Vargo and Guo's large-scale 2015 U.S. analysis — daily issue-salience networks for 2,760 news outlets built from 53.9 million machine-coded articles and probed with thousands of pairwise Granger tests — concluded that aggregated online partisan media led the news agenda while The New York Times and The Washington Post followed. The study's own central caveat remains unresolved: Granger-style temporal predictability cannot distinguish direct intermedia influence from shared exogenous forcing, because real-world events, public-opinion shifts, press releases, and government content can jointly move many outlets' agendas at once. This problem asks for a determination of how much of the observed lead-lag structure between media agendas reflects genuine outlet-to-outlet influence and how much reflects common response to external drivers. The question turns on stated identification assumptions, an explicitly defined and justified attribution estimand, and a substantively non-vacuous attribution for actually observed intermedia lead-lag data (informative partial-identification bounds are admissible where point identification is unavailable), with the original study's headline conclusions as the motivating application.

Background

Agenda-setting theory holds that news media shape which issues the public comes to regard as important (McCombs and Shaw, 1972). A companion strand, intermedia agenda setting (IAS), studies influence among news outlets themselves: which outlets' issue agendas lead and which follow. In the network agenda-setting (NAS) model — the so-called third level of the theory — salience is carried not by single issues in isolation but by bundles of issues, represented as a network whose nodes are issues and whose ties are their co-occurrence in coverage; an outlet's agenda is then the salience profile of issues in this network. Eigenvector centrality, a standard node-importance measure that scores an issue by the salience of the issues it co-occurs with, is one common way to summarize that profile as a time series. Lead-lag structure between such series is usually tested with Granger causality (Granger, 1969) — a test of temporal predictability in which a series X is said to Granger-cause Y if past values of X improve predictions of Y beyond Y's own past, typically via regression F-tests — or with the older, related technique of cross-lagged correlation across successive time panels. Neither is causal identification in the counterfactual sense: predictability can arise between two series with no direct causal link between the actors they describe.

Vargo and Guo (2017; first published online December 2016) combined these tools at unprecedented scale. Using GDELT, a machine-coded database that extracts events and themes from global news coverage, they processed 53,967,878 U.S. news articles from 2015, manually classified 2,760 outlets into media types (the elite newspapers The New York Times and The Washington Post; news agencies; other traditional media; online-only partisan outlets; emerging non-partisan online outlets), built daily weighted issue co-occurrence networks over 16 issue constructs per media type, summarized each issue's daily salience by eigenvector centrality, and ran pairwise Granger tests across media-type pairs, issues, and five daily lags. The headline findings: media agendas were highly homogeneous and reciprocal; aggregated online partisan media led the overall media agenda on most issues; and the two elite newspapers tended to follow partisan and emerging online media rather than lead them.

The identification weakness this problem concerns was flagged by the study's authors themselves. The design measures temporal predictability between aggregated agenda series but does not model or control for exogenous drivers — real-world cues and events, public-opinion shifts, press releases, and government website content — that could simultaneously move many outlets' agendas. When a major event breaks, every outlet's coverage responds to the event itself, not to each other; observed Granger links would then reflect shared external forcing rather than direct intermedia influence. Because wire services, shared sourcing, and real-world events drive coverage jointly across outlets, the fraction of the observed lead-lag structure that is attributable to genuine outlet-to-outlet influence is unknown, and conclusions such as 'partisan media lead the news agenda' may describe common response to the world rather than influence. Subsequent reviews of the IAS literature (Su and Xiao, 2021) confirm that cross-lagged correlation and Granger-type tests have remained the dominant time-series techniques in this research tradition, so the question of how to disentangle direct influence from shared forcing — and thus what the field's big-data lead-lag findings actually mean — was left open by the source study and has not been closed since.

Problem Statement

In large-scale intermedia agenda-setting analyses of the Vargo-Guo type — daily issue-salience series (such as eigenvector-centrality profiles of issue co-occurrence networks) for many outlets or outlet aggregates, tested pairwise with Granger-causality or cross-lagged correlation — determine the extent to which the observed lead-lag links between media agendas reflect direct intermedia influence versus shared forcing by exogenous drivers (for example, real-world cues and events, public-opinion shifts, press releases, government website content) that jointly move multiple media agendas.

The determination turns on an explicitly defined and justified attribution estimand — for example, the proportion of significant pairwise lead-lag links that survive conditioning on common drivers, a variance or predictive-power decomposition per link or per media-type pair, or a counterfactual contrast under an intervention on drivers — since the source study prescribes no single quantity. It turns on stated identification assumptions and on an explicit account of what remains confounded by unmeasured common causes such as wire-service copy, shared sourcing, and professional norms; where point identification is unavailable under such residual confounding, informative partial-identification bounds are admissible. The determination must be substantively about actually observed intermedia lead-lag data — the original 2015 corpus or a comparable real corpus — with uncertainty estimates and robustness checks; validation on simulated data alone, or trivial bounds such as the full interval [0,1][0,1] on real data, does not answer the question. The re-assessment of the original 2015 conclusions — whether the leading role of aggregated online partisan media and the follower position of The New York Times and The Washington Post survive once shared external forcing is taken into account — is the motivating application of this question. Multiplicity control across large numbers of pairwise tests is good practice at this scale and has been demonstrated in Granger-based intermedia designs (Stern, Livan and Smith 2020, using Benjamini-Hochberg correction), but it does not by itself resolve the influence-versus-forcing confound.

