Competing contagion model of vaccine misinformation vs. fact-check correction racing across a shared social network
The WHO coined "infodemic" in 2020 to describe an overabundance of information — including deliberate falsehoods — that spreads faster than an outbreak itself and actively undermines the public health response. Every infodemic cascade begins with a seed: a single post, usually from a low-follower, low-credibility account, that happens to combine novelty, emotional charge, and moral outrage in a way that primes it for viral amplification.
In February 2020, WHO formally introduced "infodemic management" as a core pillar of outbreak response, alongside surveillance, case management, and vaccination. The framework (WHO, 2020; refined in the 2021 Public Health Research Agenda for Managing Infodemics) treats the information environment itself as an epidemiological object with measurable dynamics: incidence (new false claims per day), prevalence (share of population exposed), and — critically for this simulation — competing "strains" of information that spread through the same contact network as the disease they describe.
The framework identifies five pillars of infodemic response: 1. Listening to community concerns and information voids 2. Detecting signals of emerging misinformation early 3. Responding with corrective, contextualized information 4. Building societal resistance (prebunking / inoculation) 5. Evaluating impact and adapting strategy
This simulation models the interaction of pillars 3 and 4 — reactive correction (fact-checking) versus proactive resistance-building (inoculation) — as two contagions competing for the same finite pool of attention on a shared social graph.
Vosoughi, Roy & Aral ("The spread of true and false news online," Science, 2018) analyzed ~126,000 cascades of rumor on Twitter spanning 2006–2017, verified by six independent fact-checking organizations. Their central finding: falsehood diffuses significantly farther, faster, deeper, and more broadly than truth in every category studied — and the effect is most pronounced for political and health-related claims.
Mechanistically, false news was found to be more novel than true news (measured via information-theoretic surprisal against a user's recent tweet history), and novel information is preferentially shared because it confers a social-status benefit to the sharer ("I know something you don't"). False stories also inspired reactions marked by higher fear, disgust, and surprise, while true stories inspired anticipation, sadness, joy, and trust — and content that provokes high-arousal negative emotion propagates faster through social networks (a finding consistent with earlier work on "moral-emotional contagion," Brady et al., PNAS 2017).
Critically, bots were found to accelerate false and true news at approximately the same rate — meaning the truth/falsehood asymmetry is driven overwhelmingly by human sharing behavior, not automated amplification. This is why patient-zero seeding matters: the seed post itself does not need bot support to achieve outsized initial reach.
With no competing correction yet in circulation, the misinformation contagion behaves like a classic SIS/SIR epidemic model on a scale-free social network: R > 1 drives exponential early growth, amplified further by algorithmic engagement-ranking systems that reward the emotionally arousing content misinformation tends to be.
Epidemiologically-inspired misinformation models (Kucharski, 2016, "Post-truth: Study epidemiology of fake news," Nature; Jin et al. 2013 "Epidemiological modeling of news and rumors on Twitter") adapt SIS/SIR compartments: Susceptible nodes have not seen the claim; Infected/Believing nodes have seen and accepted it; Recovered nodes have seen it but rejected or forgotten it. On a scale-free network — the typical topology of social platforms, where a small number of hub accounts hold disproportionately many connections — a single seed reaching even one moderately-connected hub can trigger explosive early growth, because the effective reproduction number of the claim (how many new believers each current believer generates) scales with the variance of the degree distribution, not just its mean.
During this unopposed phase, algorithmic curation compounds the raw network effect: engagement-optimized ranking systems (documented across platform transparency reports and academic audits, e.g. Huszár et al. 2022 PNAS on Twitter's algorithmic amplification) preferentially surface high-engagement content to more users, and because emotionally arousing misinformation generates disproportionate engagement (per Vosoughi et al.), the algorithm often unintentionally amplifies the very claims fact-checkers will later need to chase down.
Empirical cascade studies converge on a consistent finding: the majority of a misinformation cascade's total eventual reach is achieved within the first 24–48 hours of publication (Friggeri et al. 2014 "Rumor Cascades," ICWSM; Zannettou et al. 2019 on cross-platform meme/claim propagation). This narrow window is sometimes called the cascade's "golden hours" — the period in which intervention is cheapest and most effective, because the claim has not yet saturated its addressable audience or ossified into settled belief via repeated exposure (the "illusory truth effect," Hasher, Goldstein & Toppino 1977 — repeated statements are judged more true regardless of actual veracity, later replicated for fake news headlines by Pennycook, Cannon & Rand 2018, Journal of Experimental Psychology: General).
Every hour a correction is delayed, the pool of people who saw only the original uncorrected claim grows — and because the illusory truth effect means repetition itself increases perceived credibility, delay compounds both reach AND belief strength simultaneously.
Fact-checking is not instantaneous: claims must be identified, sourced, verified against evidence, written up, and published through an editorial process. This deployment lag — controllable via the slider in this simulation — determines how large a head start misinformation gets before any competing signal enters the network at all.
