Letter

Association of moral values with vaccine hesitancy

Received:
Accepted:
Published online:

Abstract

Clusters of unvaccinated children are particularly susceptible to outbreaks of vaccine-preventable disease1,2. Existing messaging interventions demonstrate short-term success, but some may backfire and worsen vaccine hesitancy3. Values-based messages appeal to core morality, which influences the attitudes individuals then have on topics like vaccination4,5,6,7. We must understand how underlying morals, not just attitudes, differ by hesitancy type to develop interventions that work with individual values. Here, we show in two correlational studies that harm and fairness foundations are not significantly associated with vaccine hesitancy, but purity and liberty foundations are. We found that medium-hesitancy parents were twice as likely as low-hesitancy parents to highly emphasize purity (adjusted odds ratio: 2.08; 95% confidence interval: 1.27–3.40). High-hesitancy respondents were twice as likely to strongly emphasize purity (adjusted odds ratio: 2.15; 95% confidence interval: 1.39–3.31) and liberty (adjusted odds ratio: 2.19; 95% confidence interval: 1.50–3.21). Our results demonstrate that endorsement of harm and fairness—ideas often emphasized in traditional vaccine-focused messages—are not predictive of vaccine hesitancy. This, combined with significant associations of purity and liberty with hesitancy, indicates a need for inclusion of broader themes in vaccine discussions. These findings have the potential for application to other health decisions and communications as well.

Main

Growing numbers of parents who request non-medical vaccine exemptions for their children indicate cause for concern about maintaining sufficiently high immunization coverage levels in the United States8,9,10. Although the trend appears to have stabilized and immunization coverage remains high overall11, the greater cause for concern is geographic clustering of unvaccinated or under-vaccinated children. When they are geographically clustered, partially or fully unvaccinated children are more susceptible to outbreaks of vaccine-preventable disease1,2, mainly because vaccine exemption rate levels are higher in these clusters than the overall state coverage levels12,13,14,15. Parents of such children are classified as vaccine-hesitant; the different types of hesitancy are based on the type and strength of their concerns about vaccination16,17. Individual immunization behaviours range from total vaccine acceptance to total vaccine refusal, with attitudes towards general vaccination or a specific vaccine influencing acceptance. Many vaccine-hesitant individuals demonstrate nuanced attitudes towards immunization rather than dichotomous positive or negative attitudes18. Such individuals may accept some vaccines but refuse or delay others19,20. Individuals express their hesitancy in three ways: (1) receive all vaccines but express concerns about vaccination, (2) selectively delay or refuse vaccines or (3) refuse all vaccines18.

Several classification systems exist for categorizing parental attitudes towards vaccination and identifying different types or strengths of vaccine hesitancy21,22. Gust et al.17 identified five profiles of attitudes towards vaccination—the immunization advocate, the ‘go along to get along’ parents, the health advocate, the fence sitter and the worried—based on a sample of parents with at least one child under seven years of age. Keane et al.23 classified parents using four groupings of vaccine attitudes: vaccine believers (those convinced about the benefits of vaccinating), cautious parents (those who actively seek information about vaccines but may experience anxiety about getting their child vaccinated), relaxed parents (those with relaxed parenting styles and without a particularly strong attitude towards vaccines) and unconvinced parents (those with the most negative attitudes towards vaccinations). A quantitative system, developed by Opel et al.16,19, assesses hesitancy using three different constructs—immunization behaviours, beliefs about vaccine safety and efficacy, and general attitudes towards vaccines—and classifies individuals as low, medium or high hesitancy. The hesitancy categories were validated against immunization behaviours. Children of medium-hesitancy parents were under-immunized for 14% more days than children with low-hesitancy parents, and children of high-hesitancy parents were under-immunized for 51% more days than those with low-hesitancy parents16. Another classification system, termed the 4C model and developed by Betsch, Böhm and Chapman24, classifies individuals based on the main reason for non-vaccination. The four categorizations include complacency (an individual believes vaccination is unnecessary due to a low perceived risk of disease), convenience (structural barriers prevent follow-through on individual intent to vaccinate), confidence (individuals lack trust in the effectiveness and safety of vaccines and vaccine systems) and calculation (individuals lack strong attitudes about vaccination, weighing the perceived risk of vaccination and disease against each other). Because of the variation in drivers of vaccine hesitancy, well-informed communication in both healthcare practice settings and public campaigns is key to vaccine attitude shift.

Overall, there are few evidence-based strategies to address vaccine hesitancy and even fewer based on quantitative evidence25. Recent interventions have focused on vaccine knowledge and education26,27 and message framing3,28 as methods to change vaccination attitudes. Most demonstrate short-term success, but some backfire and worsen parental hesitancy about vaccines3 or temporarily increase vaccination rates but may lead to long-term dissatisfaction and decreased intention to vaccinate28.

The apparent resistance of vaccine attitudes to interventions focused on education suggests that attitudes towards vaccines may instead be rooted in deeper intuitions that guide individual decisions. Moral Foundations Theory was developed to identify the key dimensions of moral judgement and decision-making29,30. This theory proposes that a set of innate intuitions lead humans to certain emotional responses (for example, approval or disapproval) to particular interpersonal events. Early theoretical work on this topic proposed that people often quickly make moral decisions based on emotional, intuitive processes such as these and come up with post hoc reasons and justifications31. Subjects in past studies were unable to provide justifications for their decisions regarding moral dilemmas that elicited strong intuitive reactions, but stripped away rational bases for decisions guided by such reactions32,33. Attitudes towards vaccines may be similarly based in such intuitive processes.

