Contributions
- Quantified the magnitude of COVID-19 misinformation on Twitter during the early pandemic (February 2020)
- Identified account-type patterns: informal accounts had 2.3× higher misinformation rates than other account types
- Demonstrated that verification status is a strong predictor: verified accounts had 2.5× lower misinformation rates
- Found no relationship between tweet engagement (likes, retweets) and content accuracy
- Showed hashtag/search-term variations in misinformation prevalence, with "#2019_nCov" reaching 44% misinformation
Method
Conducted a cross-sectional Twitter search on February 27, 2020 using 14 trending COVID-19 hashtags and keywords (e.g., #2019_nCov, #coronavirus, COVID-19, Corona). Selected a random sample of 50 tweets from search terms yielding >100 results, restricted to English-language tweets with ≥5 retweets. Manually assessed each tweet by classifying account type (informal individual/group, business/NGO/government, news outlet, healthcare/public health) and verification status. Tweets were coded for content tone (serious, humorous, opinion), topic type (medical/public health, financial, sociopolitical), and factuality. Factuality assessment compared tweet claims against WHO, CDC, peer-reviewed journals, and prominent news outlets; claims that could be clearly refuted were marked misinformation; those that could not be verified either way were marked unverifiable.
Results
Analyzed 673 tweets. Account composition: 66.6% informal individuals/groups, 16.5% news outlets/journalists, 10.8% healthcare/public health, 5.5% business/NGO/government; 19.2% verified accounts. Tweet content: 81.4% genuine information, 91.2% serious tone, 69.5% medical/public health topic. Misinformation rates by account type: informal 33.8% (p < 0.001), news 18.6%, healthcare 12.3%, business 6.1%. Verified vs. unverified accounts: 12.6% vs. 31.0% misinformation (p < 0.001). Unverifiable information: 17.4% overall, lowest in verified accounts (8.6%) and healthcare accounts (12.3%). No significant associations between likes/retweets and misinformation (p = 0.98 and 0.36). Hashtag variation: "#2019_nCov" had 44% misinformation; "COVID-19" had lowest at 8% misinformation; "Corona" had highest unverifiable information at 38%.
Notes
Strength: Very early quantification (February 2020, before widespread awareness of COVID-19's severity), manual annotation with explicit verification criteria against authoritative sources, rigorous statistical analysis of account and engagement characteristics. The finding that engagement metrics do not correlate with accuracy is important — it challenges assumptions that social algorithms naturally surface truth.
Limitations acknowledged: English-language only (impacts generalizability), specific hashtag search may miss relevant tweets, single-day snapshot limited temporal scope. The paper calls for interventions but does not prescribe specific solutions; leaves room for follow-up work on longitudinal evolution and intervention efficacy.
The paper is published in Cureus, an open-access multidisciplinary journal; does not carry the prestige weight of top-tier epidemiology venues but does ensure open accessibility for public health practitioners.