Herd‐immunity communication

The more people in a community that are vaccinated against a disease, the less probable it is for the disease to spread. This effect of herd immunity protects everyone but is especially important for vulnerable populations who cannot get vaccinated (such as people with serious allergies or those with weakened immune systems; Fine, Eames, & Heymann, 2011). Recent studies have shown that communicating herd immunity has the potential to increase vaccination intentions (e.g., Betsch & Böhm, 2018; Betsch, Böhm, Korn, & Holtmann, 2017; Logan et al., 2018). Specifically, communicating the social benefit (protecting others) and visually demonstrating this effect seems to have the largest impact (see also Hakim et al., 2019).

The main goal of this study was to attempt to conceptually replicate the finding that communicating the concept of herd immunity increases the willingness to get vaccinated (Betsch et al., 2017). The original study by Betsch et al. (2017) was conducted as an online experiment, with a non‐representative sample of 2,107 adult participants from seven countries (the United States, the Netherlands, Germany, India, Hong Kong, Vietnam, and South Korea). The present replication study was also conducted as an online experiment, but with a sample of participants who live in the United Kingdom.

Given that herd immunity is under‐explained and under‐utilised in vaccine advocacy (Brockmann, 2017), it is important to test if the effect of communicating herd immunity replicates. It is especially relevant to see whether this effect is stable across countries with varying vaccination laws and levels of anti‐vaccination sentiment. Furthermore, our replication study may have practical implications for the design of herd‐immunity communication. The original study used an interactive simulation. As an alternative to this, we used an animated infographic. This medium may be easier to disseminate on television and social networks and may be more familiar to participants.

Like the original study, we explored decision‐making about a hypothetical disease transmitted directly through contact with an infected person or indirectly by touching contaminated objects. The effect of herd‐immunity communication may be dependent on the mode of disease transmission. For example, in the case of sexually transmitted infections (STIs), this could be due to the extreme heterogeneity in the risk of acquiring and transmitting STIs or the fact that STIs affect sexually active people (Garnett, 2005).

Descriptive norm communication

Descriptive norms (i.e., what most others are doing) can be a powerful source of informational social influence. By signalling what will likely be an effective and reasonable course of action under the given circumstances (Cialdini et al., 2006; Cialdini, Reno, & Kallgren, 1990), descriptive norms might also motivate individual vaccination decision‐making.

According to a review by Brewer, Chapman, Rothman, Leask, and Kempe (2017), although many correlational studies have linked norms to vaccination, no field studies have evaluated the use of descriptive norms to modify vaccination behaviour (cf. Leight & Safran, 2019). There have also only been a few survey studies and laboratory experiments exploring descriptive norms as drivers of vaccination (e.g., Hershey, Asch, Thumasathit, Meszaros, & Waters, 1994; Romley, Goutam, & Sood, 2016).

In this study, we aim to expand the literature by experimentally manipulating three descriptive‐norm levels (high vaccination coverage vs. low vaccination coverage vs. no coverage information communicated) and by assessing their influence on vaccination intentions.

Although high descriptive‐norm messages have the potential to increase vaccination uptake, they can also promote a ‘backfire effect’. Employing both interactive games (e.g., Böhm, Betsch, & Korn, 2016; Ibuka, Li, Vietri, Chapman, & Galvani, 2014; Korn, Betsch, Böhm, & Meier, 2017) and hypothetical scenarios (Betsch et al., 2017; Vietri, Li, Galvani, & Chapman, 2011), previous studies have shown that learning about a high vaccine uptake prompts the individual to strategically ‘free‐ride’ on others’ protection and to refuse vaccination. This way, the ‘free‐rider’ also avoids some individual costs (e.g., money, time, inconvenience, vaccine side effects; Fine et al., 2011).

We did not expect the high descriptive norm in our study to decrease vaccination intentions in such a way. As will be detailed below, prior to learning about the descriptive norm, all of the participants learned about the social benefit of their own vaccination decision. It has been hypothesised that this framing of herd immunity activates an individual’s prosocial or other‐regarding preferences, thus preventing free‐riding (Betsch, Böhm, & Korn, 2013).

