Fear over Friends: Examining the Perceived Influence of Others on Vaccination Decisions

1 Introduction↩︎

One thing learned from COVID-19 is that the spread of a pandemic in nations around the world depended on not only on accurate epidemiological information and government responses [1][8] but also the varied behaviors of individuals, groups, and cultures [9][14]. Since culture is the context for behavior [12], [15], and cultural values vary substantially across the world [2], [16], [17], the effect of COVID-19 in terms of cases and deaths, was affected not only by government actions but by cultural values in their populations [14].

These varied behaviors have been well studied, particularly early in the COVID-19 pandemic, before vaccines were widely available. These studies have used socioeconomic and public health variables to explain COVID-19 variation within the United States [18], [19] and also globally [20]. In the context of the literature on the effect of socioeconomic factors on COVID-19 [2][12], cultural effects have been examined in concert with known risks such as obesity and advanced age, together with variables describing government efficiency and public trust in institutions [14], [19].

In terms of the behavioral dynamics of vaccination, those who are not vaccinated may feel less urgency due to a lowered perceived risk of infection [21][23]. Furthermore, even those infected with a coronavirus or flu, may not exhibit obvious symptoms to others [24] and inadequate testing of populations [6] can mean that the majority of infections in a population are undocumented [6], [24][27]. This can lead individuals to underestimate the risk of a virus, especially in lieu of more overt socio-economic concerns in people’s daily lives [28], [29]. Social influence from peers may be weak becauseunlike visible protections such as mask-wearing or conspicuous lack of people in public spacesvaccination is often comparatively less visible. In this situation, the perceived benefits of vaccination may be outweighed by economic and/or psychological costs [9], [29][31].

The complexity of the levels of voluntary vaccination lies in the multiple drivers of behavioral change, including information and social learning [10], [11], [32][35]. Ideally, decisions would be determined by their intrinsic payoffs within their socio-ecological environment [36]. In the real world, decisions are made by people who combine observational learning, which produces noisy information, and social learning, which diffuses that information to others [37], [38]. The transparency of learning is a crucial parameter [39]; the less transparent the payoffs are, the more noise and heterogeneity enters into the behavioral dynamics [40][42]. While humans are motivated to avoid cues associated with pathogens, through emotions such as fear or disgust [43], COVID-19 and similar viruses display few cues in asymptomatic cases [44].

These considerations suggest that drivers of social distancing include two dynamic factors: observable risks and observable behaviors. In everyday life, we may expect more intimate social influence, such as advice from parents or other close family members, to be the strongest for vaccinations. Vaccination can be made more visible of course through public outreach, but its acceptance will vary based on group norms and potentially large differences in risk perception due to misinformation. Acceptance may also vary based on a range of cultural memory effects, such as the past misuse of vaccination on marginalized groups. In contrast, if genuine benefits of vaccination become more transparent to the public, individual cost-benefit decisions can support the behavior spreading through social influence [11], [32], [45][47].

Here we examine one facet of this complex problem: the perceived social influence of others on vaccination decisions. While prior work has shown that social influence matters in an array of health-related decisions, including social distancing and mask wearing during the COVID-19 pandemic [48], examining social influence on COVID-19 vaccine decisions has had comparably less study, particularly across strong and weak social ties. Hence, our work seeks to fill this gap and add robustness to the current findings in the literature. More broadly, further work in this area adds to our understanding of the interactions between socio-cultural dynamics and health-related decision making.

Our four research questions are as follows:

  • RQ1: Does the perception of social influence on vaccination decisions change across vaccinated and unvaccinated populations?

  • RQ2: Does the perception of social agreement with vaccination decisions change across vaccinated and unvaccinated populations?

  • RQ3: Does the perceived danger of COVID-19 to others change across vaccinated and unvaccinated populations?

  • RQ4: If not by social influence, what other reasons do vaccinated and unvaccinated populations think influence their vaccination decision?

2 Methods↩︎

To answer our core research questions, we designed and deployed a survey on the online survey platform Prolific (prolific.co). The survey contained questions on demographics, perceived influence of varying social circles, perceived agreement of varying social circles, perceived danger of COVID-19 to varying social circles, and an open-ended reasoning question.

On May 1st 2021, we deployed a pilot survey for 15 participants to measure if our estimated survey time and payment amount were correct. There were no issues found during the pilot, so the same payment parameters and questions were used in two more batches of survey deployment on May 2nd 2021. The pilot survey was given to any Prolific user in the United States, whether they had been vaccinated for COVID-19 or not. However, to ensure balance of vaccinated and unvaccinated participants, we deployed the actual survey to two pre-screened groups provided by Prolific: participants who were vaccinated for COVID-19 and participants who were unvaccinated for COVID-19. Using these two panels of participants, we deploy the survey for approximately 500 participants in each group. According to Prolific, 1,306 participants were eligible to take our survey in the unvaccinated U.S. group and 8,569 participants were eligible to take our survey in the vaccinated U.S. group.

The median time per participant in the pilot study was just over 3 minutes, in the unvaccinated group study it was 4 minutes and in the vaccinated group study it was 3 minutes. All participants lived in the United States and were paid $1.00 for survey completion. This survey was approved by The University of Tennessee’s IRB.

We provide details about each type of question below.

First, we collected several types of demographic information for comparison to previous studies on COVID-19 vaccination and related behaviors. These demographic questions included:

  • Political leaning - We asked “What is your political leaning?” . Participants could answer on a 5-point scale: very liberal, liberal, moderate, conservative, or very conservative.

  • Rural-urban residency - We asked participants to provide the U.S. zip code in which they currently reside. This question allowed us to map participants to Rural-Urban Continuum Codes from the U.S. Department of Agriculture1. Rural-Urban Continuum Codes range from 1 - counties in metro areas of 1 million population or more - to 9 - counties that are completely rural or less than 2,500 urban population, not adjacent to a metro area.

  • Rural-urban identity - Following the questioning used in [49], we asked “Regardless of where you currently live, do you usually think of yourself as a city person, a suburban person, a small-town person, a rural or country person, or something else?”, giving us a 4-point scale from country person to city person with something else as an outlier. According to [49], rural social identification - “a psychological attachment to being from a rural area or small town” - is different than “simply living in a rural area”. Lunz Trujillo (2022) demonstrated this by showing that rural social identification predicts greater anti-intellectualism than rural residency alone. Hence, in our study we wanted to capture this psychological trait in addition to where one lives.

