Pith. sign in

REVIEW 2 major objections 5 minor 38 references

Quantile Peer Effect Models

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A structural quantile model shows peer influence varies across the outcome distribution and changes key-player rankings.

desk verdict A genuinely new quantile peer effect model with a workable IV strategy; the main vulnerability is that identifying conformity requires isolated students to be exchangeable with non-isolated ones, an assumption the paper's own data handling undermines. read the letter →

arxiv 2506.12920 v1 pith:WNFNL27H submitted 2025-06-15 econ.EM

Aristide Houndetoungan This is my paper · ORCID
classification econ.EM
keywords peereffectsquantileregressionsocialnetworksstructuralestimationkeyplayersidentificationspilloverandconformityAddHealth
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

This paper introduces a structural model in which an individual's outcome depends on several quantiles of the peer outcome distribution, not just the average. It proves that such a game has a unique Nash equilibrium and that the parameters can be recovered with an instrumental-variable strategy using quantiles of peers' characteristics. Applied to adolescent outcomes, the model finds that peers with intermediate outcomes are often the most influential, a non-monotonic pattern that linear-in-means and CES models cannot represent. As a result, rankings of key players change: who matters depends on the outcome distribution, not only on network position. If correct, the model gives applied researchers a flexible but still linear-in-parameters tool for peer-effect estimation and policy targeting.

What carries the argument

The central object is the quantile peer effect: for each $\tau$ in a finite set $T$, $q_{\tau,i}(y_{-i})$ is the sample $\tau$-quantile of friends' outcomes, and the outcome equation is $y_i = c + \sum_{\tau} \lambda_\tau q_{\tau,i} + x_i'\beta + \varepsilon_i$. Because a sample quantile is a weighted average of two ranked peer outcomes, the model remains linear in parameters, and quantiles of peers' and peers-of-peers' exogenous characteristics serve as instruments. The microfoundation is a linear-quadratic utility with spillover and conformity terms at each quantile; its best-response function is a contraction, giving a unique Nash equilibrium. Identification of the conformity share $\lambda_2$ uses isolated students: their outcomes reveal $\beta_1$ directly, and comparing $\beta_1$ to the coefficient in the non-isolated equation identifies $1-\lambda_2$. An encompassing test selects the quantile levels in $T$.

What would settle it

A Hausman-type comparison of $\beta_1$ estimated from the isolated-only equation and the value implied by the non-isolated equation would settle it: a statistically significant gap breaks the identifying restriction, and with it $\lambda_2$ and every key-player ranking derived from it.

Watch

Extended reading notes

Core claim

The central discovery is that peer influence is heterogeneous across the distribution of peer outcomes and that this heterogeneity is identifiable and empirically consequential. In the proposed model, each quantile $\tau$ in a set $T$ has its own peer-effect parameter $\lambda_\tau$; total effects split into spillover ($\lambda_1$) and conformity ($\lambda_2$) components. The paper proves a unique Nash equilibrium exists whenever $\sum_{\tau \in T}|\lambda_\tau|<1$, using a contraction argument that handles the nondifferentiability of sample quantiles. It then shows that, with enough subnetworks containing at least two isolated students and under standard instrument conditions, $\lambda_\tau$, $\lambda_1$, $\lambda_2$, $\beta_1$, and $\beta_2$ are all identified. Empirically, the quantile model uncovers non-monotonic influence patterns for outcomes such as GPA, smoking, drinking, and extracurricular activities, and shows that key-player rankings from the LIM and CES models diverge substantially when peer effects are non-monotonic.

Load-bearing premise

The key assumption is that the effect of own observable characteristics on the latent type, $\beta_1$, is the same for students with no friends and students with friends; if isolated students differ systematically from non-isolated ones in unobserved ways tied to the outcome, the estimate of the conformity parameter $\lambda_2$ and all quantile effects built on it are biased.

