Contributions
- Large-scale empirical comparison of misinformation prevalence across social media (Twitter) and traditional news media during the early COVID-19 pandemic
- Demonstrates differential media effects: social media exposure increases misperceptions about COVID-19 while news media reduces them
- Links misinformation exposure to behavioral compliance with public health measures (social distancing)
- Provides evidence that misperceptions act as a partial mediator in the relationship between social media exposure and non-compliance with distancing measures
Method
The study employs a mixed-methods sequential design combining three data sources:
Twitter Analysis: Collected 2.5 million tweets on COVID-19 from 620,000 identified Canadian users (March 26–April 6, 2020). Manually coded a random sample of tweets from four content categories: those claiming COVID-19 is no worse than the flu, that vitamin C/supplements prevent infection, that the virus originated from eating bats, or that COVID-19 is a hoax/conspiracy. Coded each tweet as propagating misinformation, debunking misinformation, or unrelated.
News Media Analysis: Collected articles from 19 prominent Canadian news outlets with active RSS feeds during the same period, yielding 9,148 articles. Applied the same keyword-based categorization and manual coding scheme to identify misinformation prevalence and health recommendations.
Survey: Nationally representative survey of 2,500 Canadian adults (April 2–6, 2020) weighted by age, gender, and region. Measured COVID-19 misperceptions via nine true/false items (vitamin C prevents infection, virus no worse than flu, etc.), COVID-19 risk perceptions (personal and societal), and social distancing compliance across ten behaviors. Media consumption assessed via self-reported exposure to 17 news outlets and 10 social media applications over the past week. Measured offline and online political discussion.
Results
Finding 1: Misinformation prevalence differs sharply by platform. Social media contained substantially more misinformation content (green bars in Figure 1) across all four categories examined (vitamin C, bat consumption, severity, conspiracy), while news media emphasized public health recommendations for hygiene and social distancing. Even accounting for longer average article length, misinformation appears less frequently in news media.
Finding 2: Social media exposure increases misperceptions; news media exposure decreases them. Among survey respondents, moving from no social media exposure to high exposure increased COVID-19 misperceptions by 0.22 on a 0–1 scale and decreased social distancing compliance by 0.12. In contrast, increasing news exposure reduced misperceptions by 0.12 and increased compliance by 0.28. The effects remain significant controlling for socioeconomic characteristics and demographics.
Finding 3: Misperceptions mediate the social-media-to-noncompliance link. COVID-19 misperceptions are powerfully associated with lower social distancing compliance (reduction of 0.39 on the 0–1 scale when moving from lowest to highest misperceptions). When including controls for science literacy and fundamental predispositions (pseudoscience beliefs, anti-intellectualism, ideology), the relationship between social media exposure and non-compliance is partially mediated by misperceptions, suggesting social media affects behavior partly through its effect on belief formation.
Notes
This is a high-quality empirical study that directly links media exposure to both belief formation and behavioral compliance in a consequential public health crisis. The mixed-methods triangulation (tweets + articles + survey) provides credible evidence. Notable strengths: large-scale data collection, nationally representative survey, measurement of actual behaviors (not just attitudes), inclusion of mechanistic mediators (misperceptions). The study was conducted early in the pandemic (April 2020), so findings may not fully generalize to later periods or to non-Canadian contexts. The focus on "misperceptions" broadly rather than specific false claims limits granularity on what misinformation actually spreads, but increases replicability.