Observational studies of diffusion in social networks cannot cleanly separate peer influence (social contagion) from homophily: linked nodes adopt similar behaviors either because adoption spreads across ties or because similar nodes form ties. Aral, Muchnik, and Sundararajan (2009) introduced a dynamic matched-sample estimation framework that separates the two on observed characteristics in a 27.4-million-user instant-messaging network, showing prior methods overstate peer influence by 300–700% and that homophily explains more than half of apparent contagion; but the authors themselves flag that their estimates are only upper bounds on influence, since unobserved and uncorrelated latent homophily and unmeasured confounding are not accounted for, and the framework does not handle settings where tie formation is itself endogenous to the behavior. This record states the still-open problem: develop identification and estimation of peer influence in dynamic networks that is robust to unobserved latent homophily and to selection-driven (behavior-contingent) tie formation, and — as an empirical application encouraged by the source — determine how influence magnitudes and homophily-mediated bias differ for products with direct network externalities.

Background

Peer influence, or social contagion, is the causal effect of one person's behavior or adoption on the behavior of their network contacts; homophily is the tendency of similar people to form social ties with one another. Both processes produce the same observable signature — assortative mixing (linked nodes resembling each other) and temporal clustering of adoptions — so network diffusion data alone cannot say which mechanism generated them. The problem matters wherever contagion claims guide action: viral marketing, epidemiology, public health campaigns, and development economics. Manski's reflection problem (the basic non-identification result for endogenous social effects in econometrics, DOI 10.2307/2298123) and the literature on peer-effect identification in networks show that separating these mechanisms requires strong assumptions. Aral, Muchnik, and Sundararajan (2009, DOI 10.1073/pnas.0908800106) proposed a dynamic matched-sample estimation framework: adopters are matched to non-adopters who are statistically indistinguishable on a rich panel of observed individual, relational, and geographic characteristics (so that homophily on observed and observed-correlated latent traits is controlled by construction), and the difference in adoption hazard between adopters' contacts and matched controls' contacts identifies influence under the assumption that no unobserved confounders remain. Applying it to day-by-day adoption of the Yahoo! Go mobile application across the Yahoo! instant-messaging network, they found previous methods overestimate peer influence by 300–700% and homophily explains over 50% of perceived contagion. The authors state three limitations that define the open problem. First, unobserved and uncorrelated latent homophily and unmeasured confounding or contextual effects (e.g., correlated advertising exposure, common unobserved contacts) are not controlled, so matched-sample estimates are interpretable only as upper bounds on influence. Second, the framework assumes tie formation is exogenous to the behavior under study (adopting Yahoo! Go does not create friendships), whereas in many contexts links form because of the behavior — the co-evolution of networks and behavior studied by the actor-oriented SIENA models (SIENA stands for Simulation Investigation for Empirical Network Analysis, a family of stochastic models for the joint longitudinal evolution of network ties and node-level behaviors) of Snijders, Steglich, and Schweinberger (2006, DOI 10.4324/9781315091655-3), which model selection and influence jointly but do not scale to networks of millions of nodes; extending scalable influence estimation to such selection processes was left open. Third, Yahoo! Go exhibits no direct network externalities — situations where a product becomes more valuable to a user as their contacts adopt it — and the authors suspect peer influence magnitudes differ for such products, encouraging empirical applications there. Later theory sharpened the first limitation into a structural obstacle: Shalizi and Thomas (2011, DOI 10.1177/0049124111404820) proved that homophily, contagion, and covariate effects are generically confounded in observational network studies, so identification requires either strong parametric assumptions or adequately rich covariates.

