key: cord-0290155-jj7azihv authors: Sudharsanan, N.; Favaretti, C.; Hachaturyan, V.; Baernighausen, T.; Vandormael, A. title: Effects of Side-Effect Risk Framing Strategies on COVID-19 Vaccine Intentions in the United States and the United Kingdom: A Randomized Controlled Trial date: 2021-10-14 journal: nan DOI: 10.1101/2021.10.12.21264877 sha: d6ffe70fc4f81eeab6d8564bd9f80b8efb114f87 doc_id: 290155 cord_uid: jj7azihv Vaccination rates have stagnated in the United States and the United Kingdom leading to the continuing spread of COVID-19. Fear and concern over vaccine side-effects is one of the main drivers of hesitancy. Drawing from behavioral science and health communication theory, we conducted a randomized controlled trial among 8998 adults to determine whether the way COVID-19 vaccine side-effects are framed and presented to individuals can influence their willingness to take a vaccine. We presented participants information on a hypothetical future COVID-19 vaccine -- including information on its side-effect rate -- and then examined the effect of three side-effect framing strategies on individuals stated willingness to take this vaccine: adding a qualitative risk label next to the numerical risk, adding comparison risks, and for those presented with comparisons, framing the comparison in relative rather than absolute terms. Based on a pre-registered and published analysis plan, we found that adding a simple descriptive risk label (very low risk) next to the numerical side-effect increased participants' willingness to take the COVID-19 vaccine by 3.0 percentage points (p = 0.003). Providing a comparison to motor vehicle mortality increased COVID-19 vaccine willingness by 2.4 percentage points (p = 0.051). These effects were independent and additive: participants that received both a qualitative risk label and comparison to motor-vehicle mortality were 6.1 percentage points (p < 0.001) more likely to report willingness to take a vaccine compared to those who did not receive a label or comparison. Taken together, our results reveal that despite increasingly strong vaccination hesitancy and exposure to large amounts of vaccine misinformation, low-cost side-effect framing strategies can meaningfully affect vaccination intentions at a population level. Vaccination is one of the main strategies for controlling the COVID-19 pandemic. However, vaccination rates have slowed and are far from target levels in countries like the United States and the United Kingdom. For example, in the United States, the share of the population that is fully vaccinated went from 1.8% in February 2021 to 40.4% in May 2021 but has since only risen to 54% in the following four months [1] . While not as stagnant, there is a similar pattern in the United Kingdom, where the share of the population that is fully vaccinated was just 65.6% as of September 27th, 2021 [1] . These vaccination trends are insufficient to prevent the spread of COVID-19, especially the Deltavariant, which has re-ignited the pandemic in both countries [1] [2] [3] . Vaccine hesitancy is not the result of a single homogenous cause and can vary for different individuals and population groups. For example, recent studies have identified several potential reasons for COVID-19 vaccine hesitancy, including a low perceived risk of COVID-19 infection and concern around how quickly the vaccines were developed [4] [5] [6] [7] . A common finding across these studies is that fear and concern about vaccine side-effects is an important reason for vaccine hesitancy. These concerns were potentially heightened by widespread media coverage of vaccine side-effects in April and May 2021, along with the pausing of vaccination efforts in several countries due to this media coverage [8, 9] . Although COVID-19 vaccine side-effects rates are extremely low (1 per 100,000 people vaccinated with AstraZeneca in the European Union) [10] , these rates were often presented by the media without context [9, [11] [12] [13] , likely leading some individuals to reject vaccine uptake [14] . In addition, the pausing of global COVID-19 vaccination efforts may have sent a strong signal that side effects are a major cause of concern, thus increasing vaccine hesitancy. Addressing public concerns over vaccine side-effects will be a key component of efforts to improve vaccine use in the United States, United Kingdom, and globally --especially as new vaccines are released. There is a large body of evidence in the health communication and the behavioral sciences that shows that how risks are framed and presented to individuals can affect their perceptions of its severity and ultimately their behavior [15] [16] [17] . For example, studies have shown that whether numerical risks are presented as percentages or natural frequencies (e.g. 1% compared to 1 out of 100), with a comparison to a commonly understood but different risk (e.g. the risk of motor-vehicle mortality), using infographics (e.g. visually showing the numbers of individuals in the numerator and denominator of a risk), with descriptive labels (e.g. putting "very low risk" or "high risk" next to the numerical risk) can all influence behavior. Therefore, simple changes to the framing of COVID-19 vaccine side-effect risks may have a meaningful influence on the number of people who choose to get vaccinated. Such "nudges" are viewed positively by policymakers since they are often affordable to implement and have the potential for wide-population reach [18] . These types of framing effects do not affect behavior by changing deeply held attitudes and beliefs, or by creating strong incentives for a certain behavior. Rather, research has shown that individuals have a limited cognitive ability to process and internalize risk information, especially rare risks like COVID-19 side-effects, and