Introduction
The COVID-19 pandemic has highlighted the critical issue of vaccine hesitancy, a challenge that predates the current crisis but has been significantly magnified in its wake. Vaccine hesitancy was identified by the World Health Organization (WHO) as one of the major threats to global health in 2019 (World Health Organization, 2019). The simultaneous proliferation of misinformation regarding vaccination has further complicated efforts to address this challenge, undermining public health initiatives aimed at controlling the spread of the virus.
Misinformation,Footnote 1 particularly through social media, spreads rapidly and influences public perception and behaviour (Vosoughi et al., Reference Vosoughi, Roy and Aral2018). The term ‘infodemic’ has been coined by the WHO to describe this overwhelming flood of information, both accurate and false, that makes it difficult for people to find trustworthy sources and reliable guidance (World Health Organization, 2021). Recent research has shown that exposure to vaccine-related misinformation leads to a significant decline in vaccine intentions (Loomba et al., Reference Loomba, de Figueiredo, Piatek, de Graaf and Larson2021). Additionally, vaccine-related misinformation on social media has been shown to propagate more swiftly than efforts to counteract it, leading to increased vaccine hesitancy (Ruggeri et al., Reference Ruggeri, Vanderslott, Yamada, Argyris, Većkalov, Boggio, Fallah, Stock and Hertwig2024). Therefore, there is a need for a deeper understanding of the impact of misinformation on vaccine decisions during pandemics.
Past research has shown that health behaviours such as vaccination are determined by an array of social and personal factors (Ajzen, Reference Ajzen1991; Fishbein and Ajzen, Reference Fishbein and Ajzen2010). One of the most influential social determinants of behaviour is social norms, defined as individuals’ perceptions of what behaviours are typical or desirable within their community (Ajzen, Reference Ajzen1991; Bicchieri, Reference Bicchieri2005; Cialdini et al., Reference Cialdini, Reno and Kallgren1990). Perceived social norms have been found to be a significant predictor of preventive behaviour during the COVID-19 (Latkin et al., Reference Latkin, Dayton, Kaufman, Schneider, Strickland and Konstantopoulos2022) and, more specifically, of vaccine uptake (Agranov et al., Reference Agranov, Elliott and Ortoleva2021; Jaffe et al., Reference Jaffe, Graupensperger, Blayney, Duckworth and Stappenbeck2022; Rimal et al., Reference Rimal, Ganjoo, Jamison, Parida and Tharmarajah2024). Additionally, recent research has shown that the influence of social norms on individuals’ behaviour during COVID-19 was found to be stronger or weaker depending on the reference group, such as family, friends, political groups and people in the city (Rabb et al., Reference Rabb, Bowers, Glick, Wilson and Yokum2022). Moreover, the descriptive normative messages highlighting the high vaccination rates within the community have been found to be successful in increasing vaccine uptake in the targeted population (Moehring et al., Reference Moehring, Collis, Garimella, Rahimian, Aral and Eckles2023). However, there has been little exploration into the informational influences on the perception of social norms during a pandemic such as how perception of social norms itself might have been influenced by exposure to misinformation about vaccines.
On the other hand, research exploring vaccine misinformation on social media has evaluated its direct impact on intentions to vaccinate while ignoring the indirect influence via social factors such as social norms, which ultimately shape vaccination intent. It would be reasonable to infer that vaccine misinformation has indirect influences on vaccine intentions because research has shown that the influence of health information on behavioural intentions during the COVID-19 pandemic operates through a complex network of mediating factors, rather than having a direct effect (Hong and Kim, Reference Hong and Kim2020). However, this complex nature of information processing has been ignored when studying the impact of vaccine misinformation on social media. When considering vaccine misinformation’s effect on intentions, there are broadly two reasons to believe that vaccine misinformation may interact with perceived social norms of vaccination to ultimately influence vaccine intent. First, the inherently ‘social’ nature of the social media makes the misinformation or the opposition to vaccination ‘common knowledge’ (Chwe, Reference Chwe2013) and it may cause people to infer that scepticism of vaccination is a norm leading to individuals aligning their intentions with the perceived norm (Ajzen, Reference Ajzen1991; Cialdini and Trost, Reference Cialdini and Trost1998). Arias (Reference Arias2019) provides experimental evidence of social influence of the public dissemination of information on individuals’ perceptions of social norms. Furthermore, Morris and Shin (Reference Morris and Shin2002) explain that publicly available information allows individuals to update both their personal beliefs and their understanding of shared beliefs within their community. Secondly, specifically for misinformation, Lewandowsky et al. (Reference Lewandowsky, Ecker, Seifert, Schwarz and Cook2012) highlight that ‘Repetition effects may create a perceived social consensus even when no consensus exists’ (p. 113). Similarly, for vaccine misinformation on social media, repetition effects may lead to a perceived consensus against vaccination when claims highlighting threatful characteristics of vaccines are repeated. So, the repeated nature of false claims about vaccines and ‘public’ nature of social media indicates that there might be a ‘social mechanism’ (Arias, Reference Arias2019) at play in how vaccine misinformation shapes intent to vaccinate. Therefore, the present study proposes and tests a theoretical model that vaccine-related misinformation shapes perceived social norms and these perceived social norms ultimately shape vaccine intentions at least partially. Understanding of this mediation mechanism can allow for the development of more effective strategies that not only counteract misinformation but also address the distorted perceptions of social norms it creates.