Study design

We ran an online experiment with a 2 (herd‐immunity explanation: present vs. absent) × 3 (descriptive norm: high vs. low vs. absent) × 2 (herd‐immunity threshold: present vs. absent) between‐subjects fractional design with seven groups (Table 1). Group 7 was the control which did not receive any experimental intervention to serve as a benchmark for the effect of herd‐immunity communication.

The study used simple randomisation. The first randomisation (1:1) served to allocate half of the participants to the control group and the other half to the rest of the groups. In the second randomisation (1:1:1:1:1:1), the participants who had not been recruited to the control group were allocated to one of the six experimental groups. Participants did not know the group to which they had been allocated and researchers were blind to the group allocation process.

Sampling plan

All participants had to meet the following inclusion criteria: (1) currently residing in the United Kingdom, (2) aged between 18 and 64 years, and (3) being confident in their English skills. Typically, individuals aged 65 or above are more susceptible to vaccine‐preventable diseases, which can be more severe than for younger people. Additionally, vaccines are less protective in older adults (Goldstein, 2012). It is possible that the community‐wide benefit emphasised in the herd‐immunity explanation would act as an incentive for younger adults to voluntarily get vaccinated to prevent illness among older adults (Chapman et al., 2012). Social‐benefit messaging, however, may not be effective among the elderly and otherwise vulnerable groups (Isler, Isler, Kopsacheilis, & Ferguson, 2020). Due to potential differential effects of herd‐immunity communication interventions associated with age, recruiting adults below 65 makes the findings of our study more directly comparable with the findings of the original study, which recruited participants from the same age group of the general population (Betsch et al., 2017).

Participants were recruited through advertisements on social media (e.g., Facebook groups, Twitter, Reddit), websites, and forums. To minimise self‐selection, the advertisements and informed consent page did not suggest that the study was related to vaccination. Participation was not compensated.

Data for this study were collected at the time of the COVID‐19 pandemic, between October 5 and 24 November 2020. The second half of the data collection period encompassed the second national lockdown (GOV.UK, 2020) but ended before COVID‐19 vaccinations were first rolled out in the United Kingdom (BBC News, 2020).

Power analyses

We decided that the sample should be powered to detect the smallest effect of herd‐immunity communication that was plausible given previous research. Analysing the raw data from the original study (Betsch, Böhm, Korn, & Holtmann, 2017), we estimated the size of the effect at Partial Eta‐Squared (η²) = 0.024, across all locations. Three subsamples were large enough to allow for country‐level analysis; the effect remained small to medium in the United States and Germany (η² = 0.049 and η² = 0.073, respectively), but was small (η² = 0.002) and did not reach statistical significance in South Korea (Cohen, 1988). The effect of communicating the social benefit of herd immunity was replicated by Betsch and Böhm (2018) among a sample of US parents; the effect sizes in the two experiments were η² = 0.042 and η² = 0.044. The target sample size is based on an a priori one‐way ANOVA power analysis using the R package {easypower} (McGarvey, 2015). Assuming α = .05, N = 531 suffices to detect the original effect size of 0.024 with .95 power. Target subsamples for experimental groups 1 through 6 was, therefore, n = 45; target subsample for the control group was n = 270. The total target sample was, thus, N = 540 participants.

We additionally conducted a sensitivity two‐way ANOVA power analysis for Hypotheses 2 and 3 using G*Power 3.1.9.4 software (Faul, Erdfelder, Lang, & Buchner, 2007). With the total sample size set at n = 270, α at .05, power at .95, the numerator degrees of freedom (df) at 2, and the number of groups at 6, our study would be able to detect a minimum effect size of η² = 0.055 of the descriptive‐norm manipulation. With the numerator df set at 1 and the rest of the parameters remaining the same, it would be able to detect a minimum effect size of η² = 0.046 of the herd‐immunity threshold manipulation. These effect sizes are small, but approaching the lower limit of what can be considered a moderate effect size, that is, η² = 0.06 (Cohen, 1988).

