Understanding and combating misinformation across 16 countries on six continents

Contributions

• Large-scale cross-cultural evidence (N = 34,286; 676,605 observations across 16 countries on 6 continents) on psychological predictors of misinformation susceptibility.
• Demonstration that analytic cognitive style and accuracy-related motivations are robust predictors of truth discernment across diverse cultural contexts.
• Evidence that both attention-based interventions (accuracy prompts) and educational interventions (digital literacy tips) can improve accuracy judgments and reduce sharing of false content.
• Finding that the disconnect between accuracy judgments and sharing intentions persists even with interventions, suggesting distinct psychological drivers of the two decisions.
• Cross-cultural consistency in the effectiveness of interventions, indicating broad-based applicability of tested solutions.

Method

Participants from 16 countries (Argentina, Australia, Brazil, China, Egypt, India, Italy, Mexico, Nigeria, Philippines, Russia, Saudi Arabia, South Africa, Spain, United Kingdom, United States) were randomly assigned to one of four conditions and evaluated COVID-19 headlines (half true, half false). The study used quota-matched sampling based on each country's national distributions of age and sex. In all conditions, participants rated either the accuracy or sharing likelihood of a headline. The three intervention conditions varied: Accuracy condition (focus on truth), Sharing condition (baseline focus on sharing), Prompt condition (accuracy prime before headlines), and Tips condition (digital literacy tips before sharing task).

The researchers examined predictors at the individual level (age, education, political ideology, analytical thinking, accuracy motivations, trust, belief in democracy, individual responsibility, moral relativism, vaccine hesitancy) and country level (GDP, power distance, corruption, collectivism) to explain variation in truth discernment. They used linear regression with demographic controls and mixed-effects models to test whether country-level variables moderated the individual-level relationships.

Results

Individual predictors of truth discernment: Across all countries, participants with more analytic cognitive styles and stronger accuracy-related motivations exhibited better truth discernment (distinguishing true from false headlines). Stronger endorsement of democracy and lower endorsement of individual responsibility over government support were also associated with better truth discernment in most countries. Age, education, and other demographic factors showed more variable relationships.

Intervention effects: The Accuracy prompt intervention increased sharing discrimination (participants were more discerning about which headlines to share) compared to the baseline Sharing condition. The effect size varied by country but was consistent. The Tips intervention (digital literacy instruction) was effective in increasing sharing discrimination, with effect sizes smaller than the accuracy prompt but still significant.

Disconnect between accuracy and sharing: People's judgments about accuracy (whether a headline is true) were substantially better than their judgments about what they would share. Even after interventions that improved accuracy judgments, sharing intentions remained relatively high for false headlines, suggesting that people distinguish between personal accuracy assessment and social-media-relevant sharing decisions.

Country-level variation: The magnitude of the disconnect between accuracy judgments and sharing intentions varied substantially by country. Countries with higher GDP per capita, lower power distance, and less corruption showed greater discernment between accuracy and sharing. However, the direction and significance of individual-level effects remained relatively stable across countries.

Crowdsourcing accuracy: When the study aggregated non-expert participant ratings of headlines using a "wisdom of crowds" approach, the collective judgment of even small groups (as few as 15 ratings per headline) was highly accurate at distinguishing true from false headlines, suggesting potential for scalable crowdsourced fact-checking.

Notes

Strengths: The unprecedented scale and cross-cultural design make this one of the largest systematic studies of psychological predictors of misinformation susceptibility. The disaggregation of accuracy judgments from sharing intentions is conceptually important — it suggests researchers and platforms should not assume that improving accuracy perceptions will automatically reduce misinformation spread. The interventions tested are simple and scalable, making the findings practically relevant.

Limitations: The study relied on headlines rather than full articles, so results may not generalize to detailed narratives. Participants were recruited via online survey platforms (Lucid Marketplace), introducing potential selection bias toward digitally literate, economically active populations. The COVID-19 context may not generalize to other misinformation domains (politics, health claims outside vaccines, conspiracy theories). The study measured stated sharing intentions rather than actual sharing behavior on social media.

Open questions: Whether the observed psychological predictors and intervention effects persist for other misinformation topics (election fraud claims, pharmaceutical misinformation, conspiracy theories). Whether actual sharing behavior on social media aligns with stated intentions. Whether the interventions can be embedded durably into social media platforms or require one-off application. How quickly treatment effects decay after the intervention ends.