Political Context of the European Vaccine Debate on Twitter

Giordano Paoletti, Lorenzo Dall'Amico, Kyriaki Kalimeri, Jacopo Lenti, Yelena Mejova, Daniela Paolotti, Michele Starnini, Michele TizzaniarXiv 2023 · DOI · arXiv

TL;DR

By analyzing 319 million tweets across 17 European countries, this study maps the intersection of political interest and COVID-19 vaccine hesitancy. The researchers found that exposure to vaccine-hesitant content peaked during early pandemic uncertainty and strongly aligned with right-wing, authoritarian, and anti-EU political leanings. Surprisingly, politicians themselves did not command outsized influence or engagement in these digital vaccine debates compared to structurally similar regular users.

What Problem Does This Paper Address?

Vaccine hesitancy is recognized as a top threat to global public health, a challenge that intensified dramatically with the rapid development and rollout of COVID-19 vaccines. While it is widely understood that public health issues can become deeply politicized, most social media research on this topic has been limited to English-only data or single-country case studies.

This paper addresses this gap by conducting a systematic, multi-country, longitudinal investigation across Europe. The authors tackle three core questions:

  1. Are individuals who endorse vaccine-hesitant content drawn to specific political parties?
  2. Does a user's overall level of politicization correlate with vaccine skepticism?
  3. Do political actors wield disproportionate influence in online vaccination debates?

Methods

The authors constructed country-specific retweet networks for four distinct time periods spanning from pre-COVID (late 2019) to the vaccine rollout (early 2021). To quantify a user's stance, they developed a pipeline to calculate a continuous Vaccine Hesitancy Endorsement (VHE) score ranging from 0 (strongly pro-vaccine) to 1 (strongly vaccine-hesitant).

The pipeline combines several key steps:

  • Label Propagation & Community Detection: A subset of tweets was manually labeled and propagated to retweeters. The authors applied spectral clustering on a regularized Laplacian matrix to identify network communities.
  • Network Perturbation: To ensure robustness against weak clustering artifacts, they ran 100 partial randomizations of the retweet networks, averaging the community scores for each user to yield a continuous VHE metric.
  • Ideological Regression: Ordinary Least Squares (OLS) regression models predicted users' VHE scores based on the political figures they follow. These political accounts were mapped to ideological dimensions (e.g., Left vs. Right, Liberty vs. Authority) using the ParlGov database.
  • Matched-Pair Influence Analysis: To evaluate the structural influence of politicians, the authors paired political accounts with non-political users matching in follower count, followees, and daily posting rates, comparing their impact via Wilcoxon signed-rank tests.

Datasets

The study leverages an extensive, multilingual data infrastructure:

  • Twitter Core Dataset: 319 million tweets collected via the Streaming API between October 2019 and March 2021, filtered using 459 vaccine-related keywords across 18 languages.
  • Annotated Benchmark: A manually annotated evaluation dataset comprising 5,667 tweets across 13 languages, classified as pro-vaccine, vaccine-hesitant, or neutral.
  • Political Mapping Data: Politician profiles identified via Politicians on Social Media and Twitter Parliament datasets, enriched with party metadata from the ParlGov database across 17 European nations.

Key Insights and Ideas

  • The Uncertainty Peak: Vaccine hesitancy endorsement was not static; it peaked globally during the early months of the pandemic (mid-2020). This suggests that digital vaccine skepticism thrives most in environments of high scientific uncertainty, before concrete clinical data and rollouts stabilize the narrative.
  • Ideological Sidedness: Online vaccine skepticism is deeply embedded in broader political worldviews. Across Europe, following right-wing, authoritarian, or anti-EU politicians consistently correlates with higher VHE scores, while left-leaning and liberal alignments correlate with pro-vaccine exposure.
  • The Myth of the Politician "Super-Spreader": Despite widespread concern over populist politicians driving anti-vaccine rhetoric, the matched-pair analysis revealed that politicians did not receive significantly more retweets or achieve higher network centrality than regular users with similar profiles. The political audience's pre-existing interests, rather than unique amplification by politicians, drive the echo chambers.