A misinformation post requires no verification step — its author can publish an unfounded claim in seconds. A rigorous fact-check requires the opposite discipline: identifying primary sources, contacting subject-matter experts, cross-referencing regulatory or scientific literature, writing a clear rebuttal, and passing an editorial review to protect the fact-checker's own credibility. The International Fact-Checking Network (IFCN, est. 2015 at the Poynter Institute) certifies signatories to a Code of Principles that mandates exactly this kind of sourcing rigor — a necessary quality bar that inherently trades off against speed.
During outbreaks specifically, WHO's EPI-WIN (Information Network for Epidemics) and national health authorities have documented turnaround times ranging from same-day (for widely-reported claims with pre-existing scientific consensus) to multiple days or weeks (for novel or technically complex claims requiring new expert consultation). Every hour of this lag is an hour in which the misinformation cascade — already advantaged by the "golden hours" dynamic — spreads uncontested.
This simulation treats fact-check deployment delay as a literal head-start parameter in a two-strain competing contagion model (structurally related to the "competitive influence maximization" literature, e.g. Budak, Agrawal & El Abbadi 2011 "Limiting the spread of misinformation in social networks," WWW): misinformation begins spreading at t=0; the fact-check strain is only permitted to begin spreading at t=delay. Because both strains, once active, spread through overlapping regions of the same network, every additional hour of delay allows misinformation to claim more of the "susceptible" population before any competing signal exists — territory the fact-check can never fully reclaim, because nodes that have already settled into a belief state exhibit resistance to the competing (corrective) strain, analogous to acquired immunity in classic SIR co-circulation models.
In this simulation, dragging the deployment-delay slider to its maximum (120 hours) allows misinfo reach to approach saturation before any fact-check node activates at all — visually demonstrating why WHO infodemic guidance treats rapid-response correction teams as time-critical infrastructure, not a nice-to-have.
Even a well-timed, well-written fact-check faces a structural reach disadvantage: corrections are shared less enthusiastically than the sensational claims they refute, are algorithmically less engaging, and are often published on different channels than the original claim reached. The result is a persistent, measurable "reach gap" between the population exposed to misinformation and the population subsequently exposed to its correction.
Vosoughi, Roy & Aral (Science, 2018) found that the top 1% of false-news cascades routinely reached between 1,000 and 100,000 people, while corrections and fact-checks — even from the same six independent fact-checking organizations used to verify the original dataset — reliably reach a much smaller fraction of that audience. The asymmetry compounds across several independent mechanisms operating simultaneously:
1. Sharing asymmetry: corrective content is less novel and less emotionally arousing than the claim it rebuts, and per the same study's findings, low-novelty/low-arousal content is shared at a measurably lower rate. 2. Channel mismatch: the original claim may go viral on a closed messaging platform (WhatsApp, Telegram) where fact-checks published on open web platforms never penetrate — a dynamic documented extensively in WHO EPI-WIN reporting on vaccine misinformation during COVID-19. 3. Timing mismatch: by the time a correction publishes, many original viewers have moved on and will never see follow-up content in their feed, since most engagement-ranked feeds heavily favor recency. 4. Selective exposure: audiences who sought out or were receptive to the original claim are, by the logic of motivated reasoning, less likely to seek out or engage with content that contradicts it (Kunda 1990, "The case for motivated reasoning," Psychological Bulletin).
This simulation visualizes the gap directly as two overlapping but mismatched contagion fronts sharing the same node population.
Even among the subset of people who ARE exposed to a correction, Lewandowsky, Ecker, Seifert, Schwarz & Cook ("Misinformation and Its Correction: Continued Influence and Successful Debiasing," Psychological Science in the Public Interest, 2012) documented the "continued influence effect": people continue to rely on debunked information in their reasoning even after clearly acknowledging the correction, particularly when the correction does not supply an alternative causal explanation to replace the one the misinformation provided. Simple negation ("X is false") is a weaker corrective format than replacement ("X is false; here is what actually happened, Y").
This means the effective "recovery" of a corrected node in the contagion model is never complete — a portion of corrected nodes retain partial residual belief, which is exactly the phenomenon Stage 6 of this simulation measures as residual false-belief.
The most effective countermeasure to the reach-gap and continued-influence problems is not to fight misinformation after it spreads, but to make the population resistant before it arrives — the psychological equivalent of vaccination. This is "prebunking," grounded in inoculation theory, first proposed for attitudes six decades before it was applied to viral falsehoods.