The protocols of power analyses are available at https://osf.io/my2gf.

Procedure and variables

The study was reviewed and approved by the Institutional Review Board at the University of Belgrade Department of Psychology (protocol #2019‐046). After informed consent, the questionnaire first assessed age, gender, country of residence, education, and socioeconomic status. After an attention check, participants received a textual explanation of herd immunity, accompanied by an animated infographic. Next, they were asked to imagine themselves in a scenario in which they had to decide whether to get vaccinated against a fictitious disease. The scenario informed participants about the disease and the vaccine, the herd‐immunity threshold, and the level of the vaccination coverage in their country. Following scenario‐recall questions, participants rated their intention to get vaccinated. Then, perceived riskiness of the infection and the disease were assessed. This was followed by a measure of vaccine hesitancy and a second attention check. Immediately after the experiment, all participants were fully debriefed and received a link to the WHO website on vaccinations for further information. It was emphasised again that all information regarding the disease and the vaccine was fictitious. The questionnaire is available at https://osf.io/hq9sv.

The online experiment was implemented in SoSci Survey. It was pre‐tested on a convenience sample of 14 people (two in each group) from the target population to ensure clarity and comprehension of the materials and fine‐tune the survey process. The data from the survey pre‐test were not included in the analyses.

Manipulated variables

Herd‐immunity explanation

Participants read a general explanation of herd immunity that emphasised the social benefit of getting vaccinated (i.e., protecting others in the community, especially the vulnerable). It did not feature the term ‘herd immunity’, but rather the term ‘community immunity’, and was 200 words long (see Appendix S1 for the full text). Participants also learned about herd immunity via a 40‐second animated infographic. It showed three environments with no vs. some vs. many people vaccinated and how the pathogen spreads in each one, infecting susceptible individuals (Figure 1). To prevent the participant from skipping the explanation and the infographic, the continue button was disabled for a specified minimum amount of time. If the participant reported any technical difficulties with starting the animation, they were shown a non‐animated infographic (depicting only the final outcome in the three environments). The control group received neither a text‐based nor an animated explanation of herd immunity. All of the materials have been developed by the authors.

Open in a separate window

Figure 1

An example slide from the animated infographic.

Note. This slide depicts the final outcome in the environment in which many people were vaccinated. The slide reads: ‘When most of the population is vaccinated … it provides a disease barrier and limits the spread of disease’. The legend shows three colours representing ‘not vaccinated but still healthy’, ‘vaccinated and healthy’, and ‘not vaccinated, sick and contagious’ individuals. All of the slides are available at https://osf.io/4hyjt. The animated infographic in full can be viewed at https://my.visme.co/projects/010jd830‐project‐animated.

Outcome variable

All participants were faced with a vaccination decision task, which informed them about a severe fictitious disease and a fictitious vaccine. The use of fictitious materials excludes potential confounding variables, such as real infections and vaccine side effects experienced or observed by an individual (e.g., Chapman & Coups, 2006; Lane, MacDonald, Marti, & Dumolard, 2018). Additionally, it allows unconstrained manipulation of descriptive‐norm and herd‐immunity threshold levels. Participants first learned about the name of the virus and the path of infection (smear infection). Following Connolly and Reb (2003), the symptoms of the infection and vaccine side effects were described as equally likely (appearing in a small number of cases) and as very similar in content to ensure equal perceived riskiness. The vaccine was described as being easily available at no out‐of‐pocket cost and as 100% effective against infection with the disease. The source of information was not disclosed, as mistrust in healthcare authorities, government, and pharmaceutical companies have been shown to affect vaccine acceptance (Yaqub, Castle‐Clarke, Sevdalis, & Chataway, 2014). Vaccination intention was assessed by asking participants ‘If you had the opportunity to get vaccinated against [name of the disease] immediately, what would you do?’, on a 7‐point scale ranging from 1 = I would definitely not get vaccinated to 7 = I would definitely get vaccinated.

