Why are people vaccine hesitant?

Vaccine hesitancy is heterogenous and context-dependent in a variety of ways. Firstly, it is important to note that while the current literature on vaccine hesitancy is focused on individuals making decisions about their own vaccination behaviors, pre-COVID vaccine hesitancy research largely focused on pediatric decisions, i.e. parents making decisions about their children [8]. While these two constructs clearly differ, research has shown that there is a significant relationship between vaccine hesitancy in the two contexts. For example, Soares and colleagues showed an association between refusal to take the flu vaccine and intention to refuse the COVID-19 vaccine [9, 10]. Secondly, it is worth noting that the exact beliefs and attitudes that underlie vaccine hesitancy shift between different diseases and even between different vaccines targeting the same disease. For example, Brewer and colleagues [11] found that parents have different levels of hesitancy and reasons for being hesitant depending on the type of vaccine concerned (in their work, MMR versus HPV). Furthermore, work by Merkley & Loewen [12] involving 6200 participants showed even within vaccines for a single disease (COVID-19), the same vaccine elicited differential levels and sources of hesitancy. Thirdly, even in cases where research is focused on a single vaccine targeting a single disease, vaccine hesitancy may not mean the same thing from one participant to another. For example, Noni MacDonald [3] has proposed that vaccine hesitancy can better be understood along a continuum that contains various attitudes such as “refuse all with conviction”, “refuse all, but unsure”, “accept some, refuse some, delay vaccination”, “accept with doubts & concerns” and “accept all with confidence”. Finally, even when people are vaccine acceptant, it may be for different reasons, unrelated to the vaccine itself (e.g., confidence in vaccine efficacy). Alfano and colleagues [13] showed that positive attitudes toward vaccines can be for individualistic or altruistic reasons.

Given how complex vaccine hesitancy is, a large amount of past research has focused on understanding its antecedents, including individual differences in vaccine hesitancy such as demographic correlates. For example, in the context of COVID-19, being a woman, being 50 years or younger, being single, being unemployed, and living in a household with five or more people, were factors associated with increased vaccine hesitancy [14]. While findings like these allow us to understand which groups might present the biggest challenge for vaccination in a public health context, they do not necessarily provide an effective strategy for targeting those groups because they shed little light on the motivating factors underlying vaccine attitude formation. For this reason (among others), recent research has started to tackle the psychological correlates of vaccine hesitancy [11, 15, 16]. For example, Betsch and colleagues [15] showed, from a group of 1000 German participants, that confidence, complacency, constraints, calculation, and collective responsibility are key psychological antecedents of vaccination. Hornsey and colleagues [16] used a sample of over 5000 people to show strong associations between vaccine hesitancy and conspiratorial thinking, reactance, disgust towards blood and needles, and individualistic world views.

Although empirical research is invaluable in allowing us to gain a better understanding of factors related to vaccine hesitancy, previous literature on behavior change [17, 18] has shown that understanding a decision environment in a holistic and mechanistic manner is beneficial to engendering behavior change. In recognition of this, a number of models have been proposed to explain causal factors related to vaccine hesitancy—some specific to vaccine hesitancy and some broadly related to uptake of health behaviors, but all applicable to vaccine uptake. These models, which we review below, have an advantage over questionnaire-style empirical reports of factors related to vaccine hesitancy in that they base themselves on large bodies of empirical findings and thus purport to map the terrain in order to create more evidence-based directions for future hypotheses and interventions. In an effort to better understand this terrain, we review existing models related to vaccine hesitancy and then use these models to create a search space for a thematic analysis of qualitative data from pandemic-era focus groups.

Frameworks focused on health behaviors

Taxonomies of uncertainty

Another set of frameworks used to explain health related behaviors comes from the study of uncertainty. The link between uncertainty (i.e., the subjective feeling of ignorance about something) and vaccine hesitancy has been well documented. In fact, past studies have shown that uncertainty is one of the most critical factors associated with vaccine hesitancy [24, 25]. In a study that looked at COVID-19 vaccination efforts, Courbage and Peter [26] found that reducing uncertainty about the vaccine reliably promotes vaccination.