Editorial extensions

If this is right

  • If the model is right, standard linear-in-means estimates can misstate both the size and the source of peer effects, because they average over quantiles with different, sometimes opposite, effects.
  • Key-player counterfactuals change: a student whose friends have extreme outcomes is less influential under the quantile model when extreme peers barely matter, so targeted interventions aimed at the 'wrong' nodes would lose effectiveness.
  • The framework yields a practical estimation path: OLS for isolated students plus IV/GMM for non-isolated students, with a formal encompassing test to choose the number of quantile levels.
  • Because the model nests the LIM model and min/max specifications, existing results can be checked for whether they depend on the equal-effects restriction.
  • With only three or four quantile levels, the model captures patterns that a saturated rank-dependent model would need dozens of parameters to represent.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial: If non-monotonic peer effects are a general feature, interventions that remove or reward specific peers should be designed with the outcome distribution in mind, not just network centrality—a consequence the paper states for key players but that extends to diffusion and norm-setting policies.
  • Editorial: The quantile specification suggests a direct testable extension: randomize the composition of peer groups along outcome levels and check whether changes in mid-level peers move behavior more than changes at the extremes, as the empirical pattern predicts.
  • Editorial: The identifying restriction that $\beta_1$ is equal for isolated and non-isolated students could be relaxed by allowing only a subset of covariates to share the coefficient; the paper notes this relaxation but does not develop a test for which components can be freed.
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

2 major / 5 minor

Summary. The paper proposes a structural model in which an agent's outcome depends on the sample quantiles of the peer outcome distribution, with the quantile set T chosen by the researcher. A linear-quadratic utility with spillover and conformity components provides microfoundations, and the paper proves existence and uniqueness of the Nash equilibrium under the contraction condition sum_{tau in T} |lambda_tau| < 1. Estimation is carried out with OLS on isolated agents and IV/GMM on non-isolated agents, using quantiles of peer characteristics as instruments. Identification of the conformity parameter lambda_2 comes from comparing the own-characteristic coefficient in the isolated equation with the corresponding coefficient in the non-isolated equation. The paper also develops an encompassing test for choosing T, tests for the validity of Type II instruments, Monte Carlo evidence, and an Add Health application with key-player counterfactuals.

Significance. If the identifying assumptions hold, this is a valuable and policy-relevant generalization of linear-in-means and CES peer-effect models. The model is linear in parameters, has a clear game-theoretic foundation, and comes with a complete estimation toolkit, including an encompassing test and an R package with replication code. The empirical finding that peer-effect profiles are often non-monotonic and that key-player rankings differ across specifications is important and potentially influential. The main caveat is that the structural identification of lambda_2, and hence of all quantile treatment parameters, rests on the exogeneity of isolation status and on the equality of beta_1 between isolated and non-isolated students; the manuscript states these assumptions but does not fully defend them against selection into isolation.

major comments (2)
  1. [Appendix B, Lemma B.1] The identification of lambda_2, and therefore of all theta_tau, rests on the OLS estimate of beta_1 from isolated agents being the same parameter that appears in the non-isolated equation (4.3). Assumption 4.1.B only imposes E(epsilon^iso_{s,i} | x^iso_{s,j}) = 0 and E(epsilon^niso_{s,i} | G_s, z^niso_{s,j}) = 0; it does not impose E(epsilon^iso_{s,i} | G_s, x^iso_{s,j}) = 0. Because isolation status is encoded in the rows of G_s, this leaves open the possibility that selection into isolation is correlated with the unobserved error, in which case the isolated OLS estimator is inconsistent for the beta_1 appearing in (4.3). The paper's own discussion of false isolates and missing friend identifiers in Section 6.1 makes this a concrete channel rather than a purely hypothetical one. Please either state the stronger condition E(epsilon^iso_{s,i} | G_s, x^iso_{s,j}) = 0, or provide an explicit selection model, and qualify Proposition 4.1 accordingly.
  2. [Appendix B, Lemma B.1] The proof of Lemma B.1, Case 2, asserts that 'the only way' for R_a(a_i) = R_tilde{a}(tilde{a}_j) to hold is the existence of an index s satisfying (B.1), but this assertion is not demonstrated. A counting argument is needed: because a_j belongs to the set of elements of a that are at least a_i, while tilde{a}_j does not belong to the set of elements of tilde{a} that are at least tilde{a}_j, equal ranks force the intersection {l : a_l <= a_i} cap {l : tilde{a}_l >= tilde{a}_j} to be nonempty. Without such a justification, the contraction proof of Proposition 3.1 is incomplete, although the lemma itself appears to be true.
minor comments (5)
  1. [Table 6.4] The header 'Isolated students only' for the first four columns is inconsistent with the surrounding text, which says the reduced-form quantile model is estimated on non-isolated students; isolated students have no peers and hence no quantile peer outcomes, so the header should read 'Non-isolated students only'.
  2. [Section 4.3] In the sentence 'If there are no isolated agents, it is still possible to identify lambda_tau for all tau in Equation (4.2)', the reference should be to Equation (4.3), since (4.2) is the isolated-agent equation with no peer terms.
  3. [Section 5] The Monte Carlo tables are numbered Table 6.1 and Table 6.2 even though they appear in Section 5, and the text in Section 5 refers to 'Table 6.1'; the tables should be renumbered consistently.
  4. [Section 4.3] The identification of lambda_2 from beta_tilde_1 = (1 - lambda_2) beta_1 requires at least one nonzero component of beta_1; Proposition 4.1 would be more precise if it stated this condition, since components of beta_1 that are zero are uninformative for lambda_2.
  5. [Online Appendix D.2] The statement in Lemma D.1 that 'the only way' omega_i != tilde{omega}_i can hold while pi_i = tilde{pi}_i is that the two vectors are sorted differently would also benefit from a counting argument, along the same lines as Lemma B.1.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the identification argument is a genuine cross-equation restriction, not a fit renamed as a prediction.

