Pith. sign in

REVIEW 3 major objections 6 minor 33 references

Inference of large scale relational state processes

T0 review · 3 major / 6 minor · reviewed 2026-07-13 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Network ties can evolve through multiple states, not just on or off, and the transition rates can be estimated efficiently from either full event histories or ordinary panel snapshots.

desk verdict Clean multi-state extension of relational-event models with a practical binary-panel shortcut that matches SAOM at far lower cost; the multi-state panel case is left open and the inertia approximation is the main soft spot. read the letter →

arxiv 2607.09363 v1 pith:QI3TI54M submitted 2026-07-10 stat.ME

Rūta Juozaitienė , Ernst C. Wit This is my paper
classification stat.ME
keywords multi-statetiesdynamicnetworksrelationalstatesstate-dependentcovariatesanchor-and-pullnestedcase-controlpaneldatalikelihood
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Most dynamic network models treat every relationship as simply present or absent. Real ties, however, often move through graded or typed states such as acquaintance, friendship and close friendship. This paper builds a continuous-time model in which each edge jumps among a finite set of states; the rate of each jump is allowed to depend on covariates that look at both the edge’s current state (anchoring forces) and its target state (pulling forces). From complete event histories the rates are recovered by a Cox partial likelihood that uses nested case-control sampling to stay computationally light; from panel snapshots a simple logistic approximation recovers the same qualitative conclusions that stochastic actor-oriented models produce, yet finishes in fractions of a second rather than hours. Simulations show accurate recovery of known linear and smooth effects; an application to adolescent friendship data reproduces established substantive findings while remaining scalable to larger networks and richer state spaces.

What carries the argument

The state-dependent intensity that factors each covariate into an anchoring term evaluated at the current state and a pulling term evaluated at the target state (the “anchor-and-pull” linear predictor), together with the nested case-control partial likelihood for full histories and the logistic inertia approximation for binary panel data.

What would settle it

Simulate or observe a binary panel network in which change rates vary strongly across dyads (for example by node degree or community membership); if the recovered effect contrasts then deviate systematically from the known true values while the full-history estimator remains accurate, the panel approximation fails.

Watch

Extended reading notes

Core claim

A continuous-time multi-state relational process whose transition intensities are driven by state-dependent covariates that decompose into anchoring (current-state) and pulling (target-state) mechanisms can be estimated from full event histories by nested case-control Cox partial likelihood and, for binary panels, by a logistic approximation that recovers the same substantive conclusions as established actor-oriented models at far lower computational cost.

Load-bearing premise

For panel data the model assumes that the probability any given tie changes inside an observation wave is essentially the same for every pair, so a single inertia shift can be absorbed into the logistic intercept.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper develops a continuous-time multi-state relational process in which each edge transitions among a finite set of states with intensities driven by state-dependent covariates, optionally decomposed into anchoring (current-state) and pulling (target-state) effects, including smooth terms. For full event histories it uses a Cox-type partial likelihood with nested case–control sampling (linear and GAM forms). For panel data it writes a CTMC/ODE likelihood and, for the binary case, approximates the two-state transition probabilities by a logistic regression with wave-level inertia shifts, recovering only effect contrasts unless SAOM-style symmetry is imposed. Simulations (20 replications, 20 nodes, 10k events) recover linear and designed nonlinear effects under full history and contrasts under panels; a re-analysis of the Teenage Friends and Lifestyle Study matches published SAOM signs/significance at much lower runtime (0.26 s).

Significance. If the estimators hold under the stated sampling regimes, the paper supplies a practical multi-state extension of relational-event thinking that preserves classical network-effect interpretability (reciprocity, degree, triadic closure) via the anchor-and-pull device, and a binary-panel procedure that reproduces SAOM substantive conclusions at logistic-regression cost. The full-history nested case–control route is a clear, scalable contribution. The binary-panel logistic approximation is the main applied payoff for the common snapshot design. Strengths include explicit simulation recovery of both linear coefficients and designed nonlinear receiving-balance curves, and a transparent side-by-side comparison with an independent SAOM implementation on a public dataset. The multi-state panel case is left open, which limits the breadth of the claim relative to the abstract and title.