Six foundations have been proposed: care/harm, authority/subversion, loyalty/betrayal, liberty/oppression, purity/degradation and fairness/cheating34,35. Reliance on these foundations is typically associated with approval of actions that uphold virtues and practices related to these foundations, vigilance against violations of the virtues and practices, and disapproval of such violations36. The care/harm foundation promotes virtues related to caring for others and protecting them from harm. It also involves compassion for those who are vulnerable or suffering and anger towards those who harm others34. The authority/subversion foundation gives rise to concerns about respect for and obedience to authority, while the loyalty/betrayal foundation is associated with virtues of in-group loyalty, patriotism and sacrificing oneself for the group36. The liberty/oppression foundation involves valuing individual freedom and liberties while resenting dominance by others35. Concerns with sanctity, chastity and avoiding contamination (spiritual and metaphorical) of oneself and one’s group stem from the purity/degradation foundation. This foundation is associated with disapproval of acts that are deemed ‘disgusting’ or ‘unnatural’.

The extent to which each of the six moral foundations is used in moral judgments and decision-making varies by individual and across cultures. Early work has shown that differential endorsement of moral foundations can explain differences in political orientation where past attempts to explain political ideology with rational self-interest models had failed36. In this study, liberals rated the care versus harm and fairness versus cheating foundations as more relevant to their moral judgements, whereas conservatives rated the authority versus subversion, loyalty versus betrayal and purity versus degradation foundations as more relevant. Moral foundations have since been associated with attitudes and attitudinal shift on topics as varied as climate change, suicide and philanthropy4,5,6,7,37,38. Differences in attitudes towards vaccination may similarly stem from differences in reliance on moral foundations.

Past research has demonstrated that framing persuasive messages using the moral foundations that underlie people’s attitudes and judgements can successfully change attitudes. For instance, one study found that conservatives were more persuaded to recognize the legitimacy of climate change and engage in conservation behaviours by messages framed using the moral foundations more strongly endorsed by conservatives (that is, purity, authority and loyalty: “Demonstrate your respect by following the examples of your religious and political leaders who defend America’s natural environment. SHOW YOUR PATRIOTISM”) than messages framed using those moral foundations more strongly endorsed by liberals (that is, care and fairness: “Do the right thing by preventing the suffering of all life-forms and making sure that no one is denied their right to a healthy planet. SHOW YOUR COMPASSION”)39. If vaccine hesitancy is indeed rooted in particular moral foundations, identifying which moral foundations are associated with vaccine hesitancy would provide a promising direction for the development of a values-based messaging intervention. More favourable attitudes towards vaccines may increase intention to vaccinate, ultimately increasing rates of immunization.

Research on the attitudes of vaccine-hesitant and vaccine-refusing parents40 and on frequent claims of anti-vaccination websites41,42 provides reason to suspect that some of the six moral foundations might relate to vaccine hesitancy. An opinion commonly expressed by these sources is that vaccines are unnatural and exposing children to diseases ‘naturally’ to build up immunity is preferable. Anti-vaccination websites also often claim that vaccines contain ‘contaminants’42. These concerns may be rooted in the purity moral foundation, with its emphasis on avoiding anything disgusting or unnatural. Another frequent message on anti-vaccine websites is that mandatory vaccination policies violate parental civil liberties42. The liberty foundation may be associated with such concerns about vaccine policy. Distrust of scientists and government officials who promote vaccinations is also common among people with unfavourable vaccine attitudes and anti-vaccination websites40,41,42. This distrust may be associated with low reliance on the authority foundation and its associated respect for and obedience to authority, or possibly high reliance on the liberty foundation and its resentment of perceived oppressors.

Other relationships between moral foundations and vaccine hesitancy are plausible as well. The care/harm foundation may underlie vaccine hesitancy to the extent that people are concerned about the harm that might result from vaccines, particularly if it affects vulnerable children. This foundation may also stimulate parents’ protective instincts regarding vaccination and disease. The fairness/cheating foundation may fuel outrage in response to the perception that pharmaceutical companies motivated by profit have an unfair voice in vaccine policy42. However, these two foundations may also be linked with concerns among individuals with pro-vaccine attitudes that unvaccinated children put others (particularly those who cannot be vaccinated for medical reasons) at risk. The loyalty/betrayal foundation may be the least likely to be related to attitudes towards vaccines. However, it is important to assess all six moral foundations to determine which are associated with attitudes towards vaccines, as past research shows that it is often not readily apparent which moral domains are relevant to one’s own judgements on a given topic38. Moreover, since the moral foundation scales used in this study have been shown to correlate with each other36, assessing the strength of such associations when controlling for other moral foundations is also key.

Here, we present the results of direct investigations into the association of the values-based foundations of moral foundations theory with parental vaccine hesitancy. The first study included 1,007 consenting parents who met eligibility criteria and answered an online survey about moral foundations and vaccine attitudes. Individuals were eligible for the study if they resided in the United States, were between 18 and 50 years of age (inclusive) and had at least one child under the age of 13. Participants were 66.4% female. Most were between 18 and 40 years of age (76.6%) and had either one (40.1%) or two (36.1%) children (see Supplementary Table 1 for full demographics and Supplementary Fig. 1 for a participant flow diagram). Of the respondents, 73.0% were classified as low hesitancy according to their Parent Attitudes about Childhood Vaccines score, while approximately 11% were classified as medium hesitancy and 16% were classified as high hesitancy (see Supplementary Table 2 for demographics stratified by hesitancy).

