pith. sign in

arxiv: 2604.07488 · v1 · submitted 2026-04-08 · 💰 econ.EM

Identification in Dynamic Dyadic Network Formation Models with Fixed Effects

Wayne Yuan Gao , Yi Niu This is my paper

Pith reviewed 2026-05-10 17:19 UTC · model grok-4.3

classification 💰 econ.EM
keywords dynamic network formationdyadic modelsfixed effectsidentificationconditional logithomophilytransitivitysubgraph statistics
0
0 comments X

The pith

Dynamic dyadic network models identify link formation parameters by integrating out pair fixed effects through inequalities and algebraic signed-subgraph comparisons.

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

The paper develops identification results for a dynamic model in which pairs of agents form links over time based on observed covariates, lagged local network statistics such as common friends or indirect connections, and unobserved time-invariant heterogeneity specific to each pair. It shows that treating each dyad as its own short panel yields inequalities that average out the fixed effects, while signed comparisons of network subgraphs across periods difference them out exactly. These semiparametric restrictions can be tightened to point identification when errors are serially independent with known distribution or when the pair fixed effect is the sum of two individual effects; under logit shocks satisfying both, the model reduces to an exact conditional logit form with explicit sufficient conditions for unique recovery of the parameters.

Core claim

In the dynamic index model of dyadic link formation that includes observed-covariate homophily, transitivity through common friends, second-order indirect-friend effects, and general local subgraph statistics, the combination of dyad-panel inequalities and signed-subgraph intertemporal comparisons delivers set identification of the parameters. When the error distribution is serially independent and known, or when pairwise fixed effects are additive individual effects, the identifying restrictions sharpen; under i.i.d. logit shocks together with additive fixed effects the representation becomes an exact conditional logit, and sufficient conditions then guarantee point identification.

What carries the argument

Signed-subgraph comparisons that algebraically eliminate dyad fixed effects through intertemporal variation within each pair, combined with inequalities obtained by treating each dyad as a short panel.

If this is right

  • The framework nests standard features of network formation such as homophily on observed traits and transitivity in a single dynamic index specification.
  • Point identification holds under the exact conditional logit representation when both serial independence and additive fixed effects are assumed together with i.i.d. logit shocks.
  • Researchers can recover the effects of lagged local statistics without parametric assumptions on the distribution of time-invariant pair heterogeneity.
  • The two complementary handling methods for fixed effects can be used separately or jointly depending on the available variation in the data.

Where Pith is reading between the lines

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

  • The approach could be applied to panel data on directed networks by redefining signed subgraphs to respect directionality.
  • If the local subgraph statistics are chosen to approximate global network properties, the identification strategy might extend to models with longer-range dependence.
  • In applied settings the conditional logit form would permit standard fixed-effects logit software to be used after suitable data transformation.
  • Testing the additive-fixed-effects restriction against a general pairwise effect could be done by comparing set-identified bounds with the point estimate obtained under additivity.

Load-bearing premise

The unobserved error terms are serially independent with a known distribution, or the pairwise fixed effect equals the sum of two individual fixed effects, or both hold.

What would settle it

Simulated data generated from the model with serially correlated errors but otherwise satisfying the other maintained conditions would produce different point estimates when the serial-independence restriction is imposed versus when it is relaxed to set identification only.

read the original abstract

This paper establishes (set) identification results in a dynamic dyadic network formation model with time-varying observed covariates, lagged local network statistics, and unobserved heterogeneity in the form of fixed effects. Our framework accommodates observed-covariate homophily, transitivity through common friends, second-order or indirect-friend effects, and more general local subgraph statistics within a single dynamic index model. The analysis combines two complementary ways of handling fixed effects: inequalities that integrate out time-invariant dyad heterogeneity by treating each dyad as a short panel, and signed-subgraph comparisons that difference out fixed effects algebraically through intertemporal variation within each dyad. We show that the semiparametric identifying restrictions can be sharpened using either or both of the following assumptions: (i) error distribution is serially independent with a known distribution, (ii) pairwise fixed effect takes the form of additive individual fixed effects. Combining (i) and (ii) under i.i.d. logit shocks, we obtain an exact conditional logit representation and provide sufficient conditions for point identification.