major comments (3)
  1. §3.2.1 (after Eq. (4)): the binary-panel likelihood replaces the dyad-specific inertia factor 1−exp(−(λ01_sr+λ10_sr)Δtk) by a single wave-level constant exp(ρk) absorbed into a logistic intercept shift ρ*_k, justified only by the claim that inertia is “approximately constant across all pairs (s,r) within an interval.” When endogenous covariates (out-degree, reciprocity, receiving balance) make total exit rates highly heterogeneous, this injects systematic bias into the identifiable contrasts Δβ. The panel simulations already report “small biases … attributable to the inertia approximation,” yet there is no sensitivity check that varies covariate-driven rate heterogeneity, interval length, or event volume relative to Δt. Because the central practical claim for the common panel regime rests on this step, a targeted stress test (or a dyad-specific inertia alternative) is needed before the a
  2. Abstract and §3.2 vs §6: the abstract advertises a “general ODE formulation for the likelihood, which leads to a particularly efficient inference procedure for binary state model” and claims the framework “scales … under both full-history and panel sampling designs” for multi-state ties. In the text, multi-state panel estimation is abandoned after noting matrix-exponential cost (§3.2), and the conclusions correctly call general multi-state panel “an open question.” The multi-state contribution is therefore fully developed only for complete histories; the panel contribution is binary-only. The abstract and title (“large scale relational state processes”) should be aligned with that scope, or a concrete multi-state panel approximation (even limited) should be supplied.
  3. §4 simulation design: both full-history and panel studies use n=20 nodes and 20 replications, with the panel DGP generating 10k changes over five equal intervals (high event density relative to temporal resolution). This is adequate for a proof-of-concept recovery check under the model class, but it does not probe the “large scale” claim, nor does it stress the inertia constancy assumption under sparse panels or strong degree/reciprocity heterogeneity. At minimum, report bias/coverage as a function of interval length and rate heterogeneity, and consider larger n for the nested case–control path.
minor comments (6)
  1. Throughout: several typos and slips (e.g., “muti-state,” “instanteneous,” “indictor,” “wheter,” “two-sate,” “care” for “case,” “Rbsr” notation in figures). A careful copy-edit is needed.
  2. §2.1–2.2: the anchor-and-pull linear predictor (Eq. 3) is clear; it would help to state explicitly which covariates enter only as target-state, only as current-state, or both, and how this interacts with SAOM-type sign restrictions when both are imposed.
  3. Table 1: report standard errors or confidence intervals alongside p-values for both methods; the inertia/rate correspondence is useful but could be stated more formally (e.g., how ρ relates to the SAOM rate under the approximation).
  4. §3.1 nested case–control: the single-control logistic form is standard; briefly note whether multiple controls or stratified sampling were considered and how baseline hazards that differ by (i,j) are handled when the common-baseline simplification is dropped.
  5. Figures 1–2 are schematic and helpful; Figure 3b axis labels and the “Estimated/True” legend could be made more readable in print.
  6. References and related work: a short comparison to continuous-time multi-state / competing-risk network models and to existing multiplex dynamic models would situate the contribution more sharply for a methods audience.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: estimators derive from standard counting-process/CTMC likelihoods; simulations recover ground truth generated from the model class; empirical comparison uses external public data and independent SAOM.

full rationale

The paper's derivation chain is self-contained and non-circular. Intensities (eq. 2) are the standard exponential hazard of a multivariate counting process (Doob–Meyer); the full-history estimator is the ordinary Cox partial likelihood reduced by nested case-control sampling to logistic regression (standard, not self-referential). The panel binary likelihood begins from the exact two-state CTMC transition probabilities (matrix exponential of Q), then introduces an explicit approximation that the dyad-specific inertia factor is roughly constant within each wave and can be absorbed into a logistic intercept shift; this is an approximation whose validity is later checked by simulation (small residual bias acknowledged), not a definitional identity or a fitted quantity re-labelled as a prediction. Anchor-and-pull restrictions and state-dependent covariates are modelling choices, not uniqueness theorems imported from prior self-work. Simulations generate data from the same intensity class and recover known parameters/contrasts (ordinary recovery, not circular reuse). The Teenage Friends application compares estimates side-by-side with an independent SAOM implementation on a public dataset; agreement of substantive conclusions is an external check, not a self-fulfilling fit. Self-citations (e.g., prior REM papers by the authors) appear only as related-work background and are not load-bearing for any uniqueness claim or for the central likelihood derivations. No step reduces a claimed prediction or first-principles result to its own inputs by construction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The central claims rest on standard continuous-time Markov assumptions plus a handful of modelling choices (anchor-and-pull linear predictors, piecewise-constant rates inside panels, constant inertia across dyads) that are introduced without external validation. No new physical entities are postulated; free parameters are the usual regression coefficients and smooth functions estimated from data.