We first developed an unconditional polytomous logistic regression model treating the degree of vaccine hesitancy (low, medium or high) as the outcome of interest. The starting model contained predictor terms for all six moral foundations, age, gender, education and number of children. We also dichotomized the moral foundation scores into high and low based on the mean study population score and age into 18–40 years and 41+ years. The rationale behind model choice and all categorizations can be found in the Methods. After stepwise backwards predictor elimination and confounding assessment for demographic terms, our final no-interaction model (model 1A) contained variables for all six moral foundations, education and age. Odds ratios (ORs) and 95% confidence intervals (95% CIs) calculated from this model can be seen in Fig. 1 (see Supplementary Table 3 for numerical values for all ORs, 95% CIs and P values for model 1A).

Fig. 1: Model 1A.
Fig. 1

ORs and 95% CIs from an unconditional polytomous logistic regression model, adjusted for education and age. a, Likelihood of high emphasis on a moral foundation relative to likelihood of low emphasis for a parent classified as medium hesitancy, as opposed to low hesitancy. b, Likelihood of high emphasis on a moral foundation relative to likelihood of low emphasis for a parent classified as high hesitancy, as opposed to low hesitancy.

Medium-hesitancy parents were twice (adjusted OR (aOR): 2.08, 95% CI: 1.27–3.40) as likely as low-hesitancy parents to place high emphasis on purity. High-hesitancy parents were also twice as likely as low-hesitancy parents to place high emphasis on purity (aOR: 2.15, 95% CI: 1.39–3.31) and liberty (aOR: 2.19, 95% CI: 1.50–3.21). Inversely, high-hesitancy parents were half as likely (aOR: 0.43, 95% CI: 0.27–0.67) as low-hesitancy parents to place high emphasis on respect for authority. The strengths of emphasis on harm and fairness were not statistically significantly (P < 0.05) associated with the degree of vaccine hesitancy. To ensure categorization did not impact our statistical results, we conducted a multiple linear regression using the continuous versions of the outcome and predictors (except for education, which was initially categorical) included in model 1A. The statistical results from this linear regression support those from model 1A (see Supplementary Table 4 for all regression coefficients (β), standard errors (s.e.) and P values). Although we did not include political ideology in our starting model to avoid politicizing an as-yet non-partisan issue, we conducted a sensitivity analysis to see if its inclusion in model 1A greatly impacted our estimated ORs. We also conducted a similar analysis with gender, even though confounding assessment had indicated it did not need to be included. The ORs remained stable in both the political ideology and gender analyses (see Supplementary Tables 5 and 6, respectively, for all ORs, 95% CIs and P values).

We repeated our modelling methodology, this time also including all two-way interaction terms in our starting model to explore possible differences in effect estimates across covariate categories. Our final interaction-containing model (model 1B) contained variables for all six moral foundations, age, education and a liberty–age interaction term. The results from model 1B, shown in Fig. 2, reveal that the age group to which an individual belonged impacted the likelihood of high liberty foundation endorsement (see Supplementary Table 7 for numerical values for all ORs, 95% CIs and P values for this model).

Fig. 2: Model 1B.
Fig. 2

ORs and 95% CIs from an unconditional polytomous logistic regression model featuring an age–liberty interaction term, adjusted for education and age. a, Likelihood of high emphasis on a moral foundation relative to likelihood of low emphasis for a parent classified as medium hesitancy, as opposed to low hesitancy. b, Likelihood of high emphasis on a moral foundation relative to likelihood of low emphasis for a parent classified as high hesitancy, as opposed to low hesitancy.

Medium-hesitancy parents over the age of 40 were one-third (aOR: 0.33, 95% CI: 0.13–0.81) as likely as their low-hesitancy counterparts to place high emphasis on liberty; however, there was no statistically significant association for liberty in the younger parents. High-hesitancy parents less than 40 years of age were two and a half (aOR: 2.48, 95% CI: 1.59–3.87) times as likely as their low-hesitancy counterparts to place high emphasis on liberty, although there was no statistically significant association for parents greater than 40 years of age. We then conducted a multiple linear regression using the continuous versions of the predictors and outcome for model 1B (see Supplementary Table 8 for all β coefficients, standard errors and P values), as described previously for model 1A. The statistical results support most of those for model 1B, although the liberty–age interaction is no longer statistically significant. We repeated the political ideology and gender sensitivity analyses, as described for model 1A, with model 1B. All but one of the estimated ORs remained stable in both analyses (see Supplementary Tables 9 and 10 for all ORs, 95% CIs and P values for the ideology and gender analyses, respectively). The liberty estimate for parents > 40 years of age in the ideology analysis decreased (aOR: 1.30, 95% CI: 0.75–3.07) but remained statistically insignificant.

A second study, conducted independently of the first, validated the main findings of study 1. This study examined the relationships between moral foundations, attitudes towards vaccines and beliefs about vaccines taken from previous research and content analyses of anti-vaccine websites. An additional goal was to determine whether vaccine-related purity- and liberty-themed violations uniquely mediated the influence of some foundations on vaccine attitudes more than others (for instance, if concerns about toxins in vaccines mediated potential relationships between the purity or harm foundations and overall attitudes).

