The Science of Misinformation

Key themes

Misinformation as a systemic threat with quantifiable harms. World Economic Forum poll of 1,500 global experts identified misinformation/disinformation as the single largest short-term global risk. Canadian data (March–November 2021): vaccine misinformation contributed to 2.35 million vaccine-hesitant individuals (nearly 1 in 10 Canadians), 198,000 fewer COVID cases, 13,000 fewer hospitalizations, 2,800 fewer deaths, $300 million in hospital cost savings. Johns Hopkins estimate: $50–300 million in daily harm costs (U.S., since May 2021). Van Bavel frames misinformation as underlying cause of polarization, institutional distrust, cyber insecurity, election integrity threats, and armed conflict.

Misinformation sows institutional distrust, not just factual confusion. Disinformation campaigns deliberately target epistemic authorities (doctors, scientists, public health officials, journalists) rather than arguing specific factual points. Goal: make people not trust any source of evidence, creating shared-reality collapse and disengagement. Example: Robert Malone's claim of "mass formation psychosis" as explanation for vaccine uptake—despite fact-checking by AP, Reuters, Vice, and academic experts finding no scientific basis, conspiracy networks reframe fact-checkers as conspirators ("Jay Van Bavel is lying because he's an architect of mass formation psychosis"), reversing the epistemic status of experts.

Emotional content is the vector of spread. Fake news spreads 70% faster and more widely than true news on Twitter (Vosoughi et al., Science). Two factors predict viral misinformation: emotional content (particularly outrage) and political relevance. Van Bavel's lab research analyzing half a million Twitter posts found that each moral-emotional word (e.g., "disgusts me," "so angry," "filled with contempt") increases sharing likelihood by ~20%; 4–5 such words doubles sharing probability. Crucially, moral-emotional language creates echo chambers—spreading only among like-minded users—while reducing cross-partisan engagement. Result: language is good for virality, bad for persuasion and dialogue.

Outgroup targeting is the single strongest predictor of social media engagement. Each term referring to a political outgroup increases odds of a post being shared by 67% across Twitter and Facebook (2.5 billion users). Dunking on other groups is the key to virality. Facebook's angry emoji is algorithmically amplified—posting the angry reaction primes the algorithm to show content to more users. Van Bavel documents evidence of coordinated amplification by foreign actors (Russian trolls increasingly use polarizing language over time; U.S. average Americans do not show this trend, suggesting foreign intent).

Algorithms and attention economy drive engagement-driven amplification, not truth. Platform default feeds are algorithmically curated (Twitter/X, Facebook, TikTok, LinkedIn). Average user scrolls ~600 times per day (equivalent to 300 feet of feed content). Content that "pops" to attention in this firehose is driven by emotional triggers (anger, moral outrage, identity). Algorithms reward high-engagement content; anger and polarizing language generate comments and responses, pushing content up feeds. Platform design is optimized for engagement metrics, not veracity. Twenty percent of TikTok's recommended videos on news topics contain misinformation; medical misinformation is even more saturated (5 of 10 top search results for mRNA vaccines contain misinformation). Demographics shifting: in 2016, older adults (65+) shared misinformation ~7 times more than young adults; TikTok's younger user base is changing this, making future health systems increasingly exposed to misinformation-primed cohorts.

Polarizing language predicts public attitudes and is weaponized by political leaders and foreign actors. Van Bavel's analysis of polarizing language use by U.S. political leaders (2008–2020) shows steady increase, with spike in recent years. Democrats led briefly (~2018), then switched to Republicans, especially those who voted to overturn elections. Anti-democratic political figures use polarizing language most heavily—evidence suggesting manipulation targeted at those least committed to democratic welfare. Russian trolls show increasing polarizing language while average Americans do not, indicating foreign targeting of division. Language is being deliberately used to "trigger and control and manipulate people."