Specifically, the present study tested whether first-order normative beliefs (individuals’ perceptions of others’ intentions) and second-order normative beliefs (individuals’ perceptions of others’ beliefs about vaccine safety) mediate the relationship between misinformation exposure and vaccine intentions. By employing a causal mediation analysis, this research sought to dissect the total effect of misinformation on intention to vaccinate into direct and indirect effects through social norm perceptions, providing a more comprehensive understanding of the mechanisms at play.
Methods
Experiment design and procedure
The experiment was programmed using oTree – an open-source platform designed for behavioural research (Chen et al., Reference Chen, Schonger and Wickens2016). Data were collected using the online experiment platform Prolific (www.prolific.com). Participants were compensated for participation at a rate of £12 per hour. On average, it took participants approximately seven to eight minutes to complete the study, based on the median time taken reported by the Prolific. The ethical approval for the present study was granted by the Biomedical Research Ethics Committee at the University of Warwick, UK.
This study used a pre-registered online survey experiment design incorporating both between-group and within-group comparisons of normative beliefs and vaccine intentions in a hypothetical pandemic scenario involving a fictitious virus, COVID-66. The use of a hypothetical virus scenario allowed the study to avoid any potential post-experiment disinformation given the false claims presented during the experiment and the ongoing COVID-19 pandemic. Participants were randomly assigned to either the ‘Misinformation’ condition or the ‘No Misinformation’ condition.
Participants in the ‘Misinformation’ group were exposed to five social media posts containing false claims about COVID-66 vaccines between the pre- and post-surveys. The social media posts were carefully designed to mimic real COVID-19 misinformation found on platforms like Facebook and Twitter and fact-checked by fullfact.org, a UK-based fact checker. The control group did not see these posts. The experimental flow, including participant group assignments and the presentation sequence of misinformation, is detailed in Figure 1. Participants were debriefed at the end of the experiment to ensure they were aware of the fictitious nature of the virus and the purpose of the misinformation exposure.
Figure 1. Experimental design showing experiment flow with randomization into the treatment and control groups.
Sample
A total of 332 participants who were all residents of the UK were recruited for an online experiment using the Prolific.com platform. The sample size was determined using G*Power 3.1 software to ensure adequate power for detecting significant effects (desired power = 0.95, alpha level = 0.05, effect size f = 0.1). Participants were pre-screened based on their opinions about COVID-19 vaccines, pre-collected by Prolific, and an equal number of participants with attitudes neutral, for and against COVID-19 were included in the study before randomization into the treatment and control groups. An equal quota of male and female participants was ensured. The demographic details of the participants, such as age, gender, ethnicity and education level, were also collected to ensure a diverse sample representative of the UK population.
Measures
The assessment of social norms included both empirical expectations (beliefs of vaccination intentions of others) (Bicchieri, Reference Bicchieri2017; Vriens et al., Reference Vriens, Tummolini and Andrighetto2023) and a second-order normative beliefs, which is the belief about others’ safety beliefs (Jachimowicz et al., Reference Jachimowicz, Hauser, O’Brien, Sherman and Galinsky2018). The personal vaccination intentions were elicited sequentially after assessing social norms. First-order normative beliefs (empirical expectations about others’ intentions to get vaccinated) were measured by asking: ‘On a scale from 0 to 100 percent, how many other people in today’s experiment will choose to get vaccinated in this scenario?’ Second-order normative beliefs (perceptions of others’ beliefs about vaccine safety) were assessed by asking: ‘On a scale from 0 to 100 percent, what are the chances that most people in today’s experiment believe vaccines are safe in the given scenario?’