  • Education - We asked “What is your level of education?”. Participants could answer on a 7-point scale: less than high school, high school graduate, some college, 2 year degree, 4 year degree, professional degree, doctorate.

  • Income - We asked “What is your approximate household income?”. Participants could answer on a 12-point scale: less than $10,000 to more than $150,000.

  • Gender - We asked “How do you describe yourself?”. Participants could answer with one of four answers: male, female, non-binary/third gender, prefer not to say.

  • Race - We asked “What is your racial ethnicity?”. Participants could answer with one of seven answers: white, black, Asian, Hispanic or Latino, native American or Alaskan, Native Hawaiian or Pacific Islander, or other

Next, we asked participants “How much influence did each social group have on your decision to get vaccinated?” or ” How much influence did each social group have on your decision to not get vaccinated?” as a matrix Likert Scale question across five groups: family members, close friends, co-workers, church or social club, and neighbors or community members. Participants could answer on a 5-point scale: none at all, a little, a moderate amount, a lot, or a great deal.

To add context to our question about social influence, we also asked participants about social agreement with their vaccination decision and their perceived danger of COVID-19 to others. Specifically, we asked participants “How much do you think each social group agrees with your decision to get vaccinated?” or “How much do you think each social group agrees with your decision to not get vaccinated?” as a matrix Likert Scale question across five groups: family members, close friends, co-workers, church or social club, and neighbors or community members. Participants could answer on a 3-point scale: disagree, do not care, or agree. We also asked participants “How likely do you think members of each social group are to get severe COVID-19 or die from COVID-19?” as a matrix Likert Scale question across five groups: family members, close friends, co-workers, church or social club, and neighbors or community members. Participants could answer on a 5-point scale: extremely unlikely, somewhat unlikely, neither likely nor unlikely, somewhat likely, or extremely likely.

Next, we wanted to understand both the size of participants social circles and how much they trust members of their social circles. Hence, we asked participants “To the best of your ability, type in the first names or relationship titles (mother, best friend, cousin, co-worker, etc.) of each person you voluntarily had a conversation with in the past week.” and “Of those people you listed in the previous question, how many of them would you turn to for advice with a major personal problem?” to approximate the number of people they interact with on a regular basis and how much they trust those individuals. This line of questioning is similar to the questions asked in prior work to approximate the size of one’s social circle [48], [50].

Lastly, we asked “In your own words, why did you get a COVID-19 vaccine?” or “In your own words, why did you not get a COVID-19 vaccine?” as an open-ended question to elicit reasoning outside of our social influence questions. Our goal with this question is to capture if participants primary reasoning for getting or not getting vaccinated is indeed based on social cues or if it is based on something else.

In the data analyzed, we use data from both the pilot deployment and actual deployment, totalling to 1015 responses. After filtering out responses that did not pass our attention check question, we had 1000 survey responses to analyze between the three deployments, with 486 unvaccinated participants and 514 vaccinated participants.

To analyze our ordered Likert Scale questions, we leverage ordinal logistic regression. Each model uses an ordered categorical question as the dependent variable and if the participant was in the vaccinated group or unvaccinated group has the independent variable. This method allows us to estimate changes in units of ordered logits between the two groups and if those changes are statistically significant. We describe this process in more detail in Section 3 and Table 3.

To analyze our open-ended reasoning question, we take a multi-round, open-coding approach. First, two of the authors individually went over the open-ended responses from the vaccinated group and the unvaccinated group to identify broad codes, continuing until the list of codes was stable. The two authors then discuss their codes to develop a combined list of codes, each with specific definitions of what reasoning fits into the category. The final set of codes can be found in Tables 5 and 6. Then the same two authors coded all of the responses for both groups (vaccinated and not). Since the respondents could describe as many reasons as they wanted, each participants response could fall into more than one category. After the second round of coding, agreement was computed, where a response is said to have agreement if all codes match across the two coders. In the unvaccinated group, 79.84% of the responses had full agreement, while the vaccinated group, 85.38% of the responses had full agreement. The two authors met again to discuss and resolve these conflicts. Disagreements in the unvaccinated group were mostly due to interpreting what entity was being distrusted (science, government, media, pharmaceutical companies, general), hence we merged these to a broader category of Distrust. For example, some response were very specific: “Because there’s not enough research to determine if it’s safe and I do not trust the government or media.” while others were more general: “i dont trust any of it”. Disagreements in the vaccinated group mostly were due to differentiating the categories Protect others versus Right thing to do. In this case, we opted for stricter definitions of each to avoid misinterpretation. For example, a response had to use the phrase “it was the right thing to do” to be included in the category Right thing to do, while calls to protecting others, such as “To protect my family, community” or “I wanted to protect others from catching Covid and help reduce the spread” were put in the category Protect others.

Definitions and examples of the final agreed upon results can be found in Tables 4, 5, and 6.

Lastly, for our questions on social circle size and trust, we manually parse and count the number of people listed by participants for each question. We do this task manually as participants format these lists in a variety of ways, making automated parsing prone to errors. The distributions of approximate social circle size and social circle trust across the vaccinated and unvaccinated groups are then analyzed quantitatively using a Mann-Whitney test.

3 Results↩︎

Figure 1: U.S. counties where at least one survey participant resides. Participants in our survey lived in 882 unique U.S. counties, with 471 counties represented in the vaccinated group and 461 counties represented in the unvaccinated group.

Table 1: No caption
image image image

In Figure 1, we show the distributions of political leaning, rural-urban identity, and education. These demographics and others are included in our ordinal logistic regression analysis found in Table 3. Additionally, in Figure 1, we show a map of the U.S. counties in which survey participants resided in at the time of the survey.

Across both the vaccinated and unvaccinated groups, participants described themselves as 40.7% male, 57.01% female, and 2.02% non-binary/third gender, and 86.1% white, 6.8% black, 3.5% Asian, 2.8% other, and 0.8% Native American. There were no significant differences found between these demographics across the two groups.