Coordinated harassment as a disinformation tactic. Van Bavel's own experience fact-checking Robert Malone's "mass formation psychosis" claim (debunked by AP, Reuters, Vice, and academic experts) led to coordinated harassment by conspiracy theorist Jack Posobiec (1.5 million followers, linked to white supremacy movement): false accusations ("Van Bavel lied," "caught as architect of mass formation psychosis"), tag flooding, meme campaigns, mobilization of followers to contact university trustees demanding termination. Result: Van Bavel shut down all social media accounts; received ~1,000 harassing messages; conspiracists targeted his wife's social media. Harassment operates to silence experts and fact-checkers through exhaustion and fear. Coordinated campaigns particularly effective when leaders have millions of followers and mobilized base that will follow through.

Misinformation susceptibility is driven by motivation, not cognitive ability. Van Bavel's experiments show misinformation sharing is not strongly predicted by cognitive reasoning ability; instead, it's about what people are motivated to believe and share. This is why intelligence alone doesn't predict resilience—smart people can be motivated to believe identity-congruent claims. The problem is not intellectual but motivational.

Accuracy incentives and social norms are powerful interventions. Van Bavel's lab demonstrates multiple evidence-based approaches, each with experimental validation:

• Monetary incentive: One dollar for spreading accurate information (across four large experiments) significantly increases accuracy, reduces misinformation belief and sharing, and reduces partisan polarization in belief by 30%. People become willing to share true information that makes their own group look bad when incentivized—overcoming the normal tendency to suppress dissonant truths.

• Social-norm incentive: Signaling that one's community values accuracy similarly reduces misinformation spreading dramatically. People are motivated by belonging and community fit; communicating group norms around accuracy leverages this motivation.

• Social-proof signal (misleadingness indicators): Showing people that others in their community found a claim misleading is 5 times more effective than accuracy nudges and reduces sharing by 25%. Effect is stronger for polarizing content and works across political divides (U.S. and U.K.). Social proof (what peers believe) is more powerful than top-down corrections.

• Trusted messengers: Surfacing what local trusted sources believe (e.g., family doctors: 99% agree COVID vaccines safe/effective for adults; 89% for children; pediatric specialists: 100%) reaches people regardless of political identity. Institutional trust remains higher in doctors than politicians or scientists.

• Prebunking/inoculation: Pre-exposure to logic of manipulation techniques (fake experts, cherry-picking, false dichotomies, logical fallacies) builds psychological resilience. Gamified versions (Bad News game, Cranky Uncle game) scale this approach. Inoculation works before misinformation is encountered (unlike corrections, which address existing beliefs).

System-level interventions are essential. Psychology-based interventions (individual and group level) must pair with systemic change: platform algorithm transparency and regulation, media ownership diversity (Australia is unusually concentrated), strong public broadcasters, science communication that is engaging, institutional accountability and transparency, trust-building through leader conduct, and education (media literacy, teaching difference between opinion and fact from early schooling). Individual resilience without systemic change leaves people vulnerable.

Consensus interventions to fight misinformation. Van Bavel, working with CDC and American Psychological Association, identified top evidence-based interventions:

1. Avoid repeating misinformation without correction.
2. Debunk misinformation repeatedly (hearing something over again makes it seem true; counter with repeated truth).
3. Prebunk via inoculation (pre-expose to techniques so people have immunity when encountered).
4. Use media and social media collaboration to amplify evidence-based messaging.

Notes

A comprehensive plenary synthesizing social psychology of misinformation at professional level. Van Bavel integrates personal experience (harassment campaign following Malone fact-check) with empirical research, making both psychological mechanisms and real-world harms concrete. His distinction between virality-driving language (moral-emotional, outgroup-targeting) and persuasion-driving language is particularly useful: engagement metrics are not truth metrics. Emphasis on social-proof signals over top-down corrections aligns with behavioral research and offers practical design guidance for platforms and communicators. The presentation of multiple intervention levels (psychological incentives, social norms, trusted sources, inoculation, systemic change) avoids false choice between individual and structural approaches. Strong on coordination tactics (Russian trolls, domestic harassers) as intentional strategy rather than organic polarization. Discussion of shifting demographics (TikTok younger users) in who spreads misinformation is forward-looking. The Clarity Foundation and BC health authority collaboration (surfacing family doctor consensus on vaccines) exemplifies translating research into practice. Appropriate for researchers, policymakers, health communicators, and platform designers.