Vaccine intentions were assessed on a 0–100% scale by posing the question: ‘On a scale from 0 to 100 percent, what are the chances that you will decide to get vaccinated in the given scenario?’ The scale was presented as 10 interval choices, such as 0–10%, 10–20%, until 90–100% to avoid focal responses of 0, 50 and 100; this has been recently used by Vriens et al. (Reference Vriens, Tummolini and Andrighetto2023). Additionally, participants were asked about their perceptions of their reference groups’ beliefs about the safety of vaccines, including family, friends, neighbours and work colleagues. Figure 2 shows some examples of misinformation used in the experiment. Specific details of the experiment can be accessed via experimental screenshots in the Supplementary material.
Figure 2. Examples of experimental stimuli used to expose participants to misinformation.
Causal mediation analysis
Causal mediation analysis was conducted using the inverse odds ratio weighting (IORW) approach (Nguyen et al., Reference Nguyen, Osypuk, Schmidt, Glymour and Tchetgen Tchetgen2015; Tchetgen Tchetgen, Reference Tchetgen Tchetgen2013), which is robust to multiple mediators and binary exposure variables. Difference scores of vaccine intentions and normative beliefs were calculated to assess the change due to misinformation exposure. Bootstrapped confidence intervals with 1,000 replicates were used to ensure robust estimation of the mediation effects. The analysis also considered the role of normative beliefs about specific reference groups, such as friends and neighbours, in mediating the effect of misinformation on vaccine intentions. Specific steps involved in conducting mediating analysis using IORW and its assumptions can be found in the Appendix.
Results
Sample characteristics
The study involved 332 participants, with equal numbers in the misinformation and control groups. The median age was 38 years, with 50% female and 50% male participants. The ethnic composition was predominantly White (84%), with other groups including Black (6.3%), Asian (5.7%), Mixed (2.7%) and Other (1.2%). Employment and literacy levels varied, ensuring a diverse sample. Key covariates like vaccine opinions, vaccine status and social media usage were balanced across groups.
Effect of misinformation on vaccine intentions
Misinformation significantly reduced vaccine intentions. A mixed ANOVA showed a significant interaction between time and condition (F(1, 330) = 5.02, p = 0.03, ηp2 = .015). In the misinformation group, average vaccine intentions dropped from 6.24 (SE = 0.29) to 5.93 (SE = 0.29) post-exposure. In contrast, the control group showed minimal change from 6.08 (SE = 0.29) to 6.02 (SE = 0.29). Post hoc comparisons confirmed a significant difference in the average change of vaccine intentions between groups, i.e., the vaccine intentions in the treatment group declined significantly more, on average, than decline in the control group (M diff = −0.25, p = 0.024, d = 0.24). Along with ANOVA Table 1 shows results of regression containing average treatment effect.
Table 1. Regressions treatment variable on changes in vaccine intentions with covariates
* p < 0.05,
** p < 0.01,
*** p < 0.001.
The mean change of own vaccine intent is shown in Figure 3. The distribution of pre–post differences in own intent in each group is illustrated in Figure 4.
Figure 3. The pre–post mean of own vaccine intentions by condition: beliefs about others' vaccine intent and beliefs about others’ vaccine safety beliefs grouped by condition.
Figure 4. The distribution of pre–post changes in vaccine intent grouped by condition.
Effect of misinformation on normative beliefs
First-order normative beliefs (empirical expectations about others’ vaccination intentions) were significantly affected by misinformation. A mixed ANOVA revealed a significant interaction between time and condition (F(1, 330) = 4.23, p = 0.04, ηp2 = .013). In the misinformation group, the average beliefs decreased from 6.55 (SE = 0.16) to 6.40 (SE = 0.17), whereas the control group saw an increase from 6.40 (SE = 0.16) to 6.59 (SE = 0.17). The difference in the pre–post change of first-order normative beliefs between groups was statistically significant (M diff = −0.35, p = 0.03, d = 0.23) as shown in Figure 5. Figure 6 shows the pre–post change in overall distribution of first- and second-order beliefs in each condition.
Figure 5. The pre–post mean of beliefs about others’ vaccine intent and beliefs bout others’ vaccine safety beliefs grouped by condition.
Figure 6. The distribution of pre–post changes in normative beliefs about others’ vaccine intent and others’ vaccine safety beliefs grouped by condition.