Vaccine decisions must often take place in an inherently uncertain environment with shifting and often conflicting recommendations. This has been all the more true during the COVID-19 pandemic, where the time horizon of recommendation changes was sometimes weeks or days [27]. Past research suggests that this type of scientific uncertainty creates distrust in scientists and public health recommendations [28], which in turn has been shown to result in low confidence in vaccines [29] and lower vaccine uptake [30]. Indeed, perceptions of uncertainty affect trust toward all disease-related information [31], particularly in individuals with lower tolerance for ambiguity and risk. Uncertainty has also been shown to make people more pessimistic about disease treatment [32] and perhaps most importantly, lower intentions to engage in health-promoting behaviors [33]. A number of efforts have been made to taxonomize the types of uncertainty relevant to healthcare decisions in an effort to better understand and support patient experience with the overall goal of improving decision making and clinical outcomes.

Varieties of uncertainty in healthcare.

Han and colleagues [34] proposed a taxonomy (Table 1) that separates uncertainty according to source and issue. Within sources of uncertainty lie ambiguity, complexity, and probability. Furthermore, within issues of uncertainty lie scientific, practical, and personal issues, each of which have been broken down into two or more categories.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Han et al.’s (2011) suggested taxonomy of uncertainty.

https://doi.org/10.1371/journal.pone.0295912.t001

Model of Uncertainty within Complex Healthcare Settings.

In response to the above proposed taxonomy, Pomare and colleagues [35] posited that uncertainties do not occur in isolation, as they argued the above taxonomy implies. The research team examined the empirical healthcare literature to see whether or not Han and colleagues’ taxonomy is applicable to the wide range of sources of uncertainty experienced by healthcare providers in the healthcare space. After examining the 94 articles that passed their criteria, their research revealed that Han and colleagues’ taxonomy [34] would have benefitted from two new sources of uncertainty: systems uncertainty and ethical uncertainty. Furthermore, Pomare and colleagues’ new model highlights the overlapping nature of uncertainty which they argue is a necessity and more realistic than disparate and unrelated sources of uncertainty. The taxonomy they suggest (Table 2), Model of Uncertainty within Complex Healthcare Settings (MUCH-S) splits healthcare uncertainty into three main (and overlapping) categories:

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Pomare et al.’s (2018) suggested taxonomy of uncertainty.

https://doi.org/10.1371/journal.pone.0295912.t002

Data collection methods

Research was carried out in collaboration with Environics Research (https://environics.ca/), a Canadian polling and research firm, for the purpose of data collection. Participants were recruited using a purposive sampling strategy between June 23rd, 2021 and August 4th, 2021 based on their self-identified vaccination status and attitudes in a national survey of COVID-19 experiences. Survey data, which we do not report on in this study, was collected from 1541 Canadians in April and May, 2021, during which time participants were asked if they were interested in participating in follow-up focus groups. Among the 758 individuals who agreed (49%), we selected 538 profiles specifically fitting a sampling strategy selected to include a balanced urban/rural split, a wide range of ages, ethnicities, geographical locations and income ranges. We contacted them by e-mail and 18 individuals agreed to participate in the focus groups. Data was collected in focus groups that were carried out virtually between September 7th and September 9th, 2021. Given that COVID-19 vaccines became available in Canada in December 2020, and were widely available by Summer 2021, focus groups were carried out at a time where vaccination decisions were top-of-mind for participants. A total of four focus group sessions were conducted, each lasting between 60 and 67 minutes with a mean length of 64.

The four focus groups consisted of participants with different vaccination statuses. Group 1 participants were unvaccinated individuals who indicated that they would either not get vaccinated at all or would wait a while before deciding. Group 2 consisted of vaccinated individuals who were initially hesitant about receiving a COVID-19 vaccine but ultimately chose to get vaccinated. Groups 3 and 4 consisted of young adults with mixed vaccination statuses.