free parameters (3)
  • β^{+}/β^{-} (or f^{+}/f^{-}) for each covariate and transition
    Linear or smooth coefficients of the state-dependent covariates; estimated by partial likelihood or logistic regression and therefore free parameters of the fitted model.
  • baseline hazards λ0^(i,j)(t) or their contrasts Δk
    Non-parametric or discrete baseline rates (or their log-ratio) that absorb overall intensity level; estimated jointly with the regression effects.
  • inertia parameters ρk (or ρ*)
    Wave-specific constants that approximate the probability of any transition inside an observation interval; fitted as logistic intercepts.
assumptions (4)
  • domain assumption Each dyad’s state process is a continuous-time Markov chain with finite state space I and no simultaneous jumps across dyads.
    Stated in §2; required for the intensity representation and the Doob–Meyer decomposition.
  • domain assumption Transition intensities take the exponential (Cox) form λ = Y λ0 exp(f(x)).
    Eq. (2); standard proportional-hazards assumption imported from survival analysis.
  • ad hoc to paper Inside each panel interval the intensities may be treated as piecewise constant, and the inertia factor is approximately constant across all dyads.
    §3.2.1; enables the logistic approximation but is not derived from first principles.
  • standard math Nested case-control sampling of a single non-event yields a consistent estimator of the partial-likelihood score.
    Cited from Borgan et al. (1995); used without re-proof for the multi-state risk sets.
invented entities (1)
  • anchor-and-pull decomposition of state-dependent covariates
    purpose: Reduces the number of free parameters while giving a substantive interpretation of forces that keep a tie in its current state versus forces that attract it to a target state.
    Introduced in §2.2 as a modelling restriction; no independent empirical validation outside the paper’s own simulations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inference of large scale relational state processes." pith.science (2026). https://pith.science/paper/QI3TI54M

@misc{pith2026260709363,
  author       = {Pith},
  title        = {Pith review of: Inference of large scale relational state processes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QI3TI54M}},
  note         = {Machine review of arXiv:2607.09363}
}
read the original abstract

Relational states refer to concepts such as friendship or collaboration, in which a relationship persists over a certain amount of time. Study of relational states often involves figuring out what factors contribute to the creation or dissolution of these relationships. However, most methods available now restrict their attention to binary states, i.e., ties that are either present or absent, even though many real-world systems evolve through multiple relational states (e.g., acquaintance, friendship, close friendship). We propose a continuous-time framework for modelling and inferring relational state networks in which each edge evolves by transitioning between two or more states. In our model, transition intensities are driven by state-dependent covariates that might be decomposed into anchoring (current-state) and pulling (target-state) mechanisms, with both linear and smooth non-linear effects. We address two common sampling regimes. With full event histories, a Cox-type partial likelihood with nested case-control sampling enables efficient estimation of both parametric and smooth effects. Instead, for panel data we derive a general ODE formulation for the likelihood, which leads to a particularly efficient inference procedure for binary state model. Simulation studies confirm accurate recovery of model parameters, and an empirical application to adolescent friendship data reproduces the substantive conclusions of established modelling techniques while offering substantial computational gains. The framework preserves the interpretability of classical network effects, generalizes them to multi-state ties, and scales to larger, more complex designs under both full-history and panel sampling designs.

Figures

Figures reproduced from arXiv: 2607.09363 by the authors.

Figure 1
Figure 1. Illustration of state-dependent reciprocity in a multi-state network. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Illustration of state-dependent transitivity in a multi-state network. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Estimates in the full-history setting. (a) Boxplots of the estimated [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Estimates in the panel data setting. Boxplots of the estimated con [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 1 linked inside Pith

  1. [1]

    The new challenges of multiplex networks: Measures and models.The European Physical Jour- nal Special Topics, 226(3):401–416, 2017

    Federico Battiston, Vincenzo Nicosia, and Vito Latora. The new challenges of multiplex networks: Measures and models.The European Physical Jour- nal Special Topics, 226(3):401–416, 2017

    2017

  2. [2]

    Multi- plex network ties and the spatial diffusion of radical innovations: Martin luther’s leadership in the early reformation.American Sociological Review, 85(5):857–894, 2020

    Sascha O Becker, Yuan Hsiao, Steven Pfaff, and Jared Rubin. Multi- plex network ties and the spatial diffusion of radical innovations: Martin luther’s leadership in the early reformation.American Sociological Review, 85(5):857–894, 2020

    2020

  3. [3]

    Relational event modeling.Annual Review of Statistics and Its Ap- plication, 11, 2024

    Federica Bianchi, Edoardo Filippi-Mazzola, Alessandro Lomi, and Ernst C Wit. Relational event modeling.Annual Review of Statistics and Its Ap- plication, 11, 2024

    2024

  4. [4]