Second-order normative beliefs (perceptions of others’ beliefs about vaccine safety) showed no significant interaction between time and condition (F(1, 330) = 1.90, p = 0.16), indicating that misinformation had a weaker effect on these beliefs (Figure 5). However, on average, the pre–post change in these beliefs was negative compared to the control group, which shows a slightly positive change as shown in Figure 5.
Mediation analysis
Causal mediation analysis using the IORW approach highlighted the joint causal mediation of first- and second-order normative beliefs. The indirect effect of misinformation on vaccine intentions through change in first-order normative beliefs alone was considered and was significant (b = −0.95, 95% CI [−2.66, −0.01]), explaining 39.52% of the total effect (b = −2.40, 95% CI [−4.72, −0.41]) (Figure 7). When changes in both general first- and second-order beliefs were included as mediators, the joint indirect effect remained significant (b = −1.05, 95% CI [−3.53, −0.11]), accounting for 43.93% of the total effect (Figure 8). These mediation effects are detailed in Table 2.
Figure 7. A directed acyclic graph representing direct, indirect and the total effect from mediation analysis. ‘*’ indicates a ρ-value less than 0.05, while ‘**’ indicate a ρ-value less than 0.01.
Figure 8. A directed acyclic graph representing joint mediation of first- and second-order normative beliefs. ‘*’ indicates a ρ-value less than 0.05, while ‘**’ indicate a ρ-value less than 0.01.
Table 2. Varying indirect effects of misinformation on intentions odds weighting approach (IORW) with 1,000 bootstrapped resampling and bias corrected and accelerated (BCa) 95% confidence intervals
Mediation analysis also assessed the role of changes in normative beliefs about specific social reference groups (friends, neighbours, work colleagues and the general population in city). The joint indirect effect of change in first-order beliefs along with change in second-order beliefs about neighbours was highest, mediating 56.91% of the misinformation effect on vaccine intentions (b = −1.36, 95% CI [−3.81, −0.12]). Similar high proportions were found for friends (56.24%) and work colleagues (52.66%), indicating that changes in perceptions about the beliefs of close social groups post misinformation exposure significantly mediated the effect of misinformation exposure on vaccine intentions. The pre–post distribution of these beliefs by reference group can be found in the Appendix in Figure A2.
Discussion
This study examined how vaccine misinformation impacts vaccine intentions in a pandemic scenario involving a hypothetical virus, COVID-66. The results align with prior research, demonstrating that misinformation significantly reduces vaccine intentions (Loomba et al., Reference Loomba, de Figueiredo, Piatek, de Graaf and Larson2021; Roozenbeek et al., Reference Roozenbeek, Schneider, Dryhurst, Kerr, Freeman, Recchia, van der Bles and van der Linden2020). The study extends previous findings by showing the indirect effect of vaccine misinformation by highlighting the role of perception of social norms as intermediaries in the impact of misinformation exposure on intentions to vaccinate during a pandemic.
Interpretation of findings
The present study showed a small effect of vaccine misinformation in changing not only individuals’ vaccine intentions but also perceived social norms which ultimately influence intentions. First-order normative beliefs (perceptions of others’ intentions) were more influential than second-order beliefs (perceptions of others’ beliefs about vaccine safety) (Cialdini et al., Reference Cialdini, Reno and Kallgren1990; Cialdini and Trost, Reference Cialdini and Trost1998; Jachimowicz et al., Reference Jachimowicz, Hauser, O’Brien, Sherman and Galinsky2018). The mediation analysis revealed that changes in first-order normative beliefs due to misinformation explained about 40% of the total effect on vaccine intentions (Fishbein and Ajzen, Reference Fishbein and Ajzen2010). When both first- and second-order beliefs were considered, the mediation effect increased to 43.93%. Furthermore, when specific reference groups were considered, changes in beliefs about neighbours mediated 56.91% of the misinformation effect on vaccine intentions, followed by friends (56.24%), work colleagues (52.66%) and city population (50.62%).