The four groups included 18 adults, consisting of 13 women and 5 men. Table 3 illustrates the primary sociodemographic characteristics of the participants, including their vaccination status at the time of the focus groups. The age of the participants ranged from 26 to 68, with an average age of 44. All participants were born in Canada and identified themselves as either white (67%) or Indigenous (33%). They were from six provinces across eastern, central, and western Canada, with the majority residing in Ontario. The participants were geographically dispersed, with some living in rural areas and some in urban areas. Ten of the female participants were mothers, while none of the male participants had children. Half of the participants were low-income, and most had at least a high school degree. To maintain confidentiality, pseudonyms were assigned to the participants. Focus groups were moderated by two researchers at Environics Research: a male researcher, who has nearly 15 years experience in the healthcare industry and market research and a female researcher with over six years experience conducting qualitative research, including in virtual platforms. Two researchers from the university investigator team were also present during focus group sessions but did not moderate. Participants were asked to sign a consent form which clearly indicated the purpose of the study and were told the identities of the interviewers. The focus groups were conducted using the guide available in the S3 Appendix, which was designed by the authors, and a recording was made, which was later transcribed. No repeat interviews were carried out and transcripts were not returned to participants for correction.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Sociodemographic characteristics and vaccination status of focus group participants.

https://doi.org/10.1371/journal.pone.0295912.t003

Data coding & analysis

Focus group recordings were transcribed verbatim. MAXQDA Analytics, a qualitative analysis software, was used to aid in the analysis of focus group transcripts. The software was in a manual capacity—i.e. no automated features were used to generate codes or thematic categories. Using MAXQDA enabled us to manually organize the transcript into the codes using the process detailed below and provided visualizations of critical elements using various built-in tools.

The study utilized an inductive approach to conduct a thematic analysis (Braun & Clarke, 2006), a process where researchers actively engage in the knowledge production process and consider their subjectivity as a valuable resource during the coding and theme development phases. This approach allowed for a systematic way of processing qualitative information using coded text and enabled us to provide a detailed and nuanced analysis of participants’ attitudes and behaviors toward COVID-19 vaccines. The steps we took, proposed by Braun and Clarke [38], are data familiarization, code formulation, generation of themes, themes review, defining and naming themes, and report formation. A single coder, XL, was used throughout the process to ensure consistency in the coding, with other members of the research team providing guidance on the coding tree. The coder has a professional and academic background in public health and qualitative research, specifically focused on vaccine hesitancy. Participants were not asked to provide feedback on the findings.

1. Data cleaning & familiarization: We transcribed the data based on audio recordings, and imported the transcripts into MAXQDA, then completed the necessary pre-processing steps: removing identifying information & formatting the data for analysis. We read and re-read the transcripts end-to-end in order to familiarize ourselves with the contents and generate initial ideas for a search space.
2. Code formulation: Using the generated ideas as well as past models used in vaccine hesitancy as a starting point for the search space, a coder read through all of the transcripts, coding representing sections related to vaccine hesitancy in a systematic fashion across the entire dataset–both sections related to the search space and novel themes were identified, with novel themes focused on expressions of hesitancy around vaccination decisions. Memos were written down to keep track of the condensed information. A codebook was developed based on this process, and then applied to the data using MAXQDA’s coding function. The data were then segmented into coded units based on the codes applied.
3. Generation of themes: The coder reviewed the coded data and refined the codes as necessary. New codes were added, and existing codes were merged or split based on the patterns that emerged from the data. The cycle was repeated several times to narrow down the number of codes and categorize them into identifiable themes. The codes were then analyzed and grouped into three central themes as stated in the results section.
4. Themes review: The complete interview data were re-read to validate that the themes were gathered in an accurate and representative way. MAXQDA was used to identify patterns within the data, and these were used to draw conclusions.
5. Defining and naming themes: The reviewed themes were then conceptualized and assigned clear definitions in the S1 Appendix, and memos were finalized within MAXQDA to ensure data transparency and reproducibility.
6. Report formation: several vital statements and features representing the data were extracted to showcase the resulting outcomes both as statements in the form of ideas and feelings and visual representations using interconnections between codes as seen in the S2 Appendix.

In order to ensure rigor in our qualitative research process and validation of our findings, we used an iterative thematic analysis process based on established analytical tools [38]. Furthermore, we recruited a diverse sample of individuals and maintained records of our processes, creating a MAXQDA codebook. Finally, we engaged in self-reflection throughout the process to better understand personal biases and report on potential sources of these biases in the limitations section.