that individuals use mental guides or "heuristics" to make sense of risk information that ultimately guides their decisions [19] [20] [21] . Therefore, framing effects work by modifying the heuristics that individuals use to understand a risk in a way that does not strongly affect their incentives or remove their agency. What remains unknown to date, however, is what the effect of such framing effects are on COVID-19 vaccine intentions, which framing strategies are most promising, and whether such nudges can even influence COVID-19 vaccination behavior. Given that vaccination has become socially and politically charged, affecting hesitancy may no longer be amenable to light-touch nudge and framing interventions. Relatedly, it is an open question whether such framing effects will still be effective given that individuals have already experienced months of side-effect-related media coverage. To answer these questions, we conducted a randomized controlled trial evaluating the effect of three framing strategies on COVID-19 vaccine intentions among 8988 adults ages 18+ (4502 adults in the United States and 4496 in the United Kingdom). We presented participants information on a hypothetical "future COVID-19 vaccine" including information on its side-effect rate (we chose a side-effect rate that was comparable to the current vaccines) [22, 23] . We examined three side-effect framing strategies: what is the effect of adding a qualitative risk label next to the numerical risk, what is the effect of adding a comparison group (along with which comparison group is most effective), and for those with comparison groups, what is the effect of framing the comparison in relative rather than absolute terms (absolute comparisons of small risks may be cognitively harder for individuals to process than relative comparisons). Based on a pre-registered and published analysis plan [24] , we then evaluated the effect of these framing strategies on two outcomes: self-reported willingness to take the hypothetical vaccine, and as a measure of the mechanism of our framing effects, individuals' perceived safety of the vaccine. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 14, 2021. ; Figure 1 presents the results of the three main framing strategies on the proportion of participants that report that they would take the hypothetical vaccine (the results are presented on the percentage point scale). Among all the strategies, we find that adding a simple qualitative risk label ("very low risk") next to the numerical risk increased vaccine intentions by 3.0 percentage points (95% CI: 1.0, 4.9; p = 0.003; control group mean: 65%). For our two comparison strategies, we found that adding a comparison to motor-vehicle mortality (Effect size: 2.4 percentage points; 95% CI: 0.0, 4.7; p = 0.051; control group mean: 65%) had a larger impact on vaccine intentions than a comparison to COVID-19 mortality (Effect size: 0.7 percentage points; 95% CI: -1.6, 3.0; p = 0.568; control group mean: 65%). Although we were not able to reject the hypothesis that these effects were statistically different, this is a surprising result as we expected the comparison to COVID-19 mortality to be more salient both because it is substantially higher than motor-vehicle mortality and because it is the form of mortality directly related to vaccination. In the US, for example, motor vehicle mortality in 2020 was 12 per 100,000 population while COVID-19 mortality in the same year was 170 per 100,000 population. Lastly, going against our expectation, we did not find strong evidence that relative, [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] . This secondary outcome serves to investigate whether the effects we observed on vaccine intentions were driven through perceptions of safety (providing more information may affect vaccine intentions, for example, by increasing individuals' trust in the source of information rather than directly affecting their perceptions of vaccine safety). Although the magnitude of the effect sizes is small, we find that the pattern of the framing effects on perceptions of vaccine safety closely mirrors the primary effects on willingness to take the vaccine. This suggests that perceptions of safety are indeed one of the pathways through which these framing effects operate, although our analyses do not exclude the possibility that by providing more information, these framing strategies also affect vaccine intentions by increasing the perceived trustworthiness or reliability of the information. Effect of the vaccine side-effect framing strategies on the probability that participants report that they would take the hypothetical vaccine and their perceptions of how safe the hypothetical vaccine is (N = 8998 for Strategy 1 and 2); Strategy 3 is estimated only among those that received a comparison risk (N = 5998). Control means for strategies 1 and 2 are 65% and 67% for strategy 3. An important question is how effective these strategies are when combined. For example, it could be the case that vaccine intentions are only movable by a fixed margin, such that even when both qualitative risk labels and comparison groups are used, the resulting change in vaccine intentions is less than the sum of the independent effects. Conversely, the effects may be synergistic such that combining strategies leads to larger effects than the sum of the independent effects. To assess this consideration, in a non-pre-registered analysis, we estimated the effect of adding both a qualitative risk label and comparison to motor vehicle mortality on willingness to take the hypothetical COVID-19 vaccine ( Table 1) . We find that at a minimum, the effects are independent and additive, with some indication that they may even be synergistic. Compared to those that received neither a quality risk label nor a comparison risk (N = 1499; control mean: 63%), those that received both strategies were 6.1 percentage points (95% CI: 2.7, 9.5; p < 0.001) more likely to state that they would take the vaccine. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; to be risky or not [25] [26] [27] , which may affect how the framing effects impact behavior. We do not find evidence that the framing strategies vary across country, sex, or education. Importantly, due to our sample sizes, we cannot rule out the possibility of potentially small interaction effects. Therefore, our results should be interpreted as not finding evidence for any large differences. Although estimated with a confidence interval that overlaps the null, we find suggestive evidence that . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; the framing strategies may have different effects by age. Among those aged 40+, a qualitative risk comparison has a 2.2 percentage point (95% CI: -6.8, 2.4; p = 0.351) smaller effect on willingness to take the vaccine, compared to those aged < 40. Conversely, adding a comparison group may have a larger effect for those aged 40+, with an effect size that is 3.9 percentage points (95% CI: -1.8, 9.6; p = 0.177) greater for a motor-vehicle comparison and 3.4 percentage points (95% CI: -2.3, 9.1; p = 0.241) greater for a COVID-19 comparison when compared to those aged <40. These results suggest that applying different framing strategies depending on the age of targeted individuals may have greater impact than blanket population strategies. However, this result is speculative and requires further evidence to tease apart. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint We find no evidence of imbalances in sociodemographic characteristics across the experimental groups (Supplemental Tables 1-3 ) and our main results remain unchanged after including these characteristics as control variables in the main regression analyses (Supplemental Table 4 ). Our results are also consistent when we examine the outcome as a four-category ordinal (rather than binary) variable (Supplemental Table 5 ), although the estimated effects become more difficult to interpret. We found that adding a simple descriptive risk label ("very low risk") next to the numerical side-effect increased participants' willingness to take the COVID-19 vaccine by 3.0 percentage points (p = 0.003). Providing a comparison to motor vehicle mortality increased COVID-19 vaccine willingness by 2.4 percentage points (p = 0.051). Importantly, we found that these effects were independent and additive: participants that received both a qualitative risk label and comparison to motor-vehicle mortality were 6.1 percentage points (p < 0.001) more likely to report willingness to take a vaccine compared to those who did not receive a label or comparison. This is an important and meaningful effect at the population level, where even small changes in vaccination rates can have large health consequences [28] . These results are especially reassuring considering the low cost and ease of implementing such framing strategies. Based on the effects on perceptions of vaccine safety, we find support for the hypothesis that these effects work by modifying individuals' judgments about how safe the vaccine is. Lastly, we found some suggestive evidence that comparison groups are more effective for older participants while descriptive risk labels were more effective for younger participants. Taken together, our results reveal that despite increasingly strong vaccination hesitancy and exposure to large amounts of vaccine information, low-cost side-effect framing strategies can meaningfully affect vaccination intentions at a population level. Given that vaccination for COVID-19 will likely remain an important priority for the foreseeable future, these insights can be valuable for increasing the uptake of future vaccination efforts. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint This was an online-based randomized controlled trial study. To be enrolled in the trial, participants had to meet the following inclusion criteria: be 18 years or older, have current residence in the United States or the United Kingdom, and be able to speak English. Participants who did not meet the inclusion criteria were not eligible to participate and were excluded from the study. We used Prolific [29] , an online-based service for recruiting participants for online-based research studies to recruit the study participants. Prolific participants are paid for their participation (we paid each participant £0.63/$0.88 for the expected 5-minute completion time). Within Prolific, we used a stratified quota sampling procedure to match the education distribution of our sample to the general population (stratified by sex and country) based on available data for each country [30, 31] . We aimed to recruit a total of 9000 total participants, split evenly by country and sex. 99% of our participants were recruited between 30 July 2021 and 10 August 2021. To reach our target quotas by sex and education, the final 1% of participants were recruited between 11 August 2021 and 4 October 2021. We previously published the protocol for our trial with a description of the experiment [24] . On the Prolific platform, potential participants were provided information that the aim of the study was to understand their willingness to take COVID-19 vaccines, the risks and benefits of the study, and how they could contact the researcher (and/or the human subjects review board at the Heidelberg University). After consenting on Prolific, participants were redirected to the Gorilla platform, an interactive web-based service for conducting behavioral science experiments. On the Gorilla page, we provided additional information on data protection. For