Implications for public health interventions
The findings of this study suggest that public health strategies aimed at correcting vaccine misinformation online should also consider misinformation’s social influence along with mitigating the informational influence through provision of accurate information. The present findings showing that misinformation may shape one’s belief system – specifically normative beliefs – which serve as an indirect pathway influencing vaccination intentions. This contributes to recent discussion on role of multiple social factors for misinformation’s continued influence even after correcting, for example factors like worldview – contradiction of misinformation correction with one’s belief system (Ecker et al., Reference Ecker, Lewandowsky, Cook, Schmid, Fazio, Brashier, Kendeou, Vraga and Amazeen2022). However, the present study suggests that misinformation might shape one’s worldview – that is, one’s belief system, particularly normative beliefs – which may not be effectively targeted when corrections address only factual inaccuracies. Hence, this study highlights the need for additional targets like reshaping ‘perception of social norms’ for misinformation corrections online to counter the indirect influences in addition to reducing ‘belief in misinformation’ by providing accurate information.
Therefore, the direct recommendation for public health messaging aimed at countering vaccine misinformation is to adopt a dual approach: correcting misinformation while also promoting clear normative messages that emphasize positive vaccine intentions and safety beliefs. Messages that explicitly highlight the high vaccination uptake among communities alongside refutation of false claims are likely to be effective in counteracting misinformation’s indirect effect and direct effects. This aligns with recent findings that pairing factual corrections with positive social norm information can more effectively change beliefs (Ecker et al., Reference Ecker, Sanderson, McIlhiney, Rowsell, Quekett, Brown and Lewandowsky2023).
Limitations and future research
Using a hypothetical virus scenario allowed controlled experimentation but may not capture real-world vaccine decisions fully. Future research should replicate these findings in naturalistic settings across diverse populations. The study’s sample was from the UK but not a representative one, so generalizability to UK and other cultural contexts needs further examination. Additionally, future studies should explore long-term effects of misinformation and changes in normative beliefs over time along with investigating potential moderating factors of indirect effects like correction to misinformation and individual susceptibility to misinformation (van der Linden, Reference van der Linden2022).
Conclusion
Vaccine misinformation reduces vaccine intentions by altering first-order and second-order normative beliefs. Public health interventions should integrate misinformation correction (Lewandowsky and van der Linden, Reference Lewandowsky and van der Linden2021) with efforts to shape positive social norms including what other people believe about the safety of vaccines along with what they intend to do. Targeted messaging leveraging the influence of close social groups may be particularly effective not only in promoting vaccine uptake but also in mitigating the indirect impact of vaccine misinformation. Future research should explore the robustness of dynamic interplay between misinformation, social norms and vaccine intentions to inform robust public health strategies.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/bpp.2025.10012.
Appendix
Assumptions of IORW
The validity of the estimates obtained through causal mediation analysis depends on an important assumption known as Sequential Ignorability (Imai et al., Reference Imai, Keele and Tingley2010). This assumption can be divided into two main conditions.
Assumption of ignorability of treatment assignment
The first assumption requires that treatment assignment is statistically independent of potential outcomes and mediators, once conditioned on observed pretreatment covariates. This is satisfied in the present study due to its randomized experimental design, which ensures that treatment assignment is independent of confounders. Randomization guarantees that, on average, any differences between the treatment and control groups are attributable to the treatment itself rather than other factors. Therefore, this aspect of sequential ignorability is inherently fulfilled by design.
Assumption of ignorability of the mediator
The second part of the sequential ignorability assumption asserts that the mediator must be independent of unobserved confounders, conditional on observed treatment and pretreatment covariates. Unlike the first assumption, this is more challenging to satisfy since mediators are not randomly assigned. In this study, this assumption implies that after accounting for treatment assignment and relevant covariates, the mediator behaves as if randomly assigned. Nevertheless, unobserved confounders may still influence the mediator–outcome relationship.
To address potential violations of this second assumption, sensitivity analysis was conducted. Sensitivity analysis helps determine how much unobserved confounding would be required to significantly alter the conclusions of the mediation analysis. This approach provides additional evidence on the robustness of the causal mediation estimates, even in the presence of unobserved confounders.
By varying the correlation (ρ) between the error terms of the mediator and outcome models, sensitivity analysis allows us to simulate how unmeasured confounders might impact the mediation effect. This helps assess whether the mediation results hold across different levels of possible confounding, providing a nuanced understanding of how unobserved factors might alter the indirect effects.
Need for sensitivity analysis – robustness checks
While the IORW approach helps mitigate the confounding impact of mediators, it still relies on the assumption that all important confounders have been accounted for. In practice, this assumption can be difficult to verify, particularly in complex behavioural or social settings where unmeasured confounders may exist. Therefore, a sensitivity analysis was performed to determine how robust the estimated indirect effects are to unobserved confounding.