Personal

Scientific

Practical

Proposed taxonomy

Our analysis of past frameworks suggests that, while they are useful in a non-pandemic context, a number of important gaps exist that limit their usefulness in a COVID-19 context and potentially in the context of future vaccine-preventable pandemics where high uncertainty vaccination decisions occur. Given the demonstrated effectiveness of targeted messaging in increasing vaccine uptake, overcoming these limitations is particularly important in order to inform the design of effective public health messaging. In an effort to address the gaps in existing models, we propose a novel taxonomy of vaccine hesitancy that is grounded in our focus group findings, is specific to COVID-19, and is potentially applicable to future vaccine-preventable pandemic contexts. This taxonomy is based on the themes that emerged from our analysis of the focus group data. This means that it differs in approach from past taxonomies, a large part of which were constructed from literature reviews of determinants of vaccine hesitancy. Given that the studies included in these literature reviews did not necessarily aim to uncover the entire landscape of vaccine hesitancy determinants but rather focused on specific hypotheses, our taxonomy represents a more naturalistic approach that is unconstrained by prior hypotheses. Our taxonomy—which we refer to as the High Uncertainty Vaccination Decision (HUVD) taxonomy—is presented in Table 5.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. The High Uncertainty Vaccination Decision (HUVD) taxonomy (full definitions of themes and sub-themes provided in the S2 Appendix).

https://doi.org/10.1371/journal.pone.0295912.t005

Notably, we found that focus group participants had strong feelings about their right to choose whether or not to be vaccinated. With these feelings, came a resistance to what is perceived as external pressure and restrictions tied to vaccination. However, while institutional pressures were expected to backfire, likely driven by low levels of trust, the influence of social circles was more effective in promoting vaccination. With regard to the development of the vaccines, even non-hesitant individuals felt a large degree of uncertainty and confusion about effectiveness and potential side effects, suggesting that the unique timeline in which the COVID-19 vaccines were developed may have contributed to perceived risk.

Comparison to past taxonomies

Our proposed HUVD taxonomy focuses on determinants of vaccine hesitancy in a COVID-19 context. Therefore, it differs in a number of ways from other models we have summarized in Table 6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. The High Uncertainty Vaccination Decision (HUVD) taxonomy compared to previous models used in vaccine hesitancy.

https://doi.org/10.1371/journal.pone.0295912.t006

In an effort to better understand the differences and similarities between our proposed taxonomy and previous taxonomies, we created the following comparison, which aims to illustrate which categories in our taxonomy might be most relevant for each category of previous taxonomies. One notable difference between our proposed taxonomy and existing taxonomies is that previous taxonomies have a much greater focus on various aspects of what we call “systems of care”—i.e. on the actual mechanics of delivering the vaccine. While systems of care are also critical in the context of COVID-19, participants in our focus groups were far more likely to bring up factors relating to uncertainty, trust and institutional pressures. These factors, which reflect the high uncertainty environment of the COVID-19 pandemic and vaccine development lifecycle, are likely to also be more relevant in future contexts where vaccine-preventable diseases emerge and require fast action. In Table 7, we further illustrate the differences between our proposed taxonomy and past taxonomies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. A mapping of categories in the HUVD taxonomy to those of previous models.

https://doi.org/10.1371/journal.pone.0295912.t007

Importantly, previous models focus on particular aspects of what our data showed to be relevant in a COVID-19 vaccine hesitancy context, but none of them cover all of the categories we identified. For example, the HBM focuses on individual perceptions and motivations but does not tie them to the social environment. The TPB connects better to social environments but does not capture (or too broadly captures) the scientific and environmental uncertainty that are inherent in high uncertainty vaccination decisions. The uncertainty taxonomies from Han and colleagues and Pomare and colleagues capture a wider breadth of categories but in a way that is too far removed from a vaccine context. Finally, the 3C, 5A and 5C models are more narrowly focused but tend to be more granular around systems of care (which our data suggests is not a significant driver) and less granular around confidence and acceptance, which limits the usefulness of these categories.