There were significant differences across other demographic traits, such as political leaning, rural-urban identity, rural residency, income, and education. Specifically, vaccinated participants were significantly more left leaning, identified more as city/suburban people, lived in more urban areas, had higher income, and more education.

These demographic differences align closely with results from previous work. For example, reports from The COVID States project2 found that people who lived in city/suburban areas, had more education, and had higher income, had significantly higher vaccination rates and greater support for vaccine requirements [51], [52]. Other studies have shown that rural residents are “significantly less likely to participate in COVID-19-related preventive health behaviors [53]” and that rural residents have higher rates of vaccine resistance [51]. Further, as stated by Green et al. (2022), “partisanship remain[s] the most stable and sizable gap” in COVID-19 behaviors and attitudes [54]. Prior work has shown that vaccine hesitancy is significantly higher for those who identify as Republications than those who identify as Democrats [55][59]. The demographic results from our survey provide further robustness to these prior findings.

Next, we examine our core survey questions capturing the perceived influence of others on vaccine decisions and the perceived agreement of others with vaccine decisions. In Table 3 we show the results of our ordinal logistic regression analysis over the perceived influence and agreement across theoretically strong and weak social ties.

At a high level, no matter the social group, vaccinated participants perceived significantly higher influence from others and significantly higher agreement with others than participants who were not vaccinated. For example, from Table 4, given that the coefficient estimates from our regression analysis are given in units of ordered logits or ordered log odds, we can say that for one unit increase in the independent variable (going from no to yes for the question “Have you received one or both shots for COVID-19?”), we expect a 1.0974 increase in the expected value of perceived influence of close friends (ordered as a 5-point scale from None at all to A great deal), meaning that vaccinated participants perceived significantly more influence from their close friends than participants who are unvaccinated. Similarly, we can say that we expect a 2.2421 increase in the value of perceived agreement of close friends with the participants vaccination decisions (ordered as a 3-point scale from disagree to agree), indicating vaccinated participants perceived that their close fiends agreed with their decision to get vaccinated significantly more than participants who were not vaccinated (i.e. close friends disagree with their decision to not get vaccinated). The only exception to this finding is the perceived influence of one’s ‘Church or Social club’ - which showed no significant differences between the vaccinated and not vaccinated groups. Further, the perceived influence and agreement weakens slightly (although not consistently) as we move from theoretically stronger relationships to weaker relationships (e.x. family to neighbors), yet statistically significant differences between the vaccination groups still remain across the social circles.

These findings align with the findings of previous work. Tunçgenç et al. (2021) showed that the perceived approval and adherence of others to social distancing during COVID-19 predicted participant’s adherence to social distancing, particularly when others within one’s close social circle approved and adhered to pandemic guidelines [48]. However, this study did not demonstrate differences in perceived social influence between groups, rather the authors demonstrated that there is a relationship between individual adherence and perceptions of social circle adherence. We add nuance to this finding, by showing perceptions of influence and approval may not be symmetrically across groups (those who follow pandemic guidelines and those who do not).

Next, we examine how the perceived danger of COVID-19 to varying social groups changes across vaccinated and unvaccinated participants. Specifically, we ask: “How likely do you think members of each social group are to get severe COVID-19 or die from COVID-19?” - capturing an indirect social influence rather than a direct influence. It is reasonable to assume that if one believes COVID-19 is dangerous to their family and close friends, they may be more likely to get vaccinated to protect them.

We find similarly divided results between the two groups. As shown in Table 4, we find that across all social groups vaccinated participants perceived COVID-19 as more dangerous to others than unvaccinated participants. While there are some variations across social groups (i.e. perceiving family as in more danger than close friends, etc.), those variations are not significantly different.

While to the best of our knowledge, perceived danger of COVID-19 to specific social groups has not been studied, our results align with more general findings in the literature. Many factors have been shown to predict COVID-19 risk perceptions, such as personal experience with the virus and hearing about the virus from friends and family [60]. Hearing about the virus through friends and family was shown to be the strongest predictor of perceived risk, where risk includes both risk of self and risk of others contracting the virus [60]. Other non-social factors have also been shown to predict personal risk to COVID-19, including psychological factors and institutional trust [60], [61].

Table 2: No caption
(a) Family (b) Close Friends
image image
(c) Co-workers (d) Neighbors
image image
Table 3: Results from 20 ordinal regression models, where being vaccinated is the independent variable. If the 95% CI does not cross 0, the parameter estimate is statistically significant, also shown by the p-value. Note, that *** indicates \(p < 0.001\) and ** indicates \(p < 0.01\). The coefficient estimates provided are given in units of ordered logits, or ordered log odds. So, for example, for the independent variable - ‘Have you received one or both shots for COVID-19?’ and the dependent variable ‘What is your political leaning?’- we would say that for a one unit increase in the IV (i.e., going from ‘No’ to ‘Yes’), we expect a \(-1.94\) decrease in the expected value of political leaning (ordered as Very Liberal to Very Conservative) on the log odds scale, meaning vaccinated participants were more politically liberal. The interpretation column indicates a trait of vaccinated participants.
Ordinal Question Coef 95% CI Interpretation
Political Leaning -1.94*** -2.19 to -1.69 More liberal
Rural-Urban Identity -0.64*** -0.87 to -0.41 More identify as city people
Rural-Urban Residency -0.44*** -0.68 to -0.21 Live in more urban areas
Income 0.63*** 0.42 to 0.85 Higher income
Education 0.95*** 0.72 to 1.18 More educated
Perceived Influence
Ordinal Question Coef 95% CI Interpretation
Family 1.07*** 0.84 to 1.31 More influenced
Close friends 1.10*** 0.84 to 1.35 More influenced
Co-workers 1.07*** 0.73 to 1.40 More influenced
Church or Social club 0.38 -0.06 to 0.82 No difference
Neighbors 1.22*** 0.86to 1.59 More influenced
Perceived Agreement
Ordinal Question Coef 95% CI Interpretation
Family 1.93*** 1.64 to 2.22 More agreement
Close friends 2.24*** 1.95 to 2.53 More agreement
Co-workers 1.91*** 1.62 to 2.21 More agreement
Church or Social club 1.05*** 0.77 to 1.33 More agreement
Neighbors 1.77*** 1.46 to 2.07 More agreement
Perceived Danger of COVID-19
Ordinal Question Coef 95% CI Interpretation
Family 0.68*** 0.46 to 0.91 More dangerous
Close friends 0.49*** 0.26 to 0.71 More dangerous
Co-workers 0.66*** 0.43 to 0.89 More dangerous
Church or Social club 0.83*** 0.60 to 1.07 More dangerous
Neighbors 0.84*** 0.61 to 1.07 More dangerous

We can further contextualize our findings by approximating the size of and trust in participants social groups. While theoretically, our survey questions measure differences across strong and weak social ties, the strength of the ties between participants and these groups likely changes across individuals. Hence, here we utilize the answers to the questions asked about voluntary conversations participants had within the past week and the county in which participants lived during the pandemic, as described in the methods section above.