Given the practical impossibility of measuring every potential confounder, sensitivity analysis has become a consensus approach to assess the robustness of mediation results. Sensitivity analysis involves varying the correlation (ρ) between the error terms of the mediator and outcome models to simulate the possible influence of unmeasured confounders. This helps determine whether the mediation effects are robust under different levels of potential confounding and provides a more nuanced understanding of how unobserved factors might alter the indirect effects.
Steps for conducting a sensitivity analysis
The sensitivity analysis was conducted using the R mediation package (Tingley et al., Reference Tingley, Yamamoto, Hirose, Keele and Imai2014) (Tingley et al., Reference Tingley, Yamamoto, Hirose, Keele and Imai2014). The analysis began with two models: a mediator model predicting FirstBelief diff (difference score in First Belief) from Condition and an outcome model predicting VaccineIntention diff (difference score in Vaccine Intention) from Condition, FirstBelief diff and their interaction. The mediate() function estimated the Average Causal Mediation Effect (ACME) through 10,000 bootstrap simulations with bias-corrected and accelerated confidence intervals.
The sensitivity analysis was performed using the medsens() function, which evaluates how the ACME would change in the presence of an unmeasured confounder. The sensitivity parameter ρ represents the correlation between the residuals of the mediator and outcome models, with values ranging from −1 to 1 examined at 0.1 intervals. For each value of ρ, the function computed the corresponding R 2 values that would be required for an unmeasured confounder to explain that level of residual correlation.
The R code for sensitivity analysis is available at https://osf.io/wrhvn/.
Sensitivity analysis – robustness checks for indirect effects in the present study
The sensitivity analysis was conducted for the first mediator, i.e. first-order normative beliefs. A large critical ρ value is needed to reverse the sign of ACME, indicating the robustness of the result to the violation of ignorability assumption. Table A1 shows the robustness assessment for first mediator. Figure A1 graphically represents the sensitivity region for the critical ρ value.
Figure A1. Sensitivity analysis plot.
Table A1. Sensitivity region for the average causal mediation effect (ACME)
Robustness assessment
Discussion
The sensitivity analysis for the treatment group’s indirect effect demonstrates notable robustness to unmeasured confounding. The effect requires a relatively strong correlation (ρ = 0.4) to change its sign from negative to positive, indicating that substantial confounding would be necessary to nullify the observed mediation effect. This transition point corresponds to an R 2_MR 2_Y value of 0.16 and an R 2_M of 0.1417, suggesting that confounding variables would need to explain approximately 14–16% of the unexplained variance in both the mediator and outcome to invalidate the mediation finding.
The graphical representation further supports this robustness, showing a gradual slope in the estimated effect across values of ρ, with relatively constrained confidence intervals near the critical region where ρ = 0.4. The width of these confidence intervals remains manageable until more extreme values of ρ, indicating reasonable precision in our estimates across a meaningful range of potential confounding.
These findings suggest that while the indirect effect in the treatment group is not immune to unmeasured confounding, it demonstrates sufficient robustness to support cautious causal interpretations. The magnitude of confounding required to overturn the observed effect is substantial enough to lend credibility to the mediation pathway identified in the treatment.
Conditional average treatment effects
As an exploratory analysis and as a robustness check, a conditional average treatment effects analysis was conducted with an interaction of treatment variable and the change in first-order normative belief on the change in vaccine intention along with demographic covariates. The results, presented in Table A2, show that the misinformation treatment’s effect on vaccine intentions is significantly moderated by changes in first-order normative beliefs (β = 0.198, p = 0.006).
Table A2. Regressions of the treatment effect of misinformation on changes in vaccine intentions conditional on the change in first-order normative beliefs
* p < 0.05,
** p < 0.01,
*** p < 0.001.
The analysis reveals that the effect of misinformation treatment on vaccine intentions is larger for a higher shift in first-order normative belief. Specifically, the baseline effect of the misinformation treatment (when there is no change in normative beliefs) is −2.35 % (p = 0.025). The significant positive interaction term indicates that for each one-unit decrease in normative beliefs, the negative treatment effect is amplified by 0.198 %. This analysis provides additional evidence for the proposed indirect mechanism: the magnitude of negative shifts in perceived social norms is associated with stronger negative effects of misinformation on vaccine intentions. These findings complement the mediation analysis by demonstrating the same relationship through a different analytical approach.
Figure A2. The pre–post distribution of Normative beliefs by reference group.
Open access
