Communication inequalities and incomplete data hinder understanding of how social media affect vaccine uptake
BMJ 2024; 385 doi: https://doi.org/10.1136/bmj-2023-076478 (Published 20 June 2024) Cite this as: BMJ 2024;385:e076478How are social media influencing vaccination? Read the full collection
- Kasisomayajula Viswanath, professor1 2,
- Edmund J Lee, assistant professor3,
- Eliza Dryer, PhD student1
- 1Harvard University, Boston, MA, USA
- 2Dana-Farber Cancer Institute, Boston, MA, USA
- 3Nanyang Technological University, Singapore
- Correspondence to: K Viswanath vish_viswanath{at}dfci.harvard.edu
The potential for exposure to misleading information on social media relating to vaccination is a matter of considerable concern in public, policy, and scientific arenas. However, we argue unpacking the causal nature of the relation between misinformation on social media and vaccination outcomes requires understanding of whether the available evidence base accounts for inequalities in communication. Once those inequalities are more fully understood, we can then examine the degree to which data absenteeism affects estimates of social media’s influence on vaccination and its potential for use in vaccination surveillance.
Gaps in evidence on social media’s effects on vaccination
The first challenge in understanding social media’s potential effects on vaccination related behaviour is measuring the audience’s exposure to anti-vaccine information. Several studies have documented the prevalence of misinformation about vaccines and their safety on social media.123 However, what proportion of social media users are actually exposed to this content is unclear, and assuming that such exposure always translates to engagement and behaviour is questionable. Furthermore, the studies that have documented misinformation on social media (web appendix) overwhelmingly focus on X (previously Twitter), with relatively few examining other platforms such as Facebook, Instagram, and WhatsApp, even though these are more widely used (fig 1). Platforms have different algorithms, different approaches to moderation (including different policies for removing or marking misinformation), and different user bases. Thus, extrapolating from documentation of anti-vaccine content on one social media platform to a generalization about widespread availability of anti-vaccine content on all social media is questionable.
Social media platforms in vaccines research over past six years (2017-23)
The next challenge is connecting the exposure to potential outcome. Few studies of the effects of social media on vaccination are designed to document causality, which would require data clearly identifying the timing of exposure and subsequent vaccination outcomes in the same individuals. Furthermore, isolating the effects of social media from those of other media and non-media sources and making causal attributions require experimental approaches such as randomized controlled trials, which are difficult given the threat of contamination (that is, the possibility of a control participant being incidentally exposed to anti-vaccine social media content). Even when experimental methods are used, the resulting data are often available only through partnership between select researchers and the platforms.4
Communication inequalities, social media, and vaccination
Challenges with understanding potential effects of social media are compounded by inequalities in communication. Such inequalities in access to, engagement with, processing of, and the ability to act on useful health information have been extensively documented along racial, socioeconomic, and geographic (for example, urban versus rural) lines.5 How these inequalities manifest in the context of social media and exposure to vaccine related content is worth critical analysis. For example, worldwide, social media usage is growing, with current estimates showing more than 4.7 billion users.6 Yet growth is not uniformly spread; people who are younger, more highly educated, or in urban areas are more likely to use social media than those who are older, less educated, or in rural areas (fig 2).7
Social media users by demographics in the United States. Survey of US adults conducted 25 January to 8 February 2021. White and Black adults include those who report being only one race and are not Hispanic; Hispanic adults are of any race. Not all numerical differences between groups shown are statistically significant (eg, no statistically significant difference exists between shares of white, Black, or Hispanic Americans who say they use Facebook). Respondents who did not give an answer are not shown. HS=high school. Reproduced with permission from Pew Research Center. Social Media Use in 2021. https://www.pewresearch.org/internet/2021/04/07/social-media-use-in-2021/
Similarly, reliable global data are hard to find, but some data sources suggest the presence of significantly more users of mobile phones, internet, and social media in high income countries than in low income countries.8 Within country variation is also extensive, with higher income or education groups having greater access than others.
Social and demographic factors implicated in communication inequalities could confound or modify observed associations between exposure to social media and vaccination behaviour (for example, educational attainment may be a “common cause” of social media exposure and attitudes towards vaccination; alternatively, the effects of exposure to social media on attitudes towards vaccination may vary across education levels). In addition to clouding efforts to identify effects, communication inequalities may hinder intervention. Because people who have greater access to digital services also have access to other reliable sources of health information, the benefits of any efforts to counter vaccine misinformation on social media through provision of more accurate, pro-vaccine information may accrue disproportionately to people who belong to privileged groups.