We find that on average, participants in the unvaccinated group had conversations with 4.02 people (median 4 people, standard deviation 2.45) in the week prior to the survey and trusted 2.27 of them (median 2 people, standard deviation 1.64). On average, participants in the vaccinated group had conversations with 5.26 people (median 4 people, standard deviation 4.83) in the week prior to the survey and trusted 2.76 of them (median 3 people, standard deviation 2.03). Both groups trusted about 60% of those they had conversations with in the past week, with 60.91% trusted in the vaccinated group and 60.60% trusted in the unvaccinated group. Note, these approximate social circle size numbers align with both pre-pandemic and during-pandemic numbers, which both showed a median close circle size of 4 people [48], [62].

To better measure differences between the two groups, we perform a Mann-Whitney test on each pair of distributions, finding significant differences between approximate social circle size (pvalue=1.6301e-07) and the number of individuals trusted (pvalue=1.8238e-05), but no significant difference was found between the proportion of social circle members trusted (pvalue=0.9807). Hence, vaccinated participants interacted with slightly more people than unvaccinated, but trust in those people is no different between the groups.

As another proxy for the number of people participants encounter, we can also look at the areas of the U.S. in which participants live. Overall, 51.62% of participants, resided in metro areas with 1 million population or more (RUCC 1). More broadly, 86.36% of participants lived in metro counties (RUCCs 1 through 3) and only 13.64% lived in nonmetro counties (RUCCs 4 through 9). Of the vaccinated participants, 57.25% of participants, resided in metro areas of 1 million population or more, 89.02% lived in metro counties, and 10.98% lived in nonmetro counties. Of the unvaccinated participants, 45.63% of participants, resided in metro areas of 1 million population or more, 83.54% lived in metro counties, and 16.46% lived in nonmetro counties.

As shown in Table 3, when examining rural-urban residency using an ordered logistic regression model, we find that the vaccinated participants live in more urban areas than unvaccinated participants. Precisely, for one unit increase in the independent variable (going from no to yes for the question “Have you received one or both shots for COVID-19?”), we expect a -0.4429 decrease in the expected value of RUCC (ordered as a 9-point scale from 1 to 9), which is statistically significant.

Hence, both the approximations of the number of people participants interacted with prior to the survey (social circle size and residency) suggest that the vaccinated population interacted with more people than the unvaccinated population did. Theoretically, it makes sense that interacting with less people can impact one’s perceptions of COVID-19, as interacting with less people reduces the chance that one comes in close contact with someone who had been severely ill from COVID-19. In general, this idea is supported in the literature. For example, when studying vaccine messaging strategies, it was found that “messages evoking harm reduction and people you know were more effective in counties where the virus is spreading more quickly. [63]

There are likely other confounding factors related to location that may have a significant impact on one’s pandemic-related decisions. For example, prior work has demonstrated that local news coverage during 2020 and 2021 was best explained by national trends rather than by local conditions, and the themes of that coverage varied across counties in the U.S. [64]. Further, we know that there were urban-rural differences in COVID-19 behaviors, and that those behaviors changed depending on the news produced locally. Specifically, Kim et al. (2020) showed that “rural residents [were] more likely to engage in social distancing behavior than otherwise similar rural residents if their local news [was] produced in a city that is more impacted by COVID-19” [65]. Other work has shown wide variations in vaccine support across areas in the United States [56].

Table 4: Percentage of responses that fit into each category across vaccination groups. Note, responses could use multiple categories of reasoning. Hence, the percentages will not add up to 100%. For definitions of each category, see Tables 6 and 5.
Vaccinated Not Vaccinated
Category % of Responses Category % of Responses
Protect Self 74.27% Fear of Vaccine 36.65%
Protect Others 54.77% Natural Immunity 7.41%
Trust in Science 14.81% Distrust 6.24%
Employer Mandate 2.53% Not True Vaccine 13.26%
Return to Normal 7.41% Apathy 7.02%
Social Pressure 2.73% Restricted Access 6.63%
Trust in Social Circle 1.75% Low Risk of COVID-19 21.64%
The Right Thing to Do 5.85% Infringement on Rights 2.73%
Other 2.14% Conspiracy Theory 5.26%
Other 2.53%

Lastly, to capture reasoning that may not have been addressed by our ordinal-scaled questions, we ask an open ended question to all participants: “In your own words, why did you (not) get a COVID-19 vaccine?”. As discussed in Section 2, we take an open-coding approach to group responses into categories. The final set of categories and corresponding percentage of responses in the category can be found in Table 4. Detailed definitions of each category can be found in Tables 6 and 5.

On average, participants in the unvaccinated group wrote more, producing 140.18 characters on average (maximum 1651 characters, minimum 7 characters) compared to only 89.77 characters on average in the vaccinated group (maximum 760 characters, minimum 12 characters). This often meant that participants in the unvaccinated group were listing many more reasons (or used very complex reasoning) for not getting vaccinated than vaccinated participants did for getting vaccinated.

The most frequent reason used by unvaccinated participants to not get vaccinated was fear of the vaccine, including fear of the vaccine itself or its side effects (36.65% of responses). For example:

“I was concerned about the side effects and cost of possible medical treatment. I’ve reacted horribly to both the tetanus vaccine and flu vaccine.”

“I am worried about potential side effects (mainly about long-term ones which are still currently unknown) and I don’t like needles.”