participants that agreed to participate, we first collected basic sociodemographic information and then set up the experiment by telling participants that we are going to show them information on a hypothetical future COVID-19 vaccine and would like to know how willing they would be to take it. At this stage, we emphasized that their answers cannot be linked back to them in any way. Participants were then presented information on the vaccine and its side effect rate on a single page (Supplemental We used a factorial randomization design to assign participants to three main factors (Supplemental Figure 2) . Factor 1 was whether there was a qualitative risk label saying "very low risk" next to the numerical risk or not. Factor 2 was whether the risk was presented with no comparison, a comparison to motor-vehicle mortality, or a comparison to COVID-19 mortality. Factor 3 was only among that received a comparison risk and varied whether the comparison was presented in an absolute or relative way. We randomized participants to each factor independently (stratified by country), such that, for example, which comparison group a participant received did not depend on whether they received a risk label or not. This means that the proportion of participants that received a risk label will be balanced across the treatment groups for the comparison risks. We repeated this independent randomization procedure for whether the comparison risk is presented as an absolute or relative comparison among the 2/3rds of the sample that were randomized to receive any comparison risk. Similarly, this means that conditional on receiving any comparison risk (either motor vehicle or COVID-19 mortality), the proportion of participants that received each type of comparison, and the proportion that received a labeled risk will be balanced across the treatment groups for absolute and relative framings. Note that in practice, the Gorilla algorithm handled the randomization using a stratified randomization approach to prevent chance imbalances in the joint distribution of the factors. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint We had complete data for all 4502 participants from the USA (the extra two participants above our target of 4500 was the result of how Prolific and Gorilla recruit individuals). 4 individuals from the UK sample did not complete the experiments and were therefore excluded due to missing data. This resulted in a final sample of 8998 individuals. Prior to recruiting any participants, we received ethical approval for the study from the Medical Faculty of Heidelberg University Ethics Committee (#S-443/2021), registered the trial, including the outcomes and treatments, on the German Clinical Trials Registry (#DRKS00025551), and published a trial protocol with a pre-analysis plan [24] . The study here is missing one sub-analysis that we registered as part of the protocol that was based on a comparison to an additional 3000 participants outside of the 9000 randomized here as part of the main study. The reason for this is that there was an error in the study text for these additional 3000 participants; this error, however, had no impact on the 9000 participants recruited for the main study activities presented here. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint Supplemental Figure 2 Detailed randomization diagram across the main study factors. We have shown the randomization here with sample sizes referring to both countries combined; in practice, we conducted this procedure stratified by country such that each cell in the diagram has exactly 50% USA and 50% UK observations. Our main outcome was a binary variable for whether participants reported "yes" to the question of "Would you take this vaccine if it were made available to you?" Importantly we asked participants to answer as if they had not been vaccinated even if they already received some form of vaccination. In Supplemental Table 5 , we show the main study results using a four-category ordinal response rather than a binary response variable and find no change to our conclusions. Our secondary outcome was participants' perception of how safe the vaccine is, based on a scale of 1 (extremely unsafe) to 10 (extremely safe). This secondary outcome served to determine if the effects we saw on vaccine willingness were at least partly driven through perceptions and judgments of the vaccine's safety. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint We first assessed if the randomization was successful by comparing means and proportions of the main sociodemographic variables across the experimental arms for each of the randomization factors. Next, we estimated the main effects of the risk label and comparison group strategies using the following linear probability regression model: In this model, ! is the binary indicator for whether participant i said "Yes" to taking the hypothetical vaccine, ! is a binary indicator for whether participant i was shown a qualitative risk label, ! is a binary indicator for whether participant i was shown a comparison to motor-vehicle mortality, and ! is a binary indicator for whether participant i was shown a comparison to COVID-19 mortality. Our main effects of interest are 1 , 2 , and 3 respectively. Note that since we factorially randomized the risk label and comparison groups, we can estimate the main effects of these separate strategies in a single regression; however, our effect estimates were virtually unchanged when estimating the risk label and comparison effects using separate regressions. We then estimated the effect of a relative compared to absolute comparison framing by estimating the following regression just among those that received either the motor-vehicle or COVID-19 mortality comparison: Here ! is a binary indicator for receiving