutorial incorporates.

pith-pipeline@v0.9.0 · 5461 in / 1270 out tokens · 64437 ms · 2026-05-10T18:10:06.028032+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    Frank Harary

    doi:10.1038/s41598-025-94969-0. Frank Harary. On the notion of balance of a signed graph.Michigan Mathematical Journal, 2(2):143–146, 1953. doi:10.1307/mmj/1028989917. Trevor Hastie and Robert Tibshirani.Generalized Additive Models. Chapman and Hall, 1990. Trevor Hastie, Robert Tibshirani, Jerome H Friedman, and Jerome H Friedman.The elements of statistic...

    work page doi:10.1038/s41598-025-94969-0 1953

  2. [2]

    2018 , journal =

    doi:10.32614/RJ-2018-009. Edzer Pebesma and Roger Bivand.Spatial Data Science: With applications in R. Chapman and Hall/CRC, 2023. doi:10.1201/9780429459016. URLhttps://r-spatial.org/book/. Patrick O. Perry and Patrick J. Wolfe. Point process modelling for directed interaction networks.Journal of the Royal Statistical Society: Series B (Statistical Method...

    work page doi:10.32614/rj-2018-009 2018

  3. [3]

    Marc Schneble.New Approaches in Statistical Modeling

    doi:10.1007/s10531-018-1535-9. Marc Schneble.New Approaches in Statistical Modeling. PhD thesis, Ludwig-Maximilians-Universität München, 2021. Hanno Seebens, Tim M Blackburn, Ellie E Dyer, Piero Genovesi, Philip E Hulme, Jonathan M Jeschke, Shyama Pagad, Petr Pyšek, Marten Winter, Margarita Arianoutsou, et al. No saturation in the accumulation of alien sp...

    work page doi:10.1007/s10531-018-1535-9 2021