“I do not believe the benefits outweigh the risks.”

“The vaccine is dangerous and proven not to be 100% effective against the virus. And COVID is nothing more than the flu.”

The second most frequent reason expressed was a perceived low risk of catching or becoming seriously ill from COVID-19 (21.64% of responses), aligning with the results from our ordinal question on the perceived dangers of COVID-19 to others.

Other reasons expressed included: claims that the vaccine was “not a true vaccine” (13.26% of responses), claims of natural immunity (7.41% of responses), distrust in institutions (6.24% of responses), and one or more conspiracy theories (5.26% of responses). As with many of the responses from the unvaccinated participants, multiple of these categories were used together. For example, often there were overlaps between the categories Ineffective/Not True Vaccine and Distrust and Conspiracy Theory; such as:

“It’s not a vaccine, it’s a biologic jab intended to kill people.”

“You would have to be a fool to not see it for what it is at this point. I did not get it because it’s poison, intentional... technological... poison. Used to kill and control. And part of a larger agenda to get us used to passports, digital IDs and social credit.”

“I do not trust the efficacy of the mRNA technology. I do not trust pharmaceutical companies that attempt to hide the results of trials for decades and have been granted total immunity from prosecution. I believe that adverse reactions have been underreported. I think the silencing and censoring of certain immunologists and virologists that have doubts about mRNA, and a substantial contingent of the medical community advocating for lockdowns and isolation except for massive gatherings for "racial justice" have seriously weakened my level of trust in the scientific and medical communities as a whole. I combine this with the illogical insistence that "masks work" despite overwhelming evidence that any type of face covering beyond type n-95 is ineffective.”

“They are not real vaccines yet. Takes a few years. The side effects and deaths are extreme with many being hidden. Once one is made that actually is an immunology then I will look into it further. Right now it is a jab of chemicals.”

“They are all under tested and mRNA vaccines were already deemed not safe for human trial due to mass death and vaccine indused autoimmune immune deficiency in all animal test groups. Good luck with that.”

Some of these findings align with prior work. For example, research from The COVID States project demonstrated that “the biggest expressed concern of the unvaccinated is the safety of the COVID-19 vaccines” and that the unvaccinated population is “more likely to be skeptical of the efficacy and safety of vaccines” [66]. Further, as others have argued, misinformation and conspiracy theories can play a role in COVID-19 decision making [67].

Overall, vaccinated participants expressed very different reasons for getting vaccinated than unvaccinated participants expressed for not getting vaccinated. The vast majority of vaccinated participants stated that they chose to get vaccinated to protect themselves and protect others (74.27% of responses contained protect self and 54.77% contained protect others), and typically used very short, simple reasoning. For example, the majority of the responses looked like the following:

“To protect myself and my family, community.”

“To protect myself and others”

Other less frequent reasons by vaccinated participants included trust in science (14.81% of responses), desire to return to normal (7.41% of responses), and getting vaccinated being “the right thing to do” (5.85% of responses). Notably, participants explicitly stating they got vaccinated due to trusting in their social circle, either through direct advice from social circle, a request to get vaccinated from social circle, or family members also getting vaccinated, only was present in 1.75% of responses. Further, social pressure, both directly from social circles or indirectly from social norms, was mentioned in only 2.73% of responses. Hence, while perceived social influence was high amongst vaccinated participants, it may not be the primary reasoning behind the decision.

4 Discussion↩︎

In this study, we found that vaccinated participants perceived more influence from and agreement with others than unvaccinated participants. The vaccinated population also perceived COVID-19 as more dangerous to others than the unvaccinated population did. For context, we approximated that participants in the vaccinated population had larger social circles on average and lived in more densely populated areas. Hence, vaccinated participants had a higher likelihood of interacting with someone who had been severely ill or died from COVID-19 than unvaccinated participants. These results expand upon and add robustness to the findings of prior work on social influence during pandemic and epidemics [48].

In contrast, social influence does not seem significant in deciding not to be vaccinated. In open-ended responses, unvaccinated participants most frequently cited fear of the vaccine and distrust in the institutions who make and support the vaccines.

While the primary goal of this study was to examine social influence, it is clear that social influence is not the only mechanism of impression in decision making. Other factors not directly measured in this study that likely also play a role in vaccination decisions include media consumption [64], [68][70] and individual, generational, and cultural memory [71][73]. Our qualitative results support this notion.

Notably, despite vaccinated participants citing social influence at varying levels in their open-ended responses, they too were motivated by fear, frequently citing that they got vaccinated to protect themselves or to avoid dying from COVID-19.

Fear and anxiety being a driver of health-related decision making is well-supported in the literature [74][76], including behaviors during the COVID-19 pandemic. For example, Harper et al. (2021) found that from a sample of 324 people from the U.K. that “the only predictor of positive behavior change (e.g., social distancing, improved hand hygiene) was fear of COVID-19 [77]”, including when controlling for political and ideological variables. When fear and anxiety motivates vaccination, this fear is functional (as Harper et al. (2021) puts it), but, as our study suggests, when fear and anxiety motivates not getting vaccinated, the fear may be counterproductive and even detrimental.

In conclusion, we found that vaccinated populations perceived more influence from their social circles than unvaccinated populations. This finding held true across various social groups, including family, close friends, co-workers, and neighbors. Similarly, we found that vaccinated participants perceived that others agreed with their decision to get vaccinated significantly more than unvaccinated participants perceived that others agreed with their decision to not get vaccinated. Indirect measures of social influence also followed this trend. Vaccinated participants perceived COVID-19 as more dangerous to their social circles than unvaccinated participants. Despite the clear differences in perceived social influence across the groups, we found through open-ended responses that both vaccinated and unvaccinated participants frequently cited fear as a motivating factor in their decision: vaccinated participants feared COVID-19, while unvaccinated participants feared the vaccine itself. Together, our results expand the current literature on vaccination behavior and add robustness to several previous findings.