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

2 major / 2 minor

Summary. The paper establishes set identification results in a dynamic dyadic network formation model with time-varying covariates, lagged local network statistics (including transitivity and indirect effects), and unobserved dyad fixed effects. It combines two strategies for handling fixed effects: inequalities that treat each dyad as a short panel to integrate out time-invariant heterogeneity, and signed-subgraph comparisons that algebraically difference out fixed effects using intertemporal variation. Under the additional assumptions of serially independent errors with known distribution and additive individual fixed effects, the model yields an exact conditional logit representation, with sufficient conditions provided for point identification.

Significance. If the derivations hold, the results advance the literature on semiparametric identification in network models by accommodating dynamics and heterogeneity in a unified index framework. The dual complementary approaches to fixed effects (panel inequalities and algebraic differencing) provide robustness, and the reduction to conditional logit under standard logit shocks is a clear strength, enabling potential empirical application to settings with homophily and transitivity.

major comments (2)
  1. [Identification arguments (around the signed-subgraph comparisons)] The abstract states that signed-subgraph comparisons difference out fixed effects algebraically through intertemporal variation, but the interaction with lagged network statistics (which are themselves dynamic) requires explicit verification that the signed comparisons remain valid without residual dependence on the fixed effects; this is load-bearing for the set-identification claim.
  2. [Point identification section (combining (i) and (ii))] Under assumptions (i) and (ii) the paper claims an exact conditional logit representation leading to point identification, but the sufficient conditions for point identification must be shown to be non-vacuous given the presence of lagged local subgraph statistics; if these conditions implicitly restrict the support of the covariates or network lags, the point-identification result would be narrower than stated.
minor comments (2)
  1. [Model setup] The abstract mentions 'more general local subgraph statistics' but does not list the precise functional forms allowed; the model section should explicitly define the class of admissible statistics to clarify the scope.
  2. [Notation and assumptions] Notation for the pairwise fixed effect and its additive decomposition should be introduced with a clear equation reference early in the paper to aid readability when comparing the two fixed-effect handling strategies.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for minor revision. We address the two major comments below by providing clarifications and committing to revisions that enhance the transparency of our identification arguments.

read point-by-point responses

  1. Referee: The abstract states that signed-subgraph comparisons difference out fixed effects algebraically through intertemporal variation, but the interaction with lagged network statistics (which are themselves dynamic) requires explicit verification that the signed comparisons remain valid without residual dependence on the fixed effects; this is load-bearing for the set-identification claim.

    Authors: We thank the referee for this insightful comment. In the manuscript, the signed-subgraph comparisons (Proposition 3) are constructed by selecting intertemporal pairs within each dyad that share identical lagged network statistics, which ensures the dynamic lags are held fixed while the fixed effects cancel algebraically. This conditioning is implicit in the definition of the signed comparisons and preserves the inequality direction without residual dependence on the fixed effects. To make the argument fully explicit, we will add a dedicated remark following the abstract and expand the proof sketch in Appendix B with an additional step verifying the cancellation under dynamic lags. revision: partial

  2. Referee: Under assumptions (i) and (ii) the paper claims an exact conditional logit representation leading to point identification, but the sufficient conditions for point identification must be shown to be non-vacuous given the presence of lagged local subgraph statistics; if these conditions implicitly restrict the support of the covariates or network lags, the point-identification result would be narrower than stated.

    Authors: We appreciate the referee raising this point. Theorem 4.2 states sufficient conditions that require only positive probability of observing changes in the time-varying covariates and lagged subgraph statistics within dyads (analogous to standard conditional logit with time-varying regressors). These conditions are non-vacuous and do not impose support restrictions that would narrow the result; they are satisfied whenever covariates have full support and network lags exhibit intertemporal variation, as is standard in dynamic network settings. We already include a simple example in Section 4.3, but we will add a short discussion paragraph clarifying that the conditions remain applicable with lagged statistics and do not restrict the model in a vacuous way. revision: yes