Data absenteeism, inequalities, and surveillance
A related challenge is data absenteeism, which results in an incomplete picture of social media and vaccination (box 1).910 Most studies of the effects of social media on vaccine behaviours and their antecedents are limited to English language media from high income countries. Of the 17 clinical trials published on this topic in the past six years that we identified from PubMed,11121314151617181920212223242526 10 were limited to participants based in the US; of these, only one enrolled participants who were not fluent in English (web appendix). A systematic review of social media’s influence on vaccination against human papillomavirus reported that most studies were conducted in the US (n=11); others were from Canada and South Korea (n=2 each) and from Taiwan, South Africa, Australia, the UK, and Romania (n=1 each).27 Another systematic review on social media interventions to promote vaccine acceptance noted a dearth of evidence from low income countries.28
Terminology
Data absenteeism
The absence of data from groups experiencing social vulnerability—whether by class, race or ethnicity, or geography—in sufficient quality and quantity
Web search queries
Web search queries are texts (ie, keywords, phrases, or sentences) that people enter into search engines in their informational search
Google Trends
Google Trends is a tool provided by Google that provides a snapshot of the popularity of certain keywords, subjects, or topics that have been searched over a certain time period and across geographic locations and languages
Geocoding
A process of identifying location with geographic coordinates
Social listening
The use of social media data to monitor public conversations online
Studies that seek to either characterise the social media content emerging from a particular geographic area or identify associations between such content and vaccine uptake in a given area almost always suffer from data absenteeism. Such studies usually rely on geocoded posts (box 1), which comprise a small proportion of overall content.29 This threatens generalisability and may bias findings, as some anti-vaccine posts are less likely to include location data than other posts.30
Data absenteeism also has implications for public health surveillance.313233 Using social media for data mining potentially has the advantage of gathering data from many users efficiently and quickly. Social media could serve as an efficient, unobtrusive, and rapid surveillance method for detecting users’ vaccine related sentiments and then facilitating timely interventions.34 “Social listening” is widely used but has limitations.35 Firstly, public health authorities need infrastructure, tools, and trained personnel to take advantage of social media data to survey and intervene. Poorer nations and smaller health departments may not have the capability to take advantage of this, another manifestation of communication inequalities at the organisational level. Secondly, equally critical, given digital inequalities, considering whose data are being mined and the groups for which inferences are being drawn is important. Drawing inferences on the basis of the data of people who are online, who tend to be better off and from high and middle income countries, constitutes “data chauvinism”—unreasonable confidence in data—leading to faulty generalisations.36
Discussion
A common belief is that social media are a site of struggle for setting and influencing public discourse on vaccines and that, when the public is exposed, the discourse is likely to influence their vaccination sentiments and behaviours to the detriment of public health. However, for public health practice and policy making, some significant gaps in evidence exist. From the perspective of public health practice, communication inequalities, data absenteeism, and lack of precision in information on who is exposed to which platforms with what effects places significant limitations on social media’s utility as both a tool for surveillance and an arena for interventions. This means that public health researchers need to be more thoughtful in using social media platforms and the “big data” generated from them. This analysis questions the current state of evidence, raises questions about the generalisability of the evidence, and argues against “broad brush” assumptions.
For instance, public health researchers could use qualitative approaches to investigate the media ecology of individuals, such as examining the kinds of social media content they are exposed to and why users believe certain anti-vaccine groups, and ultimately understand the underlying reasons why individuals or certain groups may be resistant to vaccines. In the context of older adults who were often assumed to be susceptible to misinformation, in-depth interviews examining their media use have found that they were able to identify misinformation accurately and may not be as vulnerable as they seemed.37 Also, research from qualitative studies has found that people who relied on multiple channels in their media ecology for their health information were more likely to identify misinformation; this suggests that reliance on different media ecologies may insulate one from being misled.38 These two examples are indicative that common assumptions derived from aggregated “big” data from social media may well be missing the mark if the context of individuals was not given due attention.