Participants in study 2 were recruited from Amazon Mechanical Turk (MTurk) based on their responses to a previous demographic survey used to identify parents. Individuals were eligible for the study if they were at least 18 years of age, resided in the United States and had at least one child. The final sample for analysis included 464 consented parents (see Supplementary Fig. 2 for a participant flow diagram). Respondents were 60.3% female and had a mean age of 42.3 years (ages ranged from 20 to 80). In total, 85% of participants had at least some college education and 73% had one or two children. This validation study did not stratify parental attitudes by level of hesitancy.

First, we developed a multiple regression model (model 2A) treating vaccine attitudes as the outcome of interest and included continuous variables for all six moral foundations, education and age. Both the purity (β: 0.06, s.e.: 0.01) and liberty (β: 0.03, s.e.: 0.01) foundations statistically significantly (P < 0.05) predicted vaccine attitudes (see Supplementary Table 11 for all β coefficients, s.e. and P values for this model). Next, we tested for mediation of these relationships using the Hayes approach.

The first model testing mediation (model 2B) included the purity foundation as the predictor variable, vaccine-related purity- and liberty-violation beliefs as mediator variables, and vaccine attitudes as the outcome variable. Including both purity- and liberty-violation beliefs as mediators allowed us to determine whether only purity-violation beliefs might explain the association between the purity foundation and vaccine hesitancy or both purity- and liberty-violation beliefs mediated this association. The other five moral foundations, age and education were included as covariates. Model 2B, which is presented in Table 1, yielded evidence of an indirect (mediation) effect of purity on vaccine attitudes through both purity- (β: 0.03, s.e.: 0.007) and liberty-violation beliefs (β: 0.03, s.e.: 0.007).

Table 1: Unstandardized β coefficients, s.e., P values and 95% CIs from a Hayes mediation model featuring purity (predictor), vaccine-related purity- and liberty-violation beliefs (mediators), and vaccine attitudes (outcome). β coefficients are adjusted for the other five moral foundations, age, and education.

The second model testing mediation (model 2C) included the liberty foundation as the predictor variable, vaccine-related purity- and liberty-violation beliefs as mediator variables, and vaccine attitudes as the outcome variable. Including both purity- and liberty-violation beliefs as mediators allowed us to determine whether only liberty-violation beliefs explained the association between the liberty foundation and vaccine attitudes, or both purity- and liberty-violation beliefs mediated this association. The other five moral foundations, age and education were included as covariates. Model 2C, which is presented in Table 2, provided evidence of an indirect (mediation) effect of liberty on vaccine attitudes through both purity- (β: 0.01, s.e.: 0.004) and liberty-violation beliefs (β: 0.01, s.e.: 0.006).

Table 2: Unstandardized β coefficients, s.e., P values and 95% CIs from a Hayes mediation model featuring liberty (predictor), vaccine-related purity- and liberty-violation beliefs (mediators), and vaccine attitudes (outcome). β coefficients are adjusted for the other five moral foundations, age, and education.

We demonstrated in our first study that several distinct values are associated with parental vaccine hesitancy. However, more foundations are significantly associated with high hesitancy than with medium hesitancy. For medium-hesitancy parents, the only association is high endorsement of purity, while high hesitancy is associated with high endorsement of both purity and liberty and low endorsement of authority. The influence of different generational values may also prove important, as younger high-hesitancy parents are more likely to highly endorse liberty. Although this age–liberty interaction is not statistically significant in the linearly regressed version of model 1B, we attribute this to the sparse data in each age–liberty score category (only 1,007 participants across 1,023 possible age–liberty score combinations). The strength of the associations characterized in study 1 is not impacted by either gender or political ideology. This finding is particularly useful because, although moral foundations theory originally focused on explaining tribal political attitudes, our research supports that the association of moral foundations with vaccine attitudes transcends political ideology. The phenomenon of political ideology not modifying the relationship between moral foundations and attitudes is not unique to vaccines. For example, a previous study demonstrated that moral foundations were predictive of victim-blaming attitudes, even when controlling for ideology and right-wing authoritarianism43.

Previous studies have focused on overt concerns cited by vaccine-hesitant parents44. We demonstrate in our second study that these concerns are directly linked to purity and liberty moral foundations. Past research has also found a connection between disgust and contamination concerns and negative vaccine attitudes45,46. The results from study 2 further elaborate on this work, as we found that high emphasis on the purity and liberty foundations is associated with vaccine beliefs directly tied to purity foundation violations (for example, vaccines contain poisons and toxins, while diseases like measles are natural). High emphasis on the liberty foundation is associated with vaccine beliefs directly tied to liberty foundation violations (for example, vaccine mandates violate civil liberties and are excessive government control). These results demonstrate that vaccine purity- and liberty-violation beliefs mediate the relationship between purity and liberty foundations and vaccine attitudes. These results also suggest that moral foundations may shape the beliefs one has about an attitude object in ways beyond those directly related to the moral foundation. For instance, the purity foundation may influence overall attitudes towards vaccines, which in turn promote beliefs that are in line with these attitudes—even beliefs that have seemingly little to do with purity. However, we cannot determine causality with these studies alone.

We recognize there are a few limitations to this work, as the data were collected from an online convenience sample of parents and these results may be of limited generalizability. However, our claims are bolstered by the fact that purity and liberty were statistically significant predictors of vaccine attitudes in the second study, which was independently conducted with no contact between the researchers of the first and second studies until analysis of both sets of data had been completed. Other high-quality vaccine-focused studies have also been conducted using MTurk47, further supporting our decision. We are also unable to assess causality between moral foundations and vaccine hesitancy due to the cross-sectional nature of our studies, but our characterization of patterns in associations suggests that a moral foundations approach should be further pursued to evaluate vaccine-focused messages that avoid a backfire effect.

