Social media and vaccine hesitancy¶

Authors: Steven Lloyd Wilson, Charles Wiysonge Venue: BMJ Global Health, 2020 — DOI

TL;DR¶

Large-scale cross-national analysis finds that social media use to organize offline action predicts public vaccine safety skepticism, while foreign disinformation campaigns significantly reduce vaccination coverage year-over-year. A one-point increase on the five-point disinformation scale correlates with a two-percentage-point drop in mean vaccination coverage and a 15% increase in negative vaccine tweets.

Contributions¶

Large-n cross-national framework evaluating social media effects on vaccine hesitancy globally across 137–166 countries, filling a gap in literature dominated by single-country or small-scale studies
Multidimensional operationalization of social media using: public use for organizing action (Digital Society Project), negative discourse prevalence (sentiment analysis of 258,769 geocoded tweets 2018–2019), and foreign disinformation campaigns
Dual outcome measurement via both public attitudes (Wellcome Global Monitor polling on vaccine safety) and behavioral outcomes (WHO vaccination coverage data 2000–2018)
Empirical evidence for disinformation impact on vaccination rates, controlling for socioeconomic factors (GDP, education) and regional structural differences

Method¶

Cross-country regression analysis examining two research questions:

Social media organisation and vaccine hesitancy: Does public use of social media to organize action predict population-level vaccine safety skepticism? Uses Digital Society Project expert-coded indicator (5-point Likert: "never" to "regularly") for the question "How often do average people use social media to organise offline political action of any kind?"

Foreign disinformation and vaccination rates: Do foreign government disinformation campaigns reduce vaccination coverage? Uses Digital Society Project indicator on foreign governments' use of social media to spread misleading information (5-point scale, inverted so higher = more disinformation). Measured as country-year panel data 2000–2018.

Data sources: - Sentiment analysis of geocoded tweets: 2.5 billion geocoded tweets (2018–2019) filtered to 258,769 vaccination-related tweets; classified as negative if negative word count exceeded positive (using Polyglot Python sentiment lexicon in 136 languages) - Public attitudes: Wellcome Global Monitor 2018 survey (140 countries) on statement "vaccines are safe" - Vaccination rates: WHO aggregate of top-10 most-reported vaccine doses (DTP3, MCV1, Pol3, Hib3, DTP1, HepB3, IPV1, BCG, MCV2, RCV1) - Controls: logged per capita GDP, expected years of schooling, internet penetration, region fixed effects

Specifications: Time-series OLS (Model 2, country fixed effects), cross-sectional OLS (Model 3), negative binomial count regression (Model 4 for tweet counts)

Results¶

Foreign disinformation on vaccination rates (Model 2, time series): One-point shift upward on 5-point disinformation scale → 1.93 percentage point drop in mean vaccination coverage year-over-year (p<0.01); robust across specifications
Social media organization and vaccine safety beliefs (Model 3): Strong positive association between public use of social media to organize action and percentage believing vaccines unsafe (β=1.437, p<0.05); effect stronger in younger age groups, lower education, lower income quintiles; no effect among self-identified secular individuals
Disinformation and negative vaccine discourse (Model 4, negative binomial): Foreign disinformation associated with 15% increase in negative vaccination tweets for median country (p<0.01)
Subgroup heterogeneity: Effect of social media organization on vaccine safety beliefs shows expected gradient across age, education, and income (larger effects in younger, less-educated, lower-income groups)

Connections¶

Vaccine hesitancy — focal topic; this paper provides large-scale empirical evidence for social media's role
Misinformation on social media — shared focus on online discourse; complements How Social Media Affects News Consumption methodologically
Disinformation and public health — Russian state actor role in amplifying anti-vaccine messaging; related to COVID-19 misinformation context
Related to Partisan Asymmetries in Media Consumption for Twitter methodology and sentiment analysis approaches

Notes¶

Strengths: - Uniquely global scope with 137–166 countries, moving beyond single-country case studies - Multiple data streams (sentiment-coded tweets, expert-coded political indicators, public polling, official vaccination statistics) strengthen causal inference potential - Timing is policy-relevant: written October 2020 anticipating global COVID-19 vaccine rollout - Clear practical implications for platform policy and diplomatic pressure on disinformation sources - Robustness across model specifications (with/without country fixed effects)

Weaknesses: - Foreign disinformation measure is generic (all foreign gov't disinformation, not specific to anti-vaccine content); no evidence other leading disinformation actors (China, Iran) conduct anti-vaccine campaigns - Twitter analysis limited by: (1) sentiment analysis using simple word-count polarity rather than nuanced NLP; (2) geocoded tweets represent ~1.5% of all tweets and skew toward countries with high Twitter usage; (3) non-representative of countries with low internet penetration - Vaccine safety attitudes only available as 2018 snapshot, precluding time-series analysis; 2008 vaccination rate used as proxy for pre-treatment attitudes (10-year lag is imperfect) - Cannot identify causal direction: does social media organization drive hesitancy, or do hesitant individuals self-select into organizing?

Follow-ups: - Extended to include Facebook data and post-COVID-19 vaccination campaigns - Longitudinal polling on vaccine attitudes to enable time-series analysis of disinformation impact - Deeper examination of which countries/actors drive anti-vaccine disinformation campaigns