Secondly, although public health researchers could very quickly understand the prevalent interests or sentiments computationally from publicly available social media data, they should also pay attention to web search queries (for example, by using Google Trends), as individuals’ searches are indicative of the topics that people are paying attention to, and comparing what people say publicly on social media with what they are searching for privately may highlight potential gaps in knowledge or false beliefs held by people across different regions.39 Besides focusing on “conversations” on social media, public health researchers in partnership with social media companies could identify the types of misinformation gaining traction through interaction metrics and develop fact check labels to slow the amplification of misinformation. Otherwise, a cookie cutter approach of slapping on fact check labels may have a boomerang effect, as research has shown that repetitive exposure to messages tagged as “false” may increase belief in them.40 This may severely affect certain populations such as older adults, as repetitive exposure to messages (even if they are fact checked) may induce familiarity and susceptibility to false claims.41
The task of policy makers is even more difficult. The same problems—the extent of anti-vaccine content compared with other content, the sheer variety of platforms and data, and differential exposure and effects—make developing a “one size fits all” policy difficult. A policy is a blunt instrument that cannot account for nuances in exposure and effects. However, policy makers have another avenue to impel advances in our understanding of how social media drive vaccine related behaviours in at least two ways. They can compel social media platforms to:
Promote data as a public good by: tweaking their application programming interface requirements to allow free and easy access to researchers and public health authorities to track social media content on vaccines and make accurate estimates on vaccination related content and its effects, facilitating surveillance and intervention; instituting strong governance frameworks and oversight boards comprising government, non-profit entities, academia, and select members of the community for regulatory purposes; and setting up independent “data trusts” so that citizens (particular from low socioeconomic status groups) are confident in the fiduciary responsibilities governing their social media data.
Share results of health related research that the platforms are engaged in, allowing for evidence based practice.
In examining the role of social media in vaccination and improving public health, policy makers and researchers need to straddle a fine line between overstating the claims about the impact of social media (if due consideration had not been given to underserved populations) on society and underestimating the need for understanding the unique challenges and media ecology of people from low socioeconomic status groups.
Key messages
Concern is increasing about the erosion of confidence in vaccines among public health practitioners, scholars, and policy makers; social media have been implicated in contributing to this erosion
The generalisability of evidence on the role of social media is affected by conceptual and measurement limitations
Communication inequalities and lack of data from non-English speaking countries argue against “broad brush” assumptions
Acknowledgments
KV’s work is supported by the National Cancer Institute/National Institutes of Health (5U54 CA156732-10, 5 P30 CA06516, R01 CA230355-01A), Bill and Melinda Gates Foundation, Dana-Farber Cancer Institute, Dana-Farber/Harvard Cancer Center, and Lee Kum Sheung Center for Health and Happiness, Harvard TH Chan School of Public Health.
Footnotes
Contributors and sources: KV is Lee Kum Kee professor of health communication at Harvard T.H. Chan School of Public Health and professor of population sciences at Dana-Farber Cancer Institute; his research interests are in examining how evidence in health and science is translated to influence policy and practices through different communication platforms with specific focus on equity; he has published extensively on social media and wellbeing and vaccine communications. EJL is an assistant professor at the Wee Kim Wee School of Communication and Information at Nanyang Technological University, Singapore, and assistant director at the Centre for Information Integrity and the Internet (IN-cube); his research focuses on developing health technologies to tackle health inequalities and how to take advantage of digital traces—data from social media, smartphones, wearables, and electronic health records—in an intelligent and ethical manner to understand and improve public health outcomes. ED is a PhD student at Harvard TH Chan School of Public Health; her research interests include communication inequalities and organizational health literacy. All authors contributed equally to the writing of this article. KV is the guarantor.
Competing interests: We have read and understood The BMJ policy on declaration of interests and have no interests to declare.
Provenance and peer review: Commissioned; externally peer reviewed.
The article is part of collection that was proposed by the Advancing Health Online Initiative (AHO), a consortium of partners including Meta and MSD, and several non-profit collaborators (https://www.bmj.com/social-media-influencing-vaccination). Research articles were submitted following invitations by The BMJ and associated BMJ journals, after consideration by an internal BMJ committee. Non-research articles were independently commissioned by The BMJ with advice from Sander van der Linden, Alison Buttenheim, Briony Swire-Thompson, and Charles Shey Wiysonge. Peer review, editing, and decisions to publish articles were carried out by the respective BMJ journals. Emma Veitch was the editor for this collection.
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.