full rationale

The paper's central derivation chain is self-contained. Equilibrium existence and uniqueness are proven directly in Appendix B by showing that the best-response mapping is a contraction via Lemma B.1, with no fitted inputs and no appeal to self-citations. Identification of the reduced-form parameters (λτ, β1, β̃1) is standard IV/OLS from Equations (4.2) and (4.3), and λ2 is recovered from the algebraic cross-equation restriction β̃1 = (1−λ2)β1. This is a structural restriction linking two separately identified objects, not a fitted parameter renamed as a prediction. The assumption that β1 is identical for isolated and non-isolated students is explicitly stated in Section 4.4 ('This estimation strategy relies on the assumption that β1 is the same in both Equations (4.2) and (4.3)'); it is an identifying assumption, not a circular step, and the paper even notes it can be relaxed to a subset of components. The encompassing test in Section 4.5.1 is a specification test with a well-defined null δa,b = 0 and is not used to force the structural estimates; selecting the number of quantile levels is a specification search, not a prediction that reduces to fitted values. The Section 6.1 discussion of 'false isolates' is a data limitation that is acknowledged, and it does not make the identification argument circular. Self-citations (e.g., Houndetoungan and Maoude 2024; Houndetoungan et al. 2024) concern technical estimation refinements and are not load-bearing for the main identification result. Consequently, the paper does not exhibit self-definitional, fitted-input-as-prediction, or self-citation-circularity patterns.

Assumptions & free parameters 1 free parameters · 7 assumptions · 0 invented entities

The model's central identification leverages standard econometric assumptions (network exogeneity, isolated agents, common β1). The only researcher-chosen free parameter is the quantile level set T. No new particles or entities are introduced.

free parameters (1)
  • Quantile level set T = T = {0, 1/3, 2/3, 1} in simulations and main empirical analysis
    The number and location of quantile levels is a researcher choice. The encompassing test guides the number of levels, but the uniform grid is ad hoc; results may depend on this choice.
assumptions (7)
  • standard math Contraction mapping theorem
    Used in Appendix B to prove existence and uniqueness of the Nash equilibrium under the condition sum |λτ| < 1.
  • standard math Sample quantiles are 1-Lipschitz under the infinity norm
    Invoked in Appendix B, equation (B.4), via Lemma B.1 to show the best-response map is a contraction.
  • domain assumption Exogeneity of network and covariates (Assumption 4.1.B)
    Required for validity of Type I instruments; stated in Section 4.3 and used throughout.
  • ad hoc to paper β1 is equal for isolated and non-isolated agents
    Stated in Section 4.4; this equality is necessary to identify λ2 from the ratio of coefficients between Equations (4.2) and (4.3).
  • domain assumption Positive asymptotic share of subnetworks with at least two isolated students
    Assumption in Proposition 4.1 ensuring β1 is identified from the isolated-agent regression.
  • domain assumption Bounded outcomes and compact parameter spaces (Assumption C.1)
    Needed for the consistency of the encompassing test in Appendix C.
  • domain assumption Independence across subnetworks
    Standard assumption for the central limit theorem in Sections 4 and Online Appendix E.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantile Peer Effect Models." pith.science (2026). https://pith.science/paper/WNFNL27H

@misc{pith2026250612920,
  author       = {Pith},
  title        = {Pith review of: Quantile Peer Effect Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WNFNL27H}},
  note         = {Machine review of arXiv:2506.12920}
}
read the original abstract

I propose a flexible structural model to estimate peer effects across various quantiles of the peer outcome distribution. The model allows peers with low, intermediate, and high outcomes to exert distinct influences, thereby capturing more nuanced patterns of peer effects than standard approaches that are based on aggregate measures. I establish the existence and uniqueness of the Nash equilibrium and demonstrate that the model parameters can be estimated using a straightforward instrumental variable strategy. Applying the model to a range of outcomes that are commonly studied in the literature, I uncover diverse and rich patterns of peer influences that challenge assumptions inherent in standard models. These findings carry important policy implications: key player status in a network depends not only on network structure, but also on the distribution of outcomes within the population.