Future interventions in vaccine attitudes may be similarly framed using the moral foundations associated with medium or high hesitancy, as done previously with the moral foundations associated with political ideology39. Messages could present vaccine effects and importance couched in purity and liberty foundation-stimulating terms. Terminology (the Moral Foundations Dictionary)36, vignettes48 and visual images (the Socio-Moral Image Database)49 have been tested for their stimulation of different foundations, although these resources are not specifically focused on vaccines. These resources could be used to develop interventions that describe the pros of vaccinating in foundation-framed messages or refute classic vaccine hesitations and negative attitudes using arguments framed in terms of the foundations the hesitation is based in. For example, an intervention framed in terms of the purity foundation might read, ‘Boost your child’s natural defences against diseases! Keep your child pure of infections – Vaccinate!’ This message might show a picture of a child with measles. An intervention framed in terms of the liberty foundation might read, ‘Take personal control of your child’s health! Vaccinations can help your child and others be free to live a happy and healthy life’, along with pictures of children playing. Messages could be displayed at public schools, libraries, doctors’ offices and even on the Internet, where parents frequently obtain information about vaccines and would likely encounter them42. However, it is important to note that the above messages are simple examples and have not been formally tested. Immunization programmes could also integrate such an approach into communications training for their personnel and educational modules for clinicians, as well as the development of foundation-framed talking points for quick reference.

Overall, this work quantifies the associations of moral foundations with vaccine attitudes. These results indicate a promising direction for the development of messaging interventions formed by quantitative evidence and less reliant on cues from common sense. Our data also suggest that health decisions are, to some extent, linked with moral concerns. However, the extent to which moral foundations impact health decisions would need to be replicated in studies focused on other health behaviours. More importantly, this morals-based approach can easily be applied to other health decisions and may provide a standardized, yet easily adaptable approach for public health interventions.

Methods

Study 1

Study population recruitment

After approval of the study protocol by the Emory University Institutional Review Board, parents of children under 13 years of age were recruited online through MTurk. Due to the explorative nature of the work, we used the general guideline of between 10 and 30 observations (at minimum) per intended predictor and comparison50,51,52 to determine that we needed a minimum sample size of 600. A link to the survey was posted on the MTurk website, along with a brief description of the study and the associated monetary compensation if individuals were eligible and completed the survey. However, we did not include any eligibility criteria in the MTurk posting to prevent individuals from fraudulently gaining access to the survey. Interested respondents answered a brief series of screening questions to ensure that they met eligibility criteria (between 18 and 50 years of age, resident of the United States and at least one child under 13 years of age). Screening questions were phrased to prevent participants from guessing which demographic groups we were recruiting. We explicitly stated that participants would not be paid for answering screening questions and we also prevented the same IP address from accessing the survey more than once. These measures were implemented to prevent ineligible individuals from answering the screening questions multiple times to fraudulently participate. Eligible individuals then saw a consent form displayed and were required to answer a question asking if they consented to taking part in the study. Only individuals who selected ‘Yes’ when asked if they were willing to take part in the study were allowed to answer the remainder of the survey questions.

Survey instrument

The survey was developed based on the Parent Attitudes about Childhood Vaccines short scale, Moral Foundations Questionnaire and Liberty Foundation Questionnaire. The five-item Parent Attitudes about Childhood Vaccines short scale was the vaccine hesitancy outcome analysed in this study. The Parent Attitudes about Childhood Vaccines short scale contained five items, each answered on a three-point scale (‘Yes’, ‘No’ or ‘Don’t know’, or ‘Hesitant’, ‘Not hesitant’ or ‘Not sure’, depending on the item) within three categories: vaccine behaviours, attitudes about vaccine safety and efficacy, and general health attitudes16,53.

The 30-item Moral Foundations Questionnaire contained items on a 6-point Likert scale designed to capture each moral foundation construct29 (sample items in Supplementary Table 12). Questions asked about the relevance of the statement when determining whether something was right or wrong and the importance of the presented consideration when making decisions. Answers ranged from ‘Not at all relevant’ to ‘Extremely relevant’ for relevance questions, and from ‘Strongly disagree’ to ‘Strongly agree’ for judgement questions. The nine-item Liberty Foundation Questionnaire used similar questions to assess the liberty foundation35. This scale was developed after the original five-foundation Moral Foundations Questionnaire35. Two previously validated attention checks asked the respondents to rate how relevant someone’s ability to do mathematics was to judging whether something was right or wrong (acceptable answers included ‘Not at all’ to ‘Somewhat’) and how much they agreed with the statement ‘It is better to do good than to do bad’ (acceptable answers ranged from ‘Neutral’ to ‘Strongly agree’). These items checked whether participants were paying attention and responding meaningfully, as the selection of ‘unacceptable’ answers reflected careless answering of the survey. This check was performed in keeping with previous work48. The collected demographic information included each respondent’s age, number of children, level of education, gender and political ideology.