    Methods for the analysis of sampled cohort data in the cox proportional hazards model.The Annals of Statistics, pages 1749–1778, 1995

    Ø Borgan, Larry Goldstein, and Bryan Langholz. Methods for the analysis of sampled cohort data in the cox proportional hazards model.The Annals of Statistics, pages 1749–1778, 1995

    1995

  5. [5]

    Mixed additive mod- elling of global alien species co-invasions of plants and insects.Journal of the Royal Statistical Society Series C: Applied Statistics, page qlaf034, 2025

    Martina Boschi, R¯ uta Juozaitien˙ e, and Ernst C Wit. Mixed additive mod- elling of global alien species co-invasions of plants and insects.Journal of the Royal Statistical Society Series C: Applied Statistics, page qlaf034, 2025

    2025

  6. [6]

    Introduction to relational event mod- elling.arXiv preprint arXiv:2604.07063, 2026

    Martina Boschi and Ernst C Wit. Introduction to relational event mod- elling.arXiv preprint arXiv:2604.07063, 2026

    arXiv 2026

  7. [7]

    H Bush, Patrick West, and Lynn Michell. The role of friendship groups in the uptake and maintenance of smoking amongst pre-adolescent and ado- lescent children: Distribution of frequencies.Glasgow, Scotland: Medical Research Council, 1997

    1997

  8. [8]

    Demand and congestion in multiplex transportation networks.PloS one, 11(9):e0161738, 2016

    Philip S Chodrow, Zeyad Al-Awwad, Shan Jiang, and Marta C Gonz´ alez. Demand and congestion in multiplex transportation networks.PloS one, 11(9):e0161738, 2016. 22

    2016

Show all 33 references
  1. [9]

    Regression models and life-tables.Journal of the Royal Statistical Society: Series B (Methodological), 34(2):187–202, 1972

    David R Cox. Regression models and life-tables.Journal of the Royal Statistical Society: Series B (Methodological), 34(2):187–202, 1972

    1972

  2. [10]

    Multilayer modeling and analysis of human brain networks.Giga Science, 6(5):gix004, 2017

    Manlio De Domenico. Multilayer modeling and analysis of human brain networks.Giga Science, 6(5):gix004, 2017

    2017

  3. [11]

    Social cohesion.Annu

    Noah E Friedkin. Social cohesion.Annu. Rev. Sociol., 30(1):409–425, 2004

    2004

  4. [12]

    Industrial clusters and the evolution of their knowledge networks: revisiting a chilean case.Brighton, UK: SPRU working paper series, 2008

    Elisa Giuliani and Martin Bell. Industrial clusters and the evolution of their knowledge networks: revisiting a chilean case.Brighton, UK: SPRU working paper series, 2008

    2008

  5. [13]

    Seung Hyun Han, Seung Won Yoon, and Chugnil Chae. Building social cap- ital and learning relationships through knowledge sharing: a social network approach of management students’ cases.Journal of Knowledge Manage- ment, 24(4):921–939, 2020

    2020

  6. [14]

    Discrete temporal models of social networks.The Electronic Journal of Statistics, 4:585–605, 2010

    Steve Hanneke, Wenjie Fu, and Eric P Xing. Discrete temporal models of social networks.The Electronic Journal of Statistics, 4:585–605, 2010

    2010

  7. [15]

    It’s about time: revisiting reciprocity and triadicity in relational event analysis.Journal of the Royal Statistical Society Series A: Statistics in Society, page qnae132, 2024

    R¯ uta Juozaitien˙ e and Ernst C Wit. It’s about time: revisiting reciprocity and triadicity in relational event analysis.Journal of the Royal Statistical Society Series A: Statistics in Society, page qnae132, 2024

    2024

  8. [16]

    Multilayer networks.Journal of complex networks, 2(3):203–271, 2014

    Mikko Kivel¨ a, Alex Arenas, Marc Barthelemy, James P Gleeson, Yamir Moreno, and Mason A Porter. Multilayer networks.Journal of complex networks, 2(3):203–271, 2014

    2014

  9. [17]

    Temporal ex- ponential random graph models with btergm: Estimation and bootstrap confidence intervals.Journal of Statistical Software, 83:1–36, 2018

    Philip Leifeld, Skyler J Cranmer, and Bruce A Desmarais. Temporal ex- ponential random graph models with btergm: Estimation and bootstrap confidence intervals.Journal of Statistical Software, 83:1–36, 2018

    2018

  10. [18]

    Reliability of relational event model estimates under sampling: How to fit a relational event model to 360 million dyadic events.Network Science, 8(1):97–135, 2020