William McGuire (1961, 1964) proposed that attitudes can be "inoculated" against persuasion the same way a body is immunized against a virus: by exposing a person to a weakened, refutable version of a persuasive attack, along with the tools to counter-argue it, their resistance to the FULL-STRENGTH version of that attack — or even novel attacks using the same underlying technique — increases substantially. McGuire's original experiments used uncontroversial "cultural truisms" (e.g., "brushing your teeth is good for you") and showed that pre-exposure to a mild counter-argument, refuted on the spot, made subjects markedly more resistant to a subsequent strong counter-argument than subjects given no pre-exposure, or even subjects given only supportive (non-challenging) information.
Two active ingredients are required for inoculation to work: (1) a forewarning that one is about to be, or might be, misled, and (2) a weakened refutational pre-exposure that lets the person actively practice counter-arguing. Passive warning alone ("some information you'll see may be false") is measurably weaker than active refutational pre-exposure.
Sander van der Linden and colleagues at Cambridge (Roozenbeek & van der Linden, "Fake news game confers psychological resistance against online misinformation," Palgrave Communications, 2019; and earlier "Inoculating against fake news about climate change," Frontiers in Psychology, 2017) translated McGuire's content-based inoculation into "technique-based" inoculation: rather than refuting one specific false claim, the intervention exposes people to the underlying manipulation techniques used across many different false claims — impersonation, emotional exploitation of outrage, artificial polarization, discrediting/ad hominem, conspiracy framing, and trolling.
The Bad News game (2018) is a free browser game in which players take the role of a fake-news creator, deliberately using these six manipulation techniques to build a fake follower base. By practicing the manipulative technique themselves in a low-stakes simulated environment, players build recognition-based resistance ("if I see this technique being used on me later, I recognize it") that generalizes across topics — unlike topic-specific debunking, which does not transfer to novel claims. Roozenbeek & van der Linden (2019) measured a ~21% average improvement in participants' ability to spot manipulation techniques in unrelated tweets immediately after a 15-minute play session, based on data from roughly 15,000 players, with effects persisting (with some decay) at follow-up.
Go Viral! and the WHO-endorsed "GoViral!" game applied the same technique-inoculation approach directly to COVID-19 and vaccine misinformation. Field trials embedding prebunking video ads directly into social media feeds (Roozenbeek, van der Linden et al., Science Advances, 2022, YouTube field study with millions of views) showed measurable improvement in manipulation-technique recognition at population scale, not just in lab settings.
Because prebunking must be delivered BEFORE exposure to be effective, its coverage — the percentage of the network reached by inoculation content ahead of time — is the single lever in this simulation that acts independently of the fact-check deployment lag. Raise the prebunking-coverage slider and watch shielded (ringed) nodes resist the misinfo contagion even during the unopposed early-cascade phase.
After both contagions burn out — misinformation exhausting its susceptible pool, fact-checks reaching what audience they can — the simulation totals the net outcome: how many nodes hold, and continue to hold, the false belief despite everything that happened afterward. This residual is the true public-health cost of a mismanaged infodemic.
Man-pui Sally Chan, Christopher R. Jones, Kathleen Hall Jamieson & Dolores Albarracín ("Debunking: A Meta-Analysis of the Psychological Efficacy of Messages Countering Misinformation," Psychological Science, 2017) pooled 52 experimental studies and found that debunking reliably reduces misinformation reliance on average — but the reduction is partial, not complete, and its magnitude is strongly moderated by factors including how much detailed counter-argument the debunking message provides, and how strongly the misinformation was initially rehearsed or connected to the person's pre-existing worldview or explanatory narrative.
The meta-analysis also confirmed a "backfire risk" in specific conditions (though later work, e.g. Wood & Porter 2019, "The Elusive Backfire Effect," Political Behavior, found backfire effects are much rarer and smaller than earlier single studies suggested) — reinforcing that correction alone is a necessary but insufficient public-health tool, and must be paired with upstream prevention.
This simulation's final metric — residual false-belief — is designed to make one policy point concrete: the two levers available (fact-check speed and prebunking coverage) are NOT interchangeable. Speeding up fact-check deployment reduces the head-start misinformation gets, narrowing the reach gap, but even an instantaneous correction cannot fully eliminate residual belief because of the continued influence effect documented by Lewandowsky et al. Prebunking, by contrast, prevents susceptible nodes from ever being infected in the first place — shrinking the denominator the reach gap and continued influence effect act on.
WHO's infodemic management guidance (2020–2023 iterations) explicitly frames this as pillars operating on different timescales: rapid response / fact-checking is an outbreak-control tool (fast, reactive, resource-intensive per claim), while building societal resistance through media literacy and prebunking is a public-health infrastructure investment (slower to build, but durable and claim-agnostic — a single inoculation protects against an entire family of future manipulation techniques, not just one debunked claim).
The practical takeaway echoed across the infodemiology literature: the most cost-effective point of intervention is upstream of the first exposure, not downstream of it. A population with high prebunking coverage entering an infodemic event will show a measurably smaller final residual misbelief than a population relying on fast fact-checking alone — even when the fact-check deployment lag in the latter case is minimized.