Figures

Figures reproduced from arXiv: 2506.12920 by the authors.

Figure 6.1
Figure 6.1. Influence Measure The x-axis reports student ranks based on influence measure in the quantile model, while the y-axis reports ranks based on the LIM and CES models. Each red triangle represents a student’s LIM model rank (on the y-axis) against their quantile model rank (on the x-axis). Each blue “x” marker represents a student’s CES model rank (on the y-axis) versus their quantile model rank (on the x-axis). The pr… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 25 canonical work pages

  1. [1]

    Akerlof, G. A. (1997): Social distance and social decisions, Econometrica: Journal of the Econometric Society, 1005--1027

    work page 1997

  2. [2]

    Calv \'o -Armengol, and Y

    Ballester, C., A. Calv \'o -Armengol, and Y. Zenou (2006): Who's who in networks. Wanted: The key player, Econometrica, 74, 1403--1417

    2006

  3. [3]

    Blume, L. E., W. A. Brock, S. N. Durlauf, and R. Jayaraman (2015): Linear social interactions models, Journal of Political Economy, 123, 444--496

    work page 2015

  4. [4]

    Boucher, V., F. A. Dedewanou, and A. Dufays (2022): Peer-induced beliefs regarding college participation, Economics of Education Review, 90, 102307

    work page 2022

  5. [5]

    Boucher, V. and B. Fortin (2016): Some challenges in the empirics of the effects of networks, Handbook on the Economics of Networks, 45, 48

    work page 2016

  6. [6]

    Boucher, V. and A. Houndetoungan (2023): Estimating peer effects using partial network data, Tech. rep., Centre de recherche sur les risques, les enjeux économiques et les politiques publiques

    work page 2023

  7. [7]

    Rendall, P

    Boucher, V., M. Rendall, P. Ushchev, and Y. Zenou (2024): Toward a general theory of peer effects, Econometrica, 92, 543--565

    work page 2024

  8. [8]

    Djebbari, and B

    Bramoull \'e , Y., H. Djebbari, and B. Fortin (2009): Identification of peer effects through social networks, Journal of Econometrics, 150, 41--55

    2009

Show all 38 references
  1. [9]

    --- -.1pt --- -.1pt --- (2020): Peer effects in networks: A survey, Annual Review of Economics, 12, 603--629

    2020

  2. [10]

    Kranton, and M

    Bramoull \'e , Y., R. Kranton, and M. D'amours (2014): Strategic interaction and networks, American Economic Review, 104, 898--930

    2014

  3. [11]

    Brock, W. A. and S. N. Durlauf (2007): Identification of binary choice models with social interactions, Journal of Econometrics, 140, 52--75

    2007

  4. [12]

    Patacchini, and Y

    Calv \'o -Armengol, A., E. Patacchini, and Y. Zenou (2009): Peer effects and social networks in education, The Review of Economic Studies, 76, 1239--1267

    2009

  5. [13]

    Dieye, and B

    Comola, M., R. Dieye, and B. Fortin (2022): Heterogeneous peer effects and gender-based interventions for teenage obesity, Tech. rep., Centre interuniversitaire de recherche en analyse des organisations

    2022

  6. [14]

    (2017): Econometrics of network models, in Advances in economics and econometrics: Theory and applications, eleventh world congress, Cambridge University Press Cambridge, 268--323

    De Paula, A. (2017): Econometrics of network models, in Advances in economics and econometrics: Theory and applications, eleventh world congress, Cambridge University Press Cambridge, 268--323

    2017

  7. [15]

    Patacchini, T

    D \' az, C., E. Patacchini, T. Verdier, and Y. Zenou (2021): Leaders in juvenile crime, Journal of Economic Behavior & Organization, 192, 638--667

    2021

  8. [16]

    (1954): A theory of social comparison processes, Human relations, 7, 117--140

    Festinger, L. (1954): A theory of social comparison processes, Human relations, 7, 117--140

    1954

  9. [17]

    Hausman, J. A. (1978): Specification tests in econometrics, Econometrica: Journal of the econometric society, 1251--1271

    1978

  10. [18]