Scoring and cleaning data

Survey data were analysed using SAS 9.4 (32 bit, English) software (SAS Institute; https://www.sas.com/en_us/software/sas9.html). Respondents who failed either of the two attention checks in the survey were removed from the analysis. The Parent Attitudes about Childhood Vaccines items were each scored on a scale of zero to two points, with a summary score ranging from zero to ten. For all but one item, ‘Yes’ and ‘Hesitant’ answers received two points, ‘Don’t Know’ answers received one point and ‘No’ and ‘Not hesitant’ answers received zero points. The ‘I trust the information I receive about shots’ item was reverse-scored (‘Yes’ = 0, ‘Don’t know’ = 1 and ‘No’ = 2). Summary scores were then categorized into level of vaccine hesitancy: low hesitancy (0–4), medium hesitancy (5–6) and high hesitancy (7–10) in keeping with previous work done with this scale53, which also demonstrated significant differences in under-vaccination at these cutpoints16.

Moral Foundations and Liberty Foundation Questionnaire items were scored on a 0–5 scale. For relevance questions, ‘Not at all relevant’ answers received zero points, with each sequential answer receiving one point more than the previous, all the way through to five points for an ‘Extremely relevant’ answer. Judgement questions followed a similar pattern, with ‘Strongly disagree’ answers receiving zero points and ‘Strongly agree’ answers receiving five points. Thus, moral foundation scores ranged from 0 to 30 for each foundation except liberty, as these foundations each had six total items. The liberty summary score initially ranged from 0 to 45 due to the three additional items and was linearly scaled down to range from 0 to 30 to allow direct comparison with other constructs. All moral foundation summary scores were dichotomized into low or high endorsement based on the mean sample score for each of the foundations. This was done so that any significant associations could be easily used in a clinical setting. All sociodemographic variables were categorized based on the response options except for age. Age was dichotomized into 18–40 years and 41+ years based on exploratory linear regressions with each foundation score that demonstrated a potential interaction between age and liberty. This categorization further allowed for clinical application of any potential results. Categories for number of children included one, two, three, four and five or more. Categories for level of education included high school level education at most, some college-level education, college degree, some graduate-level education, and graduate degree. Categories for political ideology included liberal, moderate, conservative, libertarian and other/don’t know.

Statistical analysis

Descriptive statistics for the distribution of age, number of children, education, gender, political ideology and high versus low foundation endorsement were calculated (reported in Supplementary Table 1). All characteristics except high versus low foundation endorsement were compared by hesitancy category using Fisher’s exact tests with a Monte Carlo simulation due to the prohibitive computational time without this simulation (see Supplementary Table 2 for results). We also calculated standardized Cronbach’s alphas (reported in Supplementary Table 13) for each of the six moral foundation constructs. A logistic model was then created to investigate associations of the six moral foundations with each Parent Attitudes about Childhood Vaccines hesitancy category. No significant deviations from model assumptions were observed. We used unconditional polytomous logistic regression models due to the exploratory nature of the work, as we did not want to artificially impose a proportional odds assumption required to conduct an ordinal logistic regression on the hesitancy categories. Thus, hesitancy category was treated as a nominal categorical variable. The scores for each moral foundation were treated as dichotomous categorical variables. Age, number of children, gender and education were included as covariates of interest in the first starting model. Confounding was assessed using an all-possible-subsets approach on covariates eligible for removal (that is, those not statistically significant). Covariates whose removal did not result in more than a 10% change in OR estimates were removed to yield the most parsimonious no-interaction model (reported in Fig. 1 and Supplementary Table 3). All originally continuous terms in the final model were returned to their continuous form to verify that categorization did not impact the statistical results of the terms in the model and a multiple linear regression was conducted (reported in Supplementary Table 4). We also conducted sensitivity analyses for both political ideology and gender (reported in Supplementary Tables 5 and 6, respectively). This entailed adding the parameter of interest to the finalized model and assessing whether OR estimates meaningfully changed (defined as a change of more than 10% from the finalized model’s estimate).

In the second starting model, interaction terms were also included to assess any difference in the magnitude of effect at the low and high level of moral foundation endorsement due to these covariates. Backwards elimination was then performed on the interaction terms, with the significance level for elimination set at P < 0.05. Removal of all non-significant interaction terms produced the fully adjusted model used as the gold standard for the assessment of confounding by the covariates of interest. Confounding was assessed using an all-possible-subsets approach on covariates eligible for removal (that is, those not featured in any statistically significant interaction terms or those not statistically significant). Covariates whose removal did not result in more than a 10% change in OR estimates were removed to yield the most parsimonious interaction-containing model (reported in Fig. 2 and Supplementary Table 7). All originally continuous terms in the final model were returned to their continuous form to verify that categorization did not impact the statistical results of the terms in the model and a multiple linear regression was conducted (reported in Supplementary Table 8). Again, we conducted sensitivity analyses for both political ideology and gender (reported in Supplementary Tables 9 and 10, respectively). This entailed adding the parameter of interest to the finalized model and assessing whether the OR estimates meaningfully changed (defined as a change of more than ~10% from the finalized model’s estimate).