    J¨ urgen Lerner and Alessandro Lomi. Reliability of relational event model estimates under sampling: How to fit a relational event model to 360 million dyadic events.Network Science, 8(1):97–135, 2020

    2020

  11. [19]

    Peer pressure to smoke: the meaning depends on the method.Health education research, 11(1):39–49, 1996

    Lynn Michell and Patrick West. Peer pressure to smoke: the meaning depends on the method.Health education research, 11(1):39–49, 1996

    1996

  12. [20]

    Homophily and assimilation among sport-active adolescent substance users.Connections, 27(1):47–63, 2006

    Michael Pearson, Christian Steglich, and Tom Snijders. Homophily and assimilation among sport-active adolescent substance users.Connections, 27(1):47–63, 2006

    2006

  13. [21]

    Anomalous: A joint modeling approach for anomaly detection on attributed networks

    Zhen Peng, Minnan Luo, Jundong Li, Huan Liu, Qinghua Zheng, et al. Anomalous: A joint modeling approach for anomaly detection on attributed networks. InIJCAI, volume 18, pages 3513–3519, 2018

    2018

  14. [22]

    Closure, connectivity and degree distributions: Exponential random graph (p*) models for directed social networks.Social networks, 31(2):105–117, 2009

    Garry Robins, Pip Pattison, and Peng Wang. Closure, connectivity and degree distributions: Exponential random graph (p*) models for directed social networks.Social networks, 31(2):105–117, 2009. 23

    2009

  15. [23]

    The statistical evaluation of social network dynamics

    Tom AB Snijders. The statistical evaluation of social network dynamics. Sociological methodology, 31(1):361–395, 2001

    2001

  16. [24]

    Stochastic actor-oriented models for network dynamics

    Tom AB Snijders. Stochastic actor-oriented models for network dynamics. Annual review of statistics and its application, 4(1):343–363, 2017

    2017

  17. [25]

    Introduction to stochastic actor-based models for network dynamics.Social networks, 32(1):44–60, 2010

    Tom AB Snijders, Gerhard G Van de Bunt, and Christian EG Steglich. Introduction to stochastic actor-based models for network dynamics.Social networks, 32(1):44–60, 2010

    2010

  18. [26]

    Applying siena

    Christian Steglich, Tom AB Snijders, and Patrick West. Applying siena. Methodology, 2(1):48–56, 2006

    2006

  19. [27]

    Evolution of prosocial behaviours in multilayer populations.Nature Human Behaviour, 6(3):338–348, 2022

    Qi Su, Alex McAvoy, Yoichiro Mori, and Joshua B Plotkin. Evolution of prosocial behaviours in multilayer populations.Nature Human Behaviour, 6(3):338–348, 2022

    2022

  20. [28]

    Evolution of cooperation with contextualized behavior.Science advances, 8(6):eabm6066, 2022

    Qi Su, Alex McAvoy, and Joshua B Plotkin. Evolution of cooperation with contextualized behavior.Science advances, 8(6):eabm6066, 2022

    2022

  21. [29]

    Interplay of competi- tion and cooperation in wind farm interorganizational projects: Relational approach.Journal of Management in Engineering, 36(1):04019034, 2020

    Jorge Renato Verschoore and Vivian Sebben Adami. Interplay of competi- tion and cooperation in wind farm interorganizational projects: Relational approach.Journal of Management in Engineering, 36(1):04019034, 2020

    2020

  22. [30]

    Rela- tional event models for longitudinal network data with an application to interhospital patient transfers.Statistics in medicine, 36(14):2265–2287, 2017

    Duy Vu, Alessandro Lomi, Daniele Mascia, and Francesca Pallotti. Rela- tional event models for longitudinal network data with an application to interhospital patient transfers.Statistics in medicine, 36(14):2265–2287, 2017

    2017

  23. [31]

    Relational event models for social learning in moocs.Social Networks, 43:121–135, 2015

    Duy Vu, Philippa Pattison, and Garry Robins. Relational event models for social learning in moocs.Social Networks, 43:121–135, 2015

    2015

  24. [32]

    Evolutionary games on multilayer networks: a colloquium.The European physical journal B, 88(5):124, 2015

    Zhen Wang, Lin Wang, Attila Szolnoki, and Matjaˇ z Perc. Evolutionary games on multilayer networks: a colloquium.The European physical journal B, 88(5):124, 2015

    2015

  25. [33]

    chap- man and hall/CRC, 2017

    Simon N Wood.Generalized additive models: an introduction with R. chap- man and hall/CRC, 2017. 24

    2017

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.