5 Limitations↩︎

While we are confident in the results presented in this study, it is not without limitations. First, our work is limited in its reliance on self-reported/perceived effects - it is unlikely that perceived influence perfectly correlates with actual influence, which may happen passively or unconsciously. A report from Altay et al. (2022) shows a clear example of this effect, stating that “[c]onspiracy believers were more likely to report relying less on social information than actually relying less on social information in the behavioral tasks” [78]. Our focus in this work is not on conspiracy believers; however, the general premise of reporting the use of social information versus actual reliance on social information still applies. From this view, it may be that both vaccinated and unvaccinated populations rely on social information, but unvaccinated populations report using less of it than vaccinated populations.

Second, as a whole, the survey data are not from a representative sample of the U.S. population, as clearly shown by the racial demographics in Section 3. This bias is due to our method of balanced sampling across the vaccinated and unvaccinated groups. Systematically sampling this imbalanced variation allowed us to capture both vaccinated and unvaccinated people’s perceptions fairly. While this survey could be re-deployed on a truly representative sample the U.S. population, there would be an imbalance in vaccinated versus unvaccinated participants, as the majority of the U.S. population is vaccinated. Despite this limitation, our survey did cover a broad set of counties and demographics.

Table 5: Qualitative Code Book definitions for the vaccinated group
Category Definition
Protect Self Protecting self, to avoid bad illness, or to avoid death
Protect Others Protecting others, including family, friends, community, neighbors, or “others”
Trust in Science Specific mentions of trust in the science or vaccines
Employer Mandate Specific mentions of employer mandate or job prospects
Desire to Return to Normal Specific mentions of “return to normal”
Social Pressure Various forms of social pressure, such as specific mentions of social pressure or indirect statements like: “not to look like an anti-vaxxer” or “not to be socially unacceptable”
Trust in Social Circle Conversations with family or friends, family member asking them to get vaccinated, family member also getting vaccinated, etc.
The Right Thing to Do Specific mentions of getting vaccinated being “right thing to do”
Other Response doesn’t fit into above categories
Table 6: Qualitative Code Book definitions for the unvaccinated group
Not Vaccinated
Category Definition
Fear of Vaccine Fear of the vaccine itself or the side effects from it. Some mention being afraid of dying from the vaccine, such as “There have been severe side effects and death after the vaccine.”
Natural Immunity Specific mentions of having “natural immunity” including because of having COVID-19 already, because they take supplements, because “God” gave them an “immune system”, or belief that they “can fight COVID-19 off naturally”
Not True Vaccine Statements of two general forms: 1. The vaccine is ineffective because people can get COVID-19 after being vaccinated, or 2. The belief that the vaccine is "not a true vaccine" for reasons related to “mRNA” not being a “true vaccine” or because it doesn’t always prevent catching COVID-19
Distrust Specific mentions of distrusting science, the media, the government, “big pharma”, or “it”
Apathy Did not care to get it or did not have time
Restricted Access Participants have the intention to get the vaccine, but cannot due to health issues, lack of transportation, or family member not allowing them to get vaccinated
Low Risk of COVID-19 Participant believes they are not at risk of getting COVID-19 or getting severely ill from COVID-19
Infringement on Rights Participant did not get vaccinated because they do not like “being forced” to get vaccinated or out of the principle that “mandates are unacceptable in a free and democratic society”
Conspiracy Theory Reasoning using one or more conspiracy theories, where there is a secret, over-arching, malicious plot by powerful groups of people against those not in power, many are framed around blaming democrats, general government, or the media
Other Response doesn’t fit into above categories

Funding: Participants in this study were paid through internal funds in the School of Information Sciences at UTK. No external funding was used for this study.
Ethical Approval: All research was performed in accordance with relevant guidelines/regulations applicable when human participants are involved (Declaration of Helsinki). The research was approved by The University of Tennessee-Knoxville Human Research Protections Program (HRPP), which determined that the application was eligible for exempt status under 45 CFR 46.104.d, Category 2 (https://www.hhs.gov/ohrp/regulations-and-policy/regulations/45-cfr-46/revised-common-rule-regulatory-text/index.html#46.101). Our application was determined to comply with proper consideration for the rights and welfare of human subjects and the regulatory requirements for the protection of human subjects.
Informed consent: As per the IRB application, informed consent was acquired from all participants in this survey. All participants were adults living the United States.
Author’s contribution: DY: original idea, survey design, qualitative analysis. RAB: writing, literature review. BDH: survey design, quantitative analysis, qualitative analysis, writing, literature review.
Conflict of interest: The authors have no competing interests related to this work.
Data availability statement: Due to the sensitive nature of survey data, the data used in this study is only available upon request.
Supplementary information Additional figures are provided in the Supplemental materials.


