cib.pdf.
Summary
This pre-registered study leverages privileged access to Meta’s platform data through the US 2020 Facebook and Instagram Election Study to measure the actual reach and audience of 49 “deceptive online networks” — 13 Coordinated Inauthentic Behavior (CIB) and 36 Financially-Motivated Operations (FMO) networks — that targeted US users around the 2020 election. The authors argue that financially-motivated networks producing political content deserve scrutiny alongside explicitly political influence operations, and they introduce “deceptive online networks” as a unifying category defined by coordination plus identity deception. Empirically, they show that these networks reached ~37M Facebook users (15% of US adults) and ~3M Instagram users, but that reach was extraordinarily concentrated in a handful of networks and overwhelmingly driven by reshares from ordinary, unaffiliated users. Crucially, observed associations between exposure and political outcomes collapse under covariate adjustment and are not robust to plausible unobserved confounding.
Key Contributions
- Introduces and operationalizes deceptive online networks as a unifying concept spanning politically- and financially-motivated coordinated identity deception.
- Provides the first large-scale, platform-measured study of actual exposure (not engagement proxies) to such networks on Facebook and Instagram during a major US election.
- Documents the underexamined prevalence of FMOs producing substantial political content, challenging their dismissal as mere spam.
- Empirically demonstrates the pivotal role of ordinary, non-network users in amplifying deceptive content via reshares.
- Releases a curated dataset and code via SOMAR/ICPSR for further research.
- Offers a methodological caution: naive exposure–outcome correlations are highly sensitive to confounding and should not be read causally.
Methods
Pre-registered observational analysis (OSF, October 2020) drawing on three data layers: (1) aggregated platform-level exposure data for ~250M Facebook and ~160M Instagram US adult users, distinguishing direct exposure from network accounts and indirect exposure via reshares by non-network accounts; (2) ~73,000 consenting FIES survey respondents with linked on-platform behavior across five waves (Aug–Dec 2020); (3) network-level metadata for 49 networks identified by Meta. Content was classified using Meta’s Topic and Civic classifiers. Diffusion was characterized via cascade size, depth, breadth, and structural virality. Associations with truth discernment, perceived election legitimacy, and partisan news clicks were estimated with entropy balancing on pre-exposure covariates, plus pre-registered sensitivity analyses for unobserved confounding. Post-hoc analyses leveraged FIES experimental arms (no-reshares feed, reverse-chronological feed, reduced like-minded content).
Findings
- Deceptive networks reached ≥36.79M unique US adult Facebook users (14.63%) and 2.98M Instagram users (1.85%) over eight months.
- Origins were diverse: CIB networks from Russia (5), the US (3), Iran (2), China, Romania, and transnational actors; FMOs concentrated in the Balkans (16) and South Asia (11).
- Politics/social issues dominated content for both CIB (65% of direct Facebook content) and FMO (32%) networks, undercutting the “FMOs are apolitical” framing.
- Reach was extremely skewed: 1% of Instagram users absorbed 96% of network views; three Facebook networks (CIB9/Rally Forge, FMO27, FMO34) drove ~70–80% of total Facebook reach.
- Indirect exposure dominated: e.g., CIB9 reached 1.3M directly vs. 13M via reshares from unaffiliated users.
- Only 5.67% of Facebook and 0.34% of Instagram viewers reshared network content, but this small group did the bulk of amplification.
- Exposed users skewed older, more conservative, heavier platform users, and more exposed to untrustworthy sources — a pattern stable across CIB/FMO and direct/indirect exposure.
- Among FIES participants, deceptive network content averaged just 0.3% of political content views in the 41 pre-election days.
- Bivariate links between exposure and worse truth discernment / more partisan news clicks vanished after entropy balancing and failed sensitivity tests for unobserved confounding.
- Only 12 of 49 networks ran ads; paid promotion was 0.6% of posts.
Connections
No related papers were provided under the assigned topics, so no wikilinks are warranted here. Substantively, the paper sits at the intersection of work on Russian-era influence operations, platform-collaborative election studies, and the misinformation-exposure literature that has shown small audiences and weak persuasive effects — extending all three by broadening the unit of analysis to include financially-motivated operations and by foregrounding reshare-driven amplification by ordinary users.