the relative framing and 1 is the corresponding effect of this framing on vaccine intentions compared to an absolute comparison. Based on our pre-registered analyses, we found evidence that adding a qualitative risk and label and adding a comparison to motor-vehicle mortality both increased the likelihood that participants reported that they would take the vaccine. In a non-pre registered analysis, we then assessed whether these two independent effects were additive by estimating the effect of receiving both strategies relative to receiving neither using the following regression model: ! is a binary indicator for those that received both a qualitative risk label and comparison to motor-vehicle mortality. To determine whether the main effects varied by country, sex, age group (below 40 and 40+), and education (Bachelor's degree or more compared to less than Bachelor's degree), we re-estimated . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint these two main regressions including interaction terms between the main experimental indicators and the heterogeneity variables (separately for each heterogeneity variable). Lastly, to heuristically determine whether these effects were partly mediated by perceptions of vaccine safety, we estimated regressions (1) and (2) using a continuous indicator for reported safety as the main outcome variable. As robustness and sensitivity analyses, we re-estimated each regression with addition controls for age, sex, country, and education (note that because participants were randomized, the addition of these controls should not change the main effects unless there were large chance imbalances across experimental groups) and using an ordinal logistic regression model with a four-category outcome variable for vaccine intention rather than a binary indicator for "yes." . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted October 14, 2021. ; . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Notes: Coefficients for the yes/no outcome are shown on the percentage point scale; coefficients for the mean safety outcome are shown as mean changes. Values in parentheses are 95% confidence intervals. Sample size for the relative to absolute comparison is smaller (N = 5998) since it is only estimated among those groups that received a comparison (omitting the no comparison group). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint Supplemental Figure 3 Screenshots of each experimental arm for United States participants. Note that we have labeled the arms for the figure but participants were not shown this label. Reference mortality information for motor-vehicle and COVID-19 fatalities were taken from the Centers for Disease Control and Prevention [32, 33] . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint Supplemental Figure 4 Screenshots of each experimental arm for United Kingdom participants. Note that we have labeled the arms for the figure but participants were not shown this label. Reference mortality information for motor-vehicle fatalities were taken from the Department of Transportation [34] and COVID-19 from the Public Health England [35] . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 14, 2021. ; https://doi.org/10.1101/2021.10.12.21264877 doi: medRxiv preprint Variants: distribution of case data COVID-19 Vaccine Hesitancy-A Scoping Review of Literature in High-Income Countries Predictors of COVID-19 vaccine hesitancy in the UK household longitudinal study COVID-19 Vaccine Hesitancy in Canada: Content Analysis of Tweets Using the Theoretical Domains Framework Understanding COVID-19 vaccine hesitancy What does Australia's new advice on the AstraZeneca Covid vaccine mean and what is the risk? The Guardian Possible Side Effects of AstraZeneca Vaccine Come Into Sharper Focus. The New York Times Global Advisory Committee on Vaccine Safety (GACVS) review of latest evidence of rare adverse blood coagulation events with AstraZeneca COVID-19 Vaccine (Vaxzevria and Covishield) Australia reports second blood clot case 'likely' linked to AstraZeneca Covid vaccine. The Guardian More blood clot cases following Johnson & Johnson COVID-19 vaccine reported. Fox News Mother-of-three, 47, dies after AstraZeneca Covid jab caused blood clots on her brain | Daily Mail Online Cable News and COVID-19 Vaccine Compliance. Social Science Research Network Current Best Practice for Presenting Probabilities in Patient Decision Aids: Fundamental Principles. Med Decis Mak Int J Soc Med Decis Mak Presenting quantitative information about decision outcomes: a risk communication primer for patient decision aid developers Evidence-Based Risk Communication Nudge: Improving Decisions about Health, Wealth, and Happiness. Rev. and expanded ed. Penguin Books Judgment Under Uncertainty: Heuristics and Biases Heuristic Decision Making Information Processing and Risk Perception: An Adaptation of the Heuristic-Systematic Model AstraZeneca's COVID-19 vaccine: benefits and risks in context. European Medicines Agency Recommends Use of Johnson & Johnson's Janssen COVID-19 Vaccine Resume The effect of framing and communicating COVID-19 vaccine side-effect risks on vaccine intentions for adults in the UK and the USA: A structured summary of a study protocol for a randomized controlled trial Gender differences in risk assessment: Why do women take fewer risks than men? Gender, race, and perceived risk: The "white male" effect Gender Differences in Fear and Risk Perception During the COVID-19 Pandemic Monitoring Incidence of COVID-19 Cases, Hospitalizations, and Deaths, by Vaccination Status -13 Educational Attainment in the United States: 2019. The United States Census Bureau Trends in Number of COVID-19 Cases and Deaths in the US Reported to CDC, by State/Territory Deaths in the UK | Coronavirus in the UK. GOV.UK. Accessed