Prather KA, Wang CC, Schooley RT. Reducing transmission of SARS-CoV-2. Science. 2020;368:1422–1424. .
Aksoy CG, Ganslmeier M, Poutvaara P. Public attention and policy responses to COVID-19 pandemic. IZA Discussion Papers. 2020;13427. .
Hale T, Petherick A, Phillips T, Webster S. Variation in Government Responses to COVID-19. Oxford Blavatnik School of Government Working Papers. 2020;BSG-WP-2020/031.
Bedford J, Enria D, Giesecke J, Heymann D, Ihekweazu C, Kobinger G, et al. COVID-19: towards controlling of a pandemic. The Lancet. 2020;395:1015–1018. .
Frey CB, Chen C. Democracy, Culture, and Contagion: Political Regimes and Countries Responsiveness to Covid-19; 2020. .
Munster VJ, Koopmans M, van Doremalen N, van Riel D, de Wit E. A novel coronavirus emerging in China—key questions for impact assessment. New England Journal of Medicine. 2020;382(8):692–694.
Chowell G, Mizumoto K. The COVID-19 pandemic in the USA: What might we expect? The Lancet. 2020;395:1093–1094.
Zhang QB Weituo and. Making decisions to mitigate COVID-19 with limited knowledge. The Lancet Infectious Diseases. 2020;20:1121–1122.
Maharaj S, Kleczkowski A. Controlling epidemic spread by social distancing: Do it well or not at all. BMC Public Health. 2012;12(1):679.
Funk S, Salathé M, Jansen VAA. Modelling the influence of human behaviour on the spread of infectious diseases: a review. J R Soc Interface. 2010;106:1247–1256.
Guiteras R, Levinsohn J, Mobarak A. Encouraging sanitation investment in the developing world: A cluster-randomized trial. Science. 2015;348:903–906.
Zhang J, Centola D. Social networks and health: new developments in diffusion, online and offline. Annual Review of Sociology. 2019;45:91–109.
Gharzai LA, Beeler WH, Jagsi R. Playing into Stereotypes: Engaging Millennials and Generation Z in the COVID-19 Pandemic Response. Advances in Radiation Oncology. 2020;.
Ruck DJ, Borycz J, Bentley RA. Cultural values predict national COVID-19 death rates. Springer Nature Social Sciences. 2021;1:article 74.
Gelfand MJ, Jackson JC, Pan X, Nau DS, Pieper D, Denison E, et al. The relationship between cultural tightness–looseness and COVID-19 cases and deaths: a global analysis. The Lancet Planetary Health. 2021;5:e135 – e144.
Inglehart R, Welzel C. Modernization, Cultural Change, and Democracy: The Human Development Sequence. Cambridge University Press.; 2005.
Ruck DJ, Matthews LJ, Kyritsis T, Atkinson QD, Bentley RA. The cultural foundations of modern democracies. Nature Human Behaviour. 2019;4:265–269.
Desmet K, Wacziarg R. Understanding Spatial Variation in Covid-19 Across the United States. CEPR Discussion Paper Series. 2020;.
Ruck DJ, Bentley RA, Borycz J. Early warning of vulnerable counties in a pandemic using socio-economic variables. Economics and Human Biology. 2021;41:100988 – 100988.
de Oliveira RP, alberto Achcar J, Nunes AA. Modeling the incidence and death rates of COVID-19 pandemic in different regions of the world. Epidemiologic Methods. 2020;9.
Verelst F, Willem L, Beutels P. Behavioural change models for infectious disease transmission: a systematic review (2010–2015). Journal of The Royal Society Interface. 2016;13(125):20160820.
Fineberg HV. The paradox of disease prevention: celebrated in principle, resisted in practice. Jama. 2013;310(1):85–90.
Jansen VA, Stollenwerk N, Jensen HJ, Ramsay M, Edmunds W, Rhodes C, et al. Measles outbreaks in a population with declining vaccine uptake. Science. 2003;301(5634):804–804.
Li R, Pei S, Chen B, et al. Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus (SARS-CoV2)[published online ahead of print March 16, 2020]. Science;10.
Bendavid E, Mulaney B, Sood N, Shah S, Ling E, Bromley-Dulfano R, et al. COVID-19 Antibody Seroprevalence in Santa Clara County, California. medRxiv. 2020;.
Baud D, Qi X, Nielsen-Saines K, Musso D, Pomar L, Favre G. Real estimates of mortality following COVID-19 infection. The Lancet infectious diseases. 2020;20(7):773.
Marchant R, Samia NI, Rosen O, Tanner MA, Cripps S. Learning as we go: An examination of the statistical accuracy of COVID19 daily death count predictions. arXiv preprint arXiv:200404734. 2020;.
Mani A, Mullainathan S, Shafir E, Zhao J. Poverty impedes cognitive function. science. 2013;341(6149):976–980.
Trueblood J, Sussman A, O’Leary D, Holmes W. A tale of two crises: Financial constraint and perceptions of COVID-19. 2020;.
Pritchard AJ, Silk MJ, Carrignon S, Bentley RA, Fefferman NH. Balancing timeliness of reporting with increasing testing probability for epidemic data. Infectious Disease Modelling. 2022;7:106 – 116.
Greenstone M, Nigam V. Does social distancing matter? University of Chicago, Becker Friedman Institute for Economics Working Paper. 2020;(2020-26).
Carrignon S, Bentley RA, Silk MJ, Fefferman NH. How social learning shapes the efficacy of preventative health behaviors in an outbreak. PLoS ONE. 2022;17.
Christakis NA, Fowler JH. Social contagion theory: examining dynamic social networks and human behavior. Statistics in medicine. 2013;32(4):556–577.
Bentley RA, Earls M, O’Brien MJ. I’ll have what she’s having: Mapping social behavior. MIT Press; 2011.
Silk MJ, Carrignon S, Bentley RA, Fefferman NH. Observations and conversations: how communities learn about infection risk can impact the success of non-pharmaceutical interventions against epidemics. BMC Public Health. 2022;22.
Gibson MA, Lawson DW. Applying evolutionary anthropology. Evolutionary Anthropology: Issues, News, and Reviews. 2015;24(1):3–14.
Hoppitt W, Laland KN. Social learning: an introduction to mechanisms, methods, and models. Princeton University Press; 2013.
Rendell L, Boyd R, Enquist M, Feldman MW, Fogarty L, Laland KN. How copying affects the amount, evenness and persistence of cultural knowledge: insights from the social learning strategies tournament. Philosophical Transactions of the Royal Society B: Biological Sciences. 2011;366(1567):1118–1128.
Hauser OP, Rand DG, Peysakhovich A, Nowak MA. Cooperating with the future. Nature. 2014;511(7508):220–223.
Li J, Lee LF. Binary Choice under Social Interactions: An Empirical Study with and without Subjective Data on Expectations. Journal of Applied Econometrics. 2009;24(2):257–281.
Caiado CC, Brock WA, Bentley RA, O’Brien MJ. Fitness landscapes among many options under social influence. Journal of Theoretical Biology. 2016;405:5–16.