Problem Statement

How can peer influence (social contagion) be identified and estimated in dynamic networks when homophily is only partly observed and tie formation responds to the behavior itself? Concretely, the limitations that Aral, Muchnik, and Sundararajan (2009, DOI 10.1073/pnas.0908800106) flag for their dynamic matched-sample framework are threefold and unified by one scientific objective — credible quantification of true contagion in networks: (i) obtain estimates of peer influence that are not merely upper bounds, by accounting for unobserved and uncorrelated latent homophily, unmeasured confounding, and contextual effects (such as correlated exposure to advertising or information from common unobserved contacts); (ii) extend influence estimation to settings where link formation is endogenous to the focal behavior — the co-evolution of network structure and behavior, or selection-on-behavior — where the matching assumption that ties are exogenous fails; and (iii) apply such validated estimation to products or behaviors with direct network externalities, where adoption directly increases utility through others' adoption, to determine whether peer influence magnitudes and the degree of homophily-mediated bias differ from the no-externalities case. The central methodological objective is an identification and estimation strategy for peer influence in dynamic networks, with its required assumptions made explicit, that is robust to (i) and (ii); the direct-externalities comparison of (iii) is an empirical application the authors encourage for such validated estimation, rather than a condition the source places on resolving (i) and (ii).

References

  1. Sinan Aral, Lev Muchnik, and Arun Sundararajan (2009), 'Distinguishing influence-based contagion from homophily-driven diffusion in dynamic networks', Proceedings of the National Academy of Sciences 106(51):21544-21549, DOI 10.1073/pnas.0908800106, https://doi.org/10.1073/pnas.0908800106
  2. Charles F. Manski (1993), 'Identification of Endogenous Social Effects: The Reflection Problem', The Review of Economic Studies 60(3):531-542, DOI 10.2307/2298123, https://doi.org/10.2307/2298123
  3. Tom A. B. Snijders, Christian Steglich, and Michael Schweinberger (2006), 'Modeling the Coevolution of Networks and Behavior', in Longitudinal Models in the Behavioral and Related Sciences, Lawrence Erlbaum Associates (Routledge reissue 2017, pp. 41-71), DOI 10.4324/9781315091655-3, https://doi.org/10.4324/9781315091655-3
  4. Cosma Rohilla Shalizi and Andrew C. Thomas (2011), 'Homophily and Contagion Are Generically Confounded in Observational Social Network Studies', Sociological Methods & Research 40(2):211-239, DOI 10.1177/0049124111404820, https://doi.org/10.1177/0049124111404820
  5. Tyler J. VanderWeele (2011), 'Sensitivity Analysis for Contagion Effects in Social Networks', Sociological Methods & Research 40(2):240-255, DOI 10.1177/0049124111404821, https://doi.org/10.1177/0049124111404821
  6. Christian E. G. Steglich, Tom A. B. Snijders, and Michael Pearson (2010), 'Dynamic Networks and Behavior: Separating Selection from Influence', Sociological Methodology 40(1):329-393, DOI 10.1111/j.1467-9531.2010.01225.x, https://doi.org/10.1111/j.1467-9531.2010.01225.x
  7. Sinan Aral and Dylan Walker (2011), 'Creating Social Contagion Through Viral Product Design: A Randomized Trial of Peer Influence in Networks', Management Science 57(9):1623-1639, DOI 10.1287/mnsc.1110.1421, https://doi.org/10.1287/mnsc.1110.1421
  8. Lev Muchnik, Sinan Aral, and Sean J. Taylor (2013), 'Social Influence Bias: A Randomized Experiment', Science 341(6146):647-651, DOI 10.1126/science.1240466, https://doi.org/10.1126/science.1240466