    Herstad, E. I. and M. Shin (2025): Identification of a rank-dependent peer effect model, arXiv preprint arXiv:2410.14317

    2025 arXiv

  11. [19]

    (2024): Count Data Models with Heterogeneous Peer Effects under Rational Expectations, arXiv preprint arXiv:2405.17290

    Houndetoungan, A. (2024): Count Data Models with Heterogeneous Peer Effects under Rational Expectations, arXiv preprint arXiv:2405.17290

    2024

  12. [20]

    Kouame, and M

    Houndetoungan, A., C. Kouame, and M. Vlassopoulos (2024): Identifying Peer Effects in Networks with Unobserved Effort and Isolated Students, arXiv preprint arXiv:2405.06850

    2024 arXiv

  13. [21]

    Houndetoungan, A. and A. H. Maoude (2024): Inference for Two-Stage Extremum Estimators, arXiv preprint arXiv:2402.05030

    2024 arXiv

  14. [22]

    Hsieh, C.-S. and L. F. Lee (2016): A social interactions model with endogenous friendship formation and selectivity, Journal of Applied Econometrics, 31, 301--319

    2016

  15. [23]

    Hyndman, R. J. and Y. Fan (1996): Sample quantiles in statistical packages, The American Statistician, 50, 361--365

    1996

  16. [24]

    Jackson, M. O. and Y. Zenou (2015): Games on networks, in Handbook of game theory with economic applications, Elsevier, vol. 4, 95--163

    2015

  17. [25]

    Johnsson, I. and H. R. Moon (2021): Estimation of peer effects in endogenous social networks: Control function approach, Review of Economics and Statistics, 103, 328--345

    2021

  18. [26]

    Kleibergen, F. and R. Paap (2006): Generalized reduced rank tests using the singular value decomposition, Journal of Econometrics, 133, 97--126

    2006

  19. [27]

    (2025): Peer Effects in Binary Outcomes: Strategic Complementarity and Taste for Conformity With Endogenous Networks, Journal of Applied Econometrics

    Lambotte, M. (2025): Peer Effects in Binary Outcomes: Strategic Complementarity and Taste for Conformity With Endogenous Networks, Journal of Applied Econometrics

    2025

  20. [28]

    (2004): Asymptotic distributions of quasi-maximum likelihood estimators for spatial autoregressive models, Econometrica, 72, 1899--1925

    Lee, L.-F. (2004): Asymptotic distributions of quasi-maximum likelihood estimators for spatial autoregressive models, Econometrica, 72, 1899--1925

    2004

  21. [29]

    Lee, L.-F., X. Liu, E. Patacchini, and Y. Zenou (2021): Who is the key player? A network analysis of juvenile delinquency, Journal of Business & Economic Statistics, 39, 849--857

    2021

  22. [30]

    (2024): Quantile effects in discrete choice with social interactions, Review of Economics and Statistics, 1--35

    Lin, Z. (2024): Quantile effects in discrete choice with social interactions, Review of Economics and Statistics, 1--35

    2024

  23. [31]

    Manski, C. F. (1993): Identification of endogenous social effects: The reflection problem, The Review of Economic Studies, 60, 531--542

    1993

  24. [32]

    Mikusheva, A. and L. Sun (2022): Inference with many weak instruments, The Review of Economic Studies, 89, 2663--2686

    2022

  25. [33]

    (2019): Heterogeneous endogenous effects in networks, arXiv preprint arXiv:1908.00663

    Peng, S. (2019): Heterogeneous endogenous effects in networks, arXiv preprint arXiv:1908.00663

    2019 arXiv

  26. [34]

    Smith, R. J. (1992): Non-nested tests for competing models estimated by generalized method of moments, Econometrica: Journal of the Econometric Society, 60, 973--980

    1992

  27. [35]

    and L.-f

    Tao, J. and L.-f. Lee (2014): A social interaction model with an extreme order statistic, The Econometrics Journal, 17, 197--240

    2014

  28. [36]

    Tatsi, E. (2015): Endogenous Social Interactions: Which Peers Matter? in VfS Annual Conference 2015 (Muenster): Economic Development-Theory and Policy, Verein f \"u r Socialpolitik/German Economic Association, 113168

    2015

  29. [37]

    Ushchev, P. and Y. Zenou (2020): Social norms in networks, Journal of Economic Theory, 185, 104969

    2020

  30. [38]

    Wang, W. and M. Kaffo (2016): Bootstrap inference for instrumental variable models with many weak instruments, Journal of Econometrics, 192, 231--268

    2016

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.