Vidiella B, Carrignon S, Bentley RA, O’Brien MJ, Valverde S. A cultural evolutionary theory that explains both gradual and punctuated change. Journal of the Royal Society Interface. 2022;19:20220570. .
Schrock JM, Snodgrass JJ, Sugiyama LS. Lassitude: The emotion of being sick. Evolution and Human Behavior. 2020;41(1):44–57.
Kronbichler A, Kresse D, Yoon S, Lee KH, Effenberger M, Shin JI. Asymptomatic patients as a source of COVID-19 infections: A systematic review and meta-analysis. International journal of infectious diseases. 2020;98:180–186.
Silk MJ, Carrignon S, Bentley RA, Fefferman NH. Improving pandemic mitigation policies across communities through coupled dynamics of risk perception and infection. Proceedings of the Royal Society B. 2021;288.
Pritchard AJ, Silk MJ, Carrignon S, Bentley RA, Fefferman NH. How reported outbreak data can shape individual behavior in a social world. Journal of Public Health Policy. 2022;43:360 – 378.
Banerjee A, Chandrasekhar AG, Duflo E, Jackson MO. The diffusion of microfinance. Science. 2013;341(6144):1236498.
Tunçgenç B, El Zein M, Sulik J, Newson M, Zhao Y, Dezecache G, et al. Social influence matters: We follow pandemic guidelines most when our close circle does. British Journal of Psychology. 2021;112(3):763–780.
Lunz Trujillo K. Rural Identity as a Contributing Factor to Anti-Intellectualism in the US. Political behavior. 2022;p. 1–24.
Dunbar RI, Spoors M. Social networks, support cliques, and kinship. Human nature. 1995;6(3):273–290.
Lazer D, Ognyanova K, Baum MA, Druckman J, Green J, Gitomer A, et al. The COVID States Project# 43: COVID-19 vaccine rates and attitudes among Americans. OSF Preprints. 2021;10.
Baum M, Quintana A, Simonson MD, Lazer D, Ognyanova K, Chwe H, et al. The COVID States Project# 52: Public support for vaccine requirements. 2021;.
Callaghan T, Lueck JA, Trujillo KL, Ferdinand AO. Rural and urban differences in COVID-19 prevention behaviors. The Journal of Rural Health. 2021;37(2):287–295.
Green J, Druckman JN, Baum MA, Lazer D, Ognyanova K, Simonson MD, et al. Using general messages to persuade on a politicized scientific issue. British Journal of Political Science. 2022;p. 1–9.
Reiter PL, Pennell ML, Katz ML. Acceptability of a COVID-19 vaccine among adults in the United States: How many people would get vaccinated? Vaccine. 2020;38(42):6500–6507.
Perlis R, Lazer D, Ognyanova K, Baum M, Santillana M, Druckman J, et al. The COVID States Project# 9: Will Americans vaccinate themselves and their children against COVID-19. 2020;.
Khubchandani J, Sharma S, Price JH, Wiblishauser MJ, Sharma M, Webb FJ. COVID-19 vaccination hesitancy in the United States: a rapid national assessment. Journal of community health. 2021;46(2):270–277.
Milligan MA, Hoyt DL, Gold AK, Hiserodt M, Otto MW. COVID-19 vaccine acceptance: Influential roles of political party and religiosity. Psychology, Health & Medicine. 2021;p. 1–11.
Clinton J, Cohen J, Lapinski J, Trussler M. Partisan pandemic: How partisanship and public health concerns affect individuals’ social mobility during COVID-19. Science advances. 2021;7(2):eabd7204.
Dryhurst S, Schneider CR, Kerr J, Freeman AL, Recchia G, Van Der Bles AM, et al. Risk perceptions of COVID-19 around the world. Journal of Risk Research. 2020;23(7-8):994–1006.
Yıldırım M, Güler A. Factor analysis of the COVID-19 Perceived Risk Scale: A preliminary study. Death studies. 2022;46(5):1065–1072.
Hill RA, Dunbar RI. Social network size in humans. Human nature. 2003;14(1):53–72.
Green J, Lazer D, Baum M, Druckman J, Uslu A, Simonson MD, et al. The covid states project# 36: Evaluation of covid-19 vaccine communication strategies. 2021;.
Joseph K, Horne BD, Green J, Wihbey JP. Local News Online and COVID in the US: Relationships among Coverage, Cases, Deaths, and Audience. In: Proceedings of the International AAAI Conference on Web and Social Media. vol. 16; 2022. p. 441–452.
Kim E, Shepherd ME, Clinton JD. The effect of big-city news on rural America during the COVID-19 pandemic. Proceedings of the National Academy of Sciences. 2020;117(36):22009–22014.
Uslu A, Lazer D, Perlis RH, Baum M, Quintana A, Ognyanova K, et al. The COVID States Project# 63: The decision to not get vaccinated, from the perspective of the unvaccinated. 2021;.
Pierri F, Perry BL, DeVerna MR, Yang KC, Flammini A, Menczer F, et al. Online misinformation is linked to early COVID-19 vaccination hesitancy and refusal. Scientific reports. 2022;12(1):5966.
Schulhofer-Wohl S, Garrido M. Do newspapers matter? Short-run and long-run evidence from the closure of The Cincinnati Post. Journal of media economics. 2013;26(2):60–81.
Hayes D, Lawless JL. As local news goes, so goes citizen engagement: Media, knowledge, and participation in US House Elections. The Journal of Politics. 2015;77(2):447–462.
Gentzkow M, Shapiro JM, Sinkinson M. The effect of newspaper entry and exit on electoral politics. American Economic Review. 2011;101(7):2980–3018.
Loftus EF. Planting misinformation in the human mind: A 30-year investigation of the malleability of memory. Learning & memory. 2005;12(4):361–366.
Lesthaeghe R. The second demographic transition: A concise overview of its development. Proceedings of the National Academy of Sciences. 2014;111(51):18112–18115.
Ruck DJ, Matthews LJ, Kyritsis T, Atkinson QD, Bentley RA. The cultural foundations of modern democracies. Nature human behaviour. 2020;4(3):265–269.
Turk T, Ewing MT, Newton FJ. Using ambient media to promote HIV/AIDS protective behaviour change. International Journal of Advertising. 2006;25(3):333–359.
Devlin E, Eadie D, Stead M, Evans K. Comparative study of young people’s response to anti-smoking messages. International Journal of Advertising. 2007;26(1):99–128.
Krishen AS, Bui M. Fear advertisements: Influencing consumers to make better health decisions. International Journal of Advertising. 2015;34(3):533–548.
Harper CA, Satchell LP, Fido D, Latzman RD. Functional fear predicts public health compliance in the COVID-19 pandemic. International journal of mental health and addiction. 2021;19:1875–1888.
Altay S, Nera K, Ejaz W, Schöpfer C, Tomas F. Conspiracy Believers Underuse Social Information (Like Everyone Else). 2022;.

  1. https://www.ers.usda.gov/data-products/rural-urban-continuum-codes.aspx↩︎

  2. https://www.covidstates.org/↩︎