Overall the differences between our HUVD taxonomy and past taxonomies are reflections of the unique decision environment of the COVID-19 pandemic. A number of factors have come together to create this environment, such as: the unprecedented speed of vaccine development; novel vaccine technologies; global scale of the illness; large amounts of media attention; large amounts of misinformation and disinformation; political polarization; unpredictable evolution of the disease and variants of concern; lockdowns and restrictions; debate about natural immunity induced by infection. Given factors like these, it is unsurprising that a taxonomy specifically adapted to a COVID-19 vaccination decision context differs substantially from previous ones. Furthermore, these factors, which are reflected in our data, contribute to a decision environment where people who are normally vaccine acceptant might refuse or, at the very least, delay a COVID-19 vaccine.

Implications of results

We believe our HUVD taxonomy provides a valuable etension of existing models to a COVID-19 context, with possible relevance to future pandemic contexts with high uncertainty vaccination decisions. Past research has shown that public health messaging that targets individuals in a personalized and stage-specific manner can improve a variety of health outcomes, such as smoking cessation [39] and vaccine uptake [40, 41]. However, successful tailoring of messages relies on having a clear understanding of the decision factors present in the target audience. Our proposed taxonomy provides a similar tool to what has been created in the past, but specific to COVID-19 vaccine hesitancy and potentially other contexts with high uncertainty vaccination decisions. Such a tool can be used to bolster existing public health strategies around COVID-19 but could also be informative for future research into factors related to vaccine hesitancy in a pandemic context where high uncertainty vaccination decisions take place.

Limitations

One major limitation of this work is that our taxonomy, while informed by qualitative data, has not yet been empirically tested in an applied setting. In addition, one must keep in mind that COVID-19 presented a unique context that may or may not generalize to other types of high uncertainty vaccination decisions–whether in a future pandemic or in other contexts. Nonetheless our taxonomy provides a useful starting point for developing public health messaging hypotheses to be tested in the particular context where they are to be applied. Additionally, while public health messaging designers can use our taxonomy to inform hypotheses they have about the types of messaging that may be most effective, they should not assume that a particular source of hesitancy is relevant without further empirical evidence.

A further limitation is the use of purposive sampling, whereby potential participants were selected based on their self-identified vaccination status and attitudes. While this approach allows us to recruit a relevant sample, it introduces the potential for bias due to self-selection by people with stronger opinions who are interested in participating in focus groups therefore potentially excluding more moderate voices. Related to this, focus group dynamics can potentially introduce bias due to groupthink. While the focus group moderators have extensive training and experience conducting qualitative research, this potential source of bias remains. A further limitation related to the focus groups is that because of the unique situation and limited sample size, we were unable to take steps to ensure data saturation.

A limitation related to the analysis of the data is the background of the qualitative data coder. While this coder has an educational and professional background in public health and qualitative analysis and vaccine hesitancy more specifically, any coder will bring a certain level of bias which must be recognized. Given the coder’s background, this bias could plausibly result in some level of confirmation bias–which we have attempted to address by carefully reviewing the raw data provided in the S2 Appendix.

Another limitation of the present study is the skew toward female and white participants in our sample. In an effort to select a sample that contains diversity in geography, educational status, income, and age, as well as a large representation of Indigenous communities, an unintended outcome was a final sample that contained an over-representation of women and an under-representation of ethnic groups other than persons of Indigenous ancestry. Finally, as noted in the Methods section, none of the male participants had children. Given that a significant amount of vaccine hesitancy attitudes and vaccination decision experiences happen in the context of childhood vaccines, this may have limited the representativeness of our sample.

Recommendations for future research

Future work looking to expand on this taxonomy should focus on understanding how generalizable it is to populations in other jurisdictions and vaccine contexts. It should also make efforts to understand how our findings may be generalized to future pandemic contexts where high uncertainty vaccination decisions are inherent. In addition, we believe that work attempting to connect the taxonomy proposed here with empirical data from communication strategies will be useful in helping us understand the extent to which the taxonomy might be relevant in a real-world context, such as public health message design.