References

  1. Vargo, Chris J., and Lei Guo (2017). Networks, Big Data, and Intermedia Agenda Setting: An Analysis of Traditional, Partisan, and Emerging Online U.S. News. Journalism & Mass Communication Quarterly 94(4):1031-1055 (first published online December 2016). DOI: 10.1177/1077699016679976. https://doi.org/10.1177/1077699016679976
  2. McCombs, Maxwell E., and Donald L. Shaw (1972). The Agenda-Setting Function of Mass Media. Public Opinion Quarterly 36(2):176-187. DOI: 10.1086/267990. https://doi.org/10.1086/267990
  3. Granger, C. W. J. (1969). Investigating Causal Relations by Econometric Models and Cross-spectral Methods. Econometrica 37(3):424-438. DOI: 10.2307/1912791. https://doi.org/10.2307/1912791
  4. Guo, Lei, and Chris J. Vargo (2017). Global Intermedia Agenda Setting: A Big Data Analysis of International News Flow. Journal of Communication 67(4):499-520. DOI: 10.1111/jcom.12311. https://doi.org/10.1111/jcom.12311
  5. Vargo, Chris J., Lei Guo, and Michelle A. Amazeen (2018). The agenda-setting power of fake news: A big data analysis of the online media landscape from 2014 to 2016. New Media & Society 20(5):2028-2049 (first published online June 2017). DOI: 10.1177/1461444817712086. https://doi.org/10.1177/1461444817712086
  6. Guo, Lei, and Chris Vargo (2020, first published online June 2018). "Fake News" and Emerging Online Media Ecosystem: An Integrated Intermedia Agenda-Setting Analysis of the 2016 U.S. Presidential Election. Communication Research 47(2):178-200. DOI: 10.1177/0093650218777177. https://doi.org/10.1177/0093650218777177
  7. Stern, Samuel, Giacomo Livan, and Robert E. Smith (2020). A network perspective on intermedia agenda-setting. Applied Network Science 5(1):31. DOI: 10.1007/s41109-020-00272-4. https://doi.org/10.1007/s41109-020-00272-4
  8. Hopp, Frederic R., Jacob T. Fisher, and René Weber (2020). Dynamic Transactions Between News Frames and Sociopolitical Events: An Integrative, Hidden Markov Model Approach. Journal of Communication 70(3):335-355. DOI: 10.1093/joc/jqaa015. https://doi.org/10.1093/joc/jqaa015
  9. Su, Yan, and Xizhu Xiao (2021). Mapping the intermedia agenda setting (IAS) literature. The Agenda Setting Journal 5(1):56-83. DOI: 10.1075/asj.20001.su. https://doi.org/10.1075/asj.20001.su
  10. Budak, Ceren, Natalie Jomini Stroud, Ashley Muddiman, Caroline C. Murray, and Yujin Kim (2023). The Stability of Cable and Broadcast News Intermedia Agenda Setting Across the COVID-19 Issue Attention Cycle. Political Communication 40(6):827-847. DOI: 10.1080/10584609.2023.2222382. https://doi.org/10.1080/10584609.2023.2222382
  11. Litterer, Benjamin, David Jurgens, and Dallas Card (2023). When it Rains, it Pours: Modeling Media Storms and the News Ecosystem. arXiv:2312.02118. https://arxiv.org/abs/2312.02118
  12. Chen, Yijing, Yaguang Liu, Lisa Singh, and Ceren Budak (2024). Intermedia Agenda Setting during the 2016 and 2020 U.S. Presidential Elections. Proceedings of the International AAAI Conference on Web and Social Media 18(1):254-275. DOI: 10.1609/icwsm.v18i1.31312. https://doi.org/10.1609/icwsm.v18i1.31312
  13. Su, Yan, and Xizhu Xiao (2024). Intermedia Attribute Agenda Setting Between the U.S. Mainstream Newspapers and Twitter: A Two-Study Analysis of the Paradigm and Driving Forces of the Agenda Flow. Journalism & Mass Communication Quarterly 101(2):451-476. DOI: 10.1177/10776990231221150. https://doi.org/10.1177/10776990231221150
  14. Vargo, Chris J., and Milad Minooie (2025, first published online April 2024). Breaking Agenda Setting Boundaries: A Multi-Dimensional Approach to Understanding Salience of Gun Control in the Polarized Public Sphere. Mass Communication and Society 28(3):530-553. DOI: 10.1080/15205436.2024.2333972. https://doi.org/10.1080/15205436.2024.2333972
  15. Ceylan, Mehmet, and Cem Yaşın (2026). Shaping the narrative: media and X (Twitter) influence during the 2020 US election. Quality & Quantity (published online July 2026). DOI: 10.1007/s11135-026-02955-z. https://doi.org/10.1007/s11135-026-02955-z