Key Results

  • VHE Distributions: Across most European nations, VHE scores skewed heavily pro-vaccine (closer to 0). Notable exceptions included France, which maintained a distribution centered around 0.5, and Poland, which exhibited a prominent vaccine-hesitant minority.
Figure 1. Vaccine Hesitancy Endorsement (VHE) score (see Algorithm 1) distribution across countries and periods. Dashed grey lines indicate scores where a user’s exposure has equal shares of vaccine-hesitant and pro-vaccine content. Score near 1 indicates more hesitant, and 0 – more pro-vaccine. Country abbreviations are ISO 3166 standard: AT Austria, BE Belgium, CZ Czech Republic, DK Denmark, FI Finland, FR France, DE Germany, GR Greece, IE Ireland, IT Italy, NL Netherlands, PL Poland, PT Portugal, SE Sweden, ES Spain, CH Switzerland, GB United Kingdom.
  • Party Affiliation Effects: In the OLS models, right-wing party families showed the strongest positive relationship with VHE, whereas Social Democracy and Liberal families consistently demonstrated a negative relationship.
Figure 2. Distribution (boxplots) of OLS coefficients modeling users’ Vaccine Hesitancy Endorsement (VSE) score (see Algorithm 1) by their interest in parties, grouped by families using ParlGov, an extensive resource on political parties in parliamentary democracies. Positive (negative) coefficients indicate that users who follow these parties are more (less) likely to engage with vaccine-hesitant rather than pro-vaccine content on Twitter. Accompanying points and whiskers indicate a 99% bootstrapped confidence interval. Numbers indicate how many parties are in each group. The parties identified as Right-wing have the strongest, and most positive, relationship with the VHE score.
  • Temporal Consistency: Within individual countries, the ideological sign of party coefficients rarely flipped over the four periods, showing that the political boundaries of the debate locked in early and remained stable.
Figure 3. Significant OLS coefficients (at p < 0.01 with Bonferroni correction) modeling the users’ VHE score using user interest in political parties, grouped in families using ParlGoV. Shaded areas indicate the 99% confidence intervals. Showing countries having sufficient model fit over 4 time periods. For the most part, we find the effect sign of a political party family to remain consistent over time.
  • Politician Influence: In 91% of the country/period experiments (40 out of 44), there was no statistically significant difference in the number of retweets received by politicians compared to their matched non-political counterparts.

Limitations and Future Work

The authors acknowledge several limitations inherent to social media studies:

  • Demographic Bias: Twitter users are more politically active and do not accurately represent the broader, offline national populations.
  • Keyword Constraints: The static keyword collection strategy may miss nuanced local phrasing or evolving topics (such as discussions surrounding "green passes").
  • Bot Activity: The pipeline does not explicitly filter out automated accounts or bots, which can artificially alter community structures.
  • Observational Nature: The data measures online exposure and endorsement (retweeting) but cannot establish causal links showing whether encountering this content actually changes a citizen's real-world medical decisions.

Future work could benefit from integrating these large-scale network metrics with longitudinal behavioral surveys to better capture the cognitive impact of digital political echo chambers.

Summary updated

Abstract

At the beginning of the COVID-19 pandemic, fears grew that making vaccination a political (instead of public health) issue may impact the efficacy of this life-saving intervention, spurring the spread of vaccine-hesitant content. In this study, we examine whether there is a relationship between the political interest of social media users and their exposure to vaccine-hesitant content on Twitter. We focus on 17 European countries using a multilingual, longitudinal dataset of tweets spanning the period before COVID, up to the vaccine roll-out. We find that, in most countries, users' endorsement of vaccine-hesitant content is the highest in the early months of the pandemic, around the time of greatest scientific uncertainty. Further, users who follow politicians from right-wing parties, and those associated with authoritarian or anti-EU stances are more likely to endorse vaccine-hesitant content, whereas those following left-wing politicians, more pro-EU or liberal parties, are less likely. Somewhat surprisingly, politicians did not play an outsized role in the vaccine debates of their countries, receiving a similar number of retweets as other similarly popular users. This systematic, multi-country, longitudinal investigation of the connection of politics with vaccine hesitancy has important implications for public health policy and communication.
BibTeX
@misc{paoletti2023political,
  author = {Giordano Paoletti and Lorenzo Dall'Amico and Kyriaki Kalimeri and Jacopo Lenti and Yelena Mejova and Daniela Paolotti and Michele Starnini and Michele Tizzani},
  title = {{Political Context of the European Vaccine Debate on Twitter}},
  year = {2023},
  eprint = {2309.03078},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2309.03078},
}

Loading related content...

Figure