Study 2

Study population recruitment

After approval of the study protocol by the Loyola University Chicago Institutional Review Board, parents were recruited through MTurk using a previous demographics survey used to identify parents who were at least 18 years of age and a resident of the United States. We aimed to recruit about 500 parents, over-powering our study to detect a small effect. After viewing the consent form and clicking a button to indicate informed consent, participants completed the 30-item Moral Foundations Questionnaire and the 9-item Liberty Foundation Questionnaire. They were then randomly assigned to view obituaries depicting hypothetical deaths due to either complications from vaccines or vaccine-preventable illnesses. Participants rated their disgust and outrage towards these deaths. The purpose of the obituary manipulation was to increase anti-vaccination attitudes in the vaccine-complication condition and decrease anti-vaccination attitudes in the vaccine-preventable illness condition to ensure sufficient variability in attitudes towards vaccines to detect potential relationships with any of the moral foundations. The manipulation was not designed to manipulate moral foundations, nor was it designed to manipulate message frames targeting different moral foundations. The type of obituary participants viewed had no appreciable effect on any outcomes; therefore, the obituary manipulation is not discussed. Participants then completed measures of vaccine purity-violation beliefs and vaccine liberty-violation beliefs. Finally, participants completed items assessing general vaccine attitudes (see below for more details).

Survey instrument

The survey featured the Moral Foundations Questionnaire, Liberty Foundation Questionnaire and two attention checks as described in study 1. Participants then completed measures of vaccine beliefs and attitudes based on past research. For example, negative attitudes towards vaccines among parents were linked to beliefs that vaccines are unsafe, ineffective or unnecessary to prevent diseases54. Content analyses of anti-vaccine website claims identified similar themes concerning the safety, effectiveness or necessity of vaccines42,55. Many anti-vaccination websites often claim that vaccines contain poisons or are unnatural, with other claims focused more on vaccine policy (for example, mandatory vaccinations in public schools as violations of civil liberties) than vaccines themselves. Borrowing items from a previous survey and analyses of anti-vaccination website claims42,54,55, we created measures of vaccine purity-violation beliefs (for example, ‘Vaccines contain poisons/toxins/contaminants’), vaccine liberty-violation beliefs (for example, ‘Vaccine mandates are excessive government control’) and general vaccine attitudes (for example, ‘Children should get vaccines on a delayed schedule’) to explore how these beliefs mediated the influence of different moral foundations on vaccine attitudes. Demographic questions included age, gender and education.

Scoring and data cleaning

Survey data were analysed using IBM SPSS 21 (http://www-01.ibm.com/support/docview.wss?uid=swg24032236) and the PROCESS macro for SPSS (http://www.processmacro.org/index.html). Respondents who were non-parents and those who failed to answer one or more questions were removed from the analysis. The Moral Foundations Questionnaire was scored as described in study 1. Scoring of the Moral Foundations Questionnaire and Liberty Foundation Questionnaire was done as described for study 1. A programming error resulted in exclusion of the items ‘Chastity is an important and valuable virtue’ from the purity scale and ‘One of the worst things a person could do is hurt a defenceless animal’ from the harm scale. Thus, the fairness, loyalty and authority scales each had six items, as in study 1. Moral foundation scores ranged from 0 to 30, except the liberty foundation scores, which ranged from 0 to 45. Items assessing vaccine purity-violation beliefs (n = 7), vaccine liberty-violation beliefs (n = 7) and general vaccine attitudes (n = 6) were separately averaged to create summary scores for each construct. Values for each construct ranged from 1 to 7.

Statistical analysis

Cronbach’s alphas (reported in Supplementary Table 12) for moral foundation constructs, vaccine purity- and liberty-violation beliefs and general vaccine attitudes were calculated. A multiple regression model (reported in Supplementary Table 11) was created to investigate the associations of the six moral foundations with the three constructs of interest: vaccine purity-violation beliefs, vaccine liberty-violation beliefs and vaccine attitudes. The six moral foundations were regressed on vaccine attitudes, with age and level of education as covariates. Hayes mediation analyses were conducted to test for mediation using the PROCESS macro56. The first mediation analysis (reported in Table 1) included the purity foundation as the predictor variable, vaccine purity-violation beliefs and liberty-violation beliefs as mediator variables, and vaccine attitudes as the outcome variable. The other five moral foundations, age and education were included as covariates. The second mediation analysis (reported in Table 2) included the liberty foundation as the predictor variable, vaccine purity-violation beliefs and liberty-violation beliefs as mediator variables, and vaccine attitudes as the outcome variable. The other five moral foundations, age and level of education were again included as covariates. Before the primary analyses, data were examined for violations of assumptions of the statistical tests conducted (that is, multiple regression). Visual inspection of quantile–quantile plots and histograms suggested that the data were approximately normally distributed, including the regression residuals, and no significant deviations from any other assumptions were observed.


Life Sciences Reporting Summary

Further information on experimental design is available in the Life Sciences Reporting Summary.


Code availability

The code used to analyse the datasets in this study is available from the corresponding author on reasonable request.


Data availability

The datasets generated and analysed in this study are available from the corresponding author on reasonable request.

Additional Information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  1. 1.

    Omer, S. B., Salmon, D. A., Orenstein, W. A., deHart, M. P. & Halsey, N. Vaccine refusal, mandatory immunization, and the risks of vaccine-preventable diseases. N. Engl. J. Med. 360, 1981–1988 (2009).

  2. 2.

    Parker, A. A. et al. Implications of a 2005 measles outbreak in Indiana for sustained elimination of measles in the United States. N. Engl. J. Med. 355, 447–455 (2006).

  3. 3.

    Nyhan, B., Reifler, J., Richey, S. & Freed, G. L. Effective messages in vaccine promotion: a randomized trial. Pediatrics 133, E835–E842 (2014).

  4. 4.

    Day, M. V., Fiske, S. T., Downing, E. L. & Trail, T. E. Shifting liberal and conservative attitudes using moral foundations theory. Pers. Soc. Psychol. Bull. 40, 1559–1573 (2014).

  5. 5.

