Social Media Vaccine Trust Rose Then Fell During Pandemic

A new study tracking a decade of vaccine discussions on social media reveals a dramatic emotional rollercoaster during the COVID-19 pandemic. Researchers analyzed 18.7 million English-language posts on X (formerly Twitter) from 2013 to 2022, finding that public sentiment toward vaccines first surged with optimism and trust when COVID-19 vaccines were developed, then plummeted into skepticism and negativity as the pandemic progressed. This comprehensive analysis provides unprecedented insight into how major health crises reshape public discourse, with for how health organizations communicate during future emergencies.

The research team from Kansas State University, University of Illinois, and the National Research Council Canada discovered that emotional language around vaccines underwent significant shifts during the pandemic years. While negative emotion word usage decreased overall by 13.22% during COVID-19 compared to pre-pandemic years, this masked a more complex pattern. The study found statistically significant increases in words associated with surprise (up 21.66%) and trust (up 10.64%), while emotions like disgust decreased by 27.81% and fear dropped by 23.89%. These changes reflect the initial hope surrounding vaccine development followed by evolving public reactions as vaccines became widely available.

To conduct their analysis, the researchers collected 129 million posts using vaccine-related keywords, then filtered them down to 18.7 million high-quality posts through rigorous preprocessing. They employed established psychological frameworks, using the NRC Emotion Lexicon and Words of Warmth Lexicon to analyze emotional content through the Utterance Emotion Dynamics metric framework. For stance detection, they used Meta's Llama 3.3 model with 70 billion parameters to classify posts as favoring vaccines, opposing them, or neutral, achieving 66.32% accuracy compared to human annotations on a sample of 500 posts. ology focused on linguistic content rather than engagement metrics, ensuring directly reflected how vaccines were framed in public discourse.

The data reveals clear patterns in how vaccine discourse evolved. Figure 3 shows that during pre-COVID-19 years, negative word usage varied more significantly than positive word usage, with English-speaking X users generally using more negative words. However, during early 2021, negative word usage dropped significantly below positive word usage when vaccines were being developed and released. From late 2021 onward, negative word usage increased rapidly while positive word usage declined. Figure 4 illustrates how fear started high in early 2020, dropped significantly at the beginning of 2021, then gradually increased again in late 2021 and throughout 2022.

From a social cognition perspective, the discourse shifted notably. Figure 5 shows that pre-COVID-19 discussions about vaccine effectiveness were relatively stable, while during COVID-19 years, there was marked variability around competence. Figure 6A demonstrates that warmth word usage increased during the COVID-19 period from 2020 to early 2021 before decreasing consistently until the end of 2022, with similar trends in trust and sociability components. Figure 6B shows competence increased during the first year of COVID-19 before decreasing consistently after early 2020. These shifts suggest public trust in vaccines increased during the early pandemic stages but declined once vaccines became readily available.

The study also tracked changing stances toward vaccines over time. Figure 7 shows the estimated proportion of posts favoring vaccines had a slight increasing trend between 2013 and 2020, then experienced substantial decreases and increases during the pandemic years. The percentage of posts against vaccines showed a slight increasing trend from 2013 until about 2020, declined in late 2020, then sharply increased from 2021 onward. By late 2022, the LLM-based estimates suggest posts against vaccines surpassed those in favor. Table 4 reveals that among posts favoring vaccines, emotion word density associated with low warmth decreased by 6.4% during COVID-19, while anti-vaccine posts showed a 1.3% increase in low warmth emotion word density.

These have significant for public health communication. The research demonstrates that vaccine discourse became more emotionally charged during the pandemic, with early optimism giving way to polarization. The dataset itself represents a valuable resource for future research, enabling examination of public vaccine discourse spanning over a decade. The patterns observed suggest health organizations need to anticipate emotional shifts in public response during health emergencies and develop communication strategies that address both trust and competence perceptions as situations evolve.

The study has several important limitations. It focuses exclusively on English posts on X from 2013 to 2022, so conclusions apply to this dataset rather than the general population. X users tend to be younger and more technologically savvy than average, and most posts aren't geo-tagged, limiting regional analysis. The research doesn't incorporate social media engagement metrics like likes or comments, which could provide additional insights into content diffusion. Additionally, the stance detection using LLMs, while reasonable for identifying broad trends, shouldn't be used to draw conclusions about individual posters due to nuances in language use including sarcasm and humor. Future research could address these limitations by incorporating data from multiple platforms, analyzing non-English content, and extending the timeframe beyond 2022.