  9. Ravi Bapna and Akhmed Umyarov (2015), 'Do Your Online Friends Make You Pay? A Randomized Field Experiment on Peer Influence in Online Social Networks', Management Science 61(8):1902-1920, DOI 10.1287/mnsc.2014.2081, https://doi.org/10.1287/mnsc.2014.2081
  10. Ran Xu (2018), 'Alternative estimation methods for identifying contagion effects in dynamic social networks: A latent-space adjusted approach', Social Networks 54:101-117, DOI 10.1016/j.socnet.2018.01.002, https://doi.org/10.1016/j.socnet.2018.01.002
  11. Edward McFowland III and Cosma Rohilla Shalizi (2023, online first 2021), 'Estimating Causal Peer Influence in Homophilous Social Networks by Inferring Latent Locations', Journal of the American Statistical Association 118(541):707-718, DOI 10.1080/01621459.2021.1953506, https://doi.org/10.1080/01621459.2021.1953506
  12. Xi Chen, Yan Liu, and Cheng Zhang (2022), 'Distinguishing Homophily from Peer Influence Through Network Representation Learning', INFORMS Journal on Computing 34(4):1958-1969, DOI 10.1287/ijoc.2022.1171, https://doi.org/10.1287/ijoc.2022.1171
  13. Naoki Egami (2024), 'Identification of causal diffusion effects using placebo outcomes under structural stationarity', Journal of the Royal Statistical Society Series A: Statistics in Society 187(4):944-966, DOI 10.1093/jrsssa/qnae014, https://doi.org/10.1093/jrsssa/qnae014
  14. Anton Badev (2021), 'Nash Equilibria on (Un)Stable Networks', Econometrica 89(3):1179-1206, DOI 10.3982/ecta12576, https://doi.org/10.3982/ecta12576
  15. Prasanta Bhattacharya, Tuan Q. Phan, Xue Bai, and Edoardo M. Airoldi (2019), 'A Coevolution Model of Network Structure and User Behavior: The Case of Content Generation in Online Social Networks', Information Systems Research 30(1):117-132, DOI 10.1287/isre.2018.0790, https://doi.org/10.1287/isre.2018.0790
  16. Wenshuo Wang, Edvard Bakhitov, and Dominic Coey (2026), 'Experimentation on Endogenous Graphs', Proceedings of the AAAI Conference on Artificial Intelligence 40(31):26498-26507, DOI 10.1609/aaai.v40i31.39857, https://doi.org/10.1609/aaai.v40i31.39857
  17. Seungha Um, Tracy Sweet, and Samrachana Adhikari (2024), 'A Bayesian Approach to Estimate Causal Peer Influence Accounting for Latent Network Homophily', arXiv:2405.14789, https://arxiv.org/abs/2405.14789
  18. Hanh T. D. Pham and Daniel K. Sewell (2024), 'Homophily-adjusted social influence estimation', arXiv:2405.18413, https://arxiv.org/abs/2405.18413
  19. Duncan A. Clark (2026), 'A Sensitivity Framework for Identifying Contagion under Latent Homophily for Fixed-in-Time Network Analyses, with an Application to U.S. House Congressional Voting', arXiv:2606.18197, https://arxiv.org/abs/2606.18197
  20. Zhibin Wu, Yueyuan Li, and Witold Pedrycz (2026), 'Peer influence in social network: a review of identification, modeling and application', Journal of Business Research 203:115806, DOI 10.1016/j.jbusres.2025.115806, https://doi.org/10.1016/j.jbusres.2025.115806
  21. Romain Ferrali, Guy Grossman, Melina R. Platas, and Jonathan Rodden (2020, online first 2019), 'It Takes a Village: Peer Effects and Externalities in Technology Adoption', American Journal of Political Science 64(3):536-553, DOI 10.1111/ajps.12471, https://doi.org/10.1111/ajps.12471
  22. Liangfei Qiu, Qian Tang, and Andrew B. Whinston (2015), 'Two Formulas for Success in Social Media: Learning and Network Effects', Journal of Management Information Systems 32(4):78-108, DOI 10.1080/07421222.2015.1138368, https://doi.org/10.1080/07421222.2015.1138368
  23. Dorian Quelle, Frederic Denker, Prashant Garg, and Alexandre Bovet (2025), 'Simple contagion drives population-scale platform migration', arXiv:2505.24801, https://arxiv.org/abs/2505.24801