Circularity Check

0 steps flagged

No significant circularity; identification derived from stated assumptions

full rationale

The paper's core claims rest on semiparametric identification arguments that treat each dyad as a short panel and use signed-subgraph comparisons to difference out fixed effects, then sharpen to conditional logit under serial independence plus additive individual effects. These steps are explicitly conditional on the listed modeling restrictions (i.i.d. logit shocks, known error distribution, or additive FE form) rather than being fitted to data or defined in terms of the target result. No equations reduce by construction to inputs, no parameters are estimated and relabeled as predictions, and no load-bearing uniqueness theorems are imported via self-citation. The derivation is self-contained within the stated framework and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claims rest on three modeling assumptions about the error process and the structure of unobserved heterogeneity; no free parameters or new invented entities are introduced.

axioms (3)
  • domain assumption Error distribution is serially independent with a known distribution
    Used to sharpen the semiparametric identifying restrictions into tighter bounds or point identification.
  • domain assumption Pairwise fixed effect takes the form of additive individual fixed effects
    Alternative assumption that allows algebraic differencing to remove the fixed effect.
  • domain assumption i.i.d. logit shocks
    Combined with the above to obtain an exact conditional logit representation.

pith-pipeline@v0.9.0 · 5471 in / 1511 out tokens · 37191 ms · 2026-05-10T17:19:55.283661+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

  1. [1]

    Candelaria, L. E. (2017): A Semiparametric Network Formation Model with Multiple Linear Fixed Effects, Working paper, The University of Edinburgh

    work page 2017

  2. [2]

    (2020 a ): Econometric Models of Network Formation, Annual Review of Economics, 12, pp

    de Paula, A. (2020 a ): Econometric Models of Network Formation, Annual Review of Economics, 12, pp. 775--799

    work page 2020

  3. [3]

    --- -.1pt --- -.1pt --- (2020 b ): Strategic network formation, in The Econometric Analysis of Network Data, ed. by B. Graham and A. de Paula, Academic Press, 41--61

    work page 2020

  4. [4]

    Gao, W. Y. (2020): Nonparametric identification in index models of link formation, Journal of Econometrics, 215, 399--413

    work page 2020

  5. [5]

    Gao, W. Y., M. Li, and S. Xu (2023): Logical differencing in dyadic network formation models with nontransferable utilities, Journal of Econometrics, 235, 302--324

    work page 2023

  6. [6]

    Gao, W. Y., M. Li, and Z. Xu (2026): Tractable Identification of Strategic Network Formation Models with Unobserved Heterogeneity, arXiv preprint arXiv:2603.08634

    work page arXiv 2026

  7. [7]

    Gao, W. Y. and R. Wang (2026): Identification in nonlinear dynamic panel models under partial stationarity, Journal of Econometrics, 253, 106185

    work page 2026

  8. [8]

    Graham, B. S. (2016): Homophily and Transitivity in Dynamic Network Formation , Tech. rep., National Bureau of Economic Research

    work page 2016

  9. [9]

    --- -.1pt --- -.1pt --- (2017): An Econometric Model of Network Formation with Degree Heterogeneity , Econometrica, 85, 1033--1063

    work page 2017

  10. [10]

    --- -.1pt --- -.1pt --- (2020): Dyadic regression, in The Econometric Analysis of Network Data, ed. by B. Graham and A. de Paula, Academic Press, 23--40

    work page 2020

  11. [11]

    Honor \'e , B. E. and E. Kyriazidou (2000): Panel data discrete choice models with lagged dependent variables, Econometrica, 68, 839--874

    work page 2000

  12. [12]

    (2018): Semiparametric Analysis of Network Formation, Journal of Business & Economic Statistics, 36, 705--713

    Jochmans, K. (2018): Semiparametric Analysis of Network Formation, Journal of Business & Economic Statistics, 36, 705--713

    work page 2018

  13. [13]

    (2017): Semiparametric Estimation in Network Formation Models with Homophily and Degree Heterogeneity, Available at SSRN 2988698

    Toth, P. (2017): Semiparametric Estimation in Network Formation Models with Homophily and Degree Heterogeneity, Available at SSRN 2988698

    work page 2017