Other measured variables

Hypotheses testing

Communicating herd immunity significantly increased vaccination intentions compared to the control (M = 5.7, SD = 1.7 vs. M = 5.3, SD = 2.0), F(1,541) = 6.97, p = .009, η² = 0.013 (Figure 2), supporting Hypothesis 1. The effect remained significant after controlling for sociodemographic variables, F(1,519) = 5.92, p = .018, η² = 0.011. After excluding the groups where the herd‐immunity threshold was communicated (and without any covariates included in the model), the effect was no longer significant, although it remained in the same direction, F(1,405) = 3.48, p = .063, η² = 0.009.

Open in a separate window

Figure 2

Vaccination intentions depending on whether herd‐immunity explanation was provided.

Note. Communicating herd immunity via text and animated infographic was effective in increasing vaccination intentions. The figure shows a raincloud plot with the distribution of the data and jittered raw data; the box plot indicates the interquartile range from the 25th to the 75th percentile, including the median; the mean (with 95% confidence interval) is plotted on top of the jittered points.

Exposure to descriptive norms did not influence vaccination intentions, F(2,267) = 0.05, p = .956, η² < 0.001, not supporting Hypothesis 2 (Figure 3). Neither low (M = 5.7, SD = 1.8) nor high norms (M = 5.8, SD = 1.7) were significantly different from the no‐coverage message (M = 5.7, SD = 1.7) (estimate = 0.05, SE = 0.26, p = .977, 95% CI [−0.55, 0.66] and estimate = 0.07, SE = 0.26, p = .954, 95% CI [−0.53, 0.68], respectively). There was also no difference between low and high norms, estimate = −0.02, SE = 0.26, p = .996, 95% CI [−0.63, 0.58]. Hypotheses 2a, 2b, and 2c were, therefore, not supported. The main effect of norms did not change after controlling for sociodemographic variables (F(2,255) = 0.05, p = .951, η² < 0.001), and neither did the differences between the levels.

Open in a separate window

Figure 3

Vaccination intentions depending on the levels of the descriptive norm message.

Note. Communicating different descriptive norm messages (no‐coverage vs. low‐coverage [20%] vs. high‐coverage [80%] message) alongside herd immunity was not effective in increasing vaccination intentions. The figure shows a raincloud plot with the distribution of the data and jittered raw data; the box plot indicates the interquartile range from the 25th to the 75th percentile, including the median; the mean (with 95% confidence interval) is plotted on top of the jittered points.

The presence of the herd‐immunity threshold did not influence vaccination intentions, F(1,267) = 0.22, p = .639, η² = 0.001, not supporting Hypothesis 3 (Figure 4). Intentions of the participants who were informed about the threshold (M = 5.8, SD = 1.8) were not significantly different from the intentions of the participants who were not informed about it (M = 5.7, SD = 1.7). This effect did not change after controlling for sociodemographic variables, F(1,255) = 0.45, p = .501, η² = 0.002.

Open in a separate window

Figure 4

Vaccination intentions depending on whether the herd‐immunity threshold was provided.

Note. Communicating the herd‐immunity threshold alongside herd immunity was not effective in increasing vaccination intentions. The figure shows a raincloud plot with the distribution of the data and jittered raw data; the box plot indicates the interquartile range from the 25th to the 75th percentile, including the median; the mean (with 95% confidence interval) is plotted on top of the jittered points.

This project was initially supported through the Junior Researcher Programme (http://jrp.pscholars.org) at the 2019 jSchool. AL was funded by the Ministry of Education, Science and Technological Development of the Republic of Serbia [179018]. This work is part of AL’s PhD thesis at the University of Belgrade under the supervision of Dr. Iris Žeželj, who claimed no authorship. We wish to thank Dr. Robert Böhm for his encouragement and insightful comments on the Stage 1 protocol.