    Dickinson, J. L., McLeod, P., Bloomfield, R. & Allred, S. Which moral foundations predict willingness to make lifestyle changes to avert climate change in the USA? PLoS ONE 11, e0163852 (2016).

  6. 6.

    Feinberg, M. & Willer, R. The moral roots of environmental attitudes. Psychol. Sci. 24, 56–62 (2012).

  7. 7.

    Feinberg, M. & Willer, R. From gulf to bridge: when do moral arguments facilitate political influence? Pers. Soc. Psychol. Bull. 41, 1665–1681 (2015).

  8. 8.

    Omer, S. B. et al. Nonmedical exemptions to school immunization requirements: secular trends and association of state policies with pertussis incidence. JAMA 296, 1757–1763 (2006).

  9. 9.

    Omer, S. B., Richards, J. L., Ward, M. & Bednarczyk, R. A. Vaccination policies and rates of exemption from immunization, 2005–2011. N. Engl. J. Med. 367, 1170–1171 (2012).

  10. 10.

    Wang, E., Clymer, J., Davis-Hayes, C. & Buttenheim, A. Nonmedical exemptions from school immunization requirements: a systematic review. Am. J. Public Health 104, e62–e84 (2014).

  11. 11.

    Seither, R. et al. Vaccination coverage among children in kindergarten—United States, 2015–2016 school year. MMWR Morb. Mortal. Wkly Rep. 65, 1057–1064 (2016).

  12. 12.

    Lieu, T. A., Ray, G. T., Klein, N. P., Chung, C. & Kulldorff, M. Geographic clusters in underimmunization and vaccine refusal. Pediatrics 135, 280–289 (2015).

  13. 13.

    Seither, R. et al. Vaccination coverage among children in kindergarten—United States, 2013–2014 school year. MMWR Morb. Mortal. Wkly Rep. 63, 913–920 (2014).

  14. 14.

    Birnbaum, M. S., Jacobs, E. T., Ralston-King, J. & Ernst, K. C. Correlates of high vaccination exemption rates among kindergartens. Vaccine 31, 750–756 (2013).

  15. 15.

    Buttenheim, A., Jones, M. & Baras, Y. Exposure of California kindergartners to students with personal belief exemptions from mandated school entry vaccinations. Am. J. Public Health 102, e59–e67 (2012).

  16. 16.

    Opel, D. J. et al. Validity and reliability of a survey to identify vaccine-hesitant parents. Vaccine 29, 6598–6605 (2011).

  17. 17.

    Gust, D. et al. Immunization attitudes and beliefs among parents: beyond a dichotomous perspective. Am. J. Health Behav. 29, 81–92 (2005).

  18. 18.

    Larson, H. J., Jarrett, C., Eckersberger, E., Smith, D. M. D. & Paterson, P. Understanding vaccine hesitancy around vaccines and vaccination from a global perspective: a systematic review of published literature, 2007–2012. Vaccine 32, 2150–2159 (2014).

  19. 19.

    Opel, D. J. et al. Development of a survey to identify vaccine-hesitant parents: the Parent Attitudes about Childhood Vaccines survey. Hum. Vaccin. 7, 419–425 (2011).

  20. 20.

    Benin, A. L., Wisler-Scher, D. J., Colson, E., Shapiro, E. D. & Holmboe, E. S. Qualitative analysis of mothers’ decision-making about vaccines for infants: the importance of trust. Pediatrics 117, 1532–1541 (2006).

  21. 21.

    Kestenbaum, L. A. & Feemster, K. A. Identifying and addressing vaccine hesitancy. Pediatr. Ann. 44, e71–e75 (2015).

  22. 22.

    Dubé, E. et al. Vaccine hesitancy: an overview. Hum. Vaccin. Immunother. 9, 1763–1773 (2013).

  23. 23.

    Keane, M. T. et al. Confidence in vaccination: a parent model. Vaccine 23, 2486–2493 (2005).

  24. 24.

    Betsch, C., Böhm, R. & Chapman, G. B. Using behavioral insights to increase vaccination policy effectiveness. Policy Insights Behav. Brain Sci. 2, 61–73 (2015).

  25. 25.

    Jarrett, C., Wilson, R., O’Leary, M., Eckersberger, E. & Larson, H. J. Strategies for addressing vaccine hesitancy—a systematic review. Vaccine 33, 4180–4190 (2015).

  26. 26.

    Spleen, A. M., Kluhsman, B. C., Clark, A. D., Dignan, M. B. & Lengerich, E. J. An increase in HPV-related knowledge and vaccination intent among parental and non-parental caregivers of adolescent girls, age 9–17 years, in Appalachian Pennsylvania. J. Cancer Educ. 27, 312–319 (2012).

  27. 27.

    Sales, J. M. et al. Rural parents’ vaccination-related attitudes and intention to vaccinate middle and high school children against influenza following educational influenza vaccination intervention. Hum. Vaccin. 7, 1146–1152 (2011).

  28. 28.

    Opel, D. J. et al. The influence of provider communication behaviors on parental vaccine acceptance and visit experience. Am. J. Public Health 105, 1998–2004 (2015).

  29. 29.

    Graham, J. et al. Mapping the moral domain. J. Pers. Soc. Psychol. 101, 366–385 (2011).

  30. 30.

    Haidt, J. & Graham, J. When morality opposes justice: conservatives have moral intuitions that liberals may not recognize. Soc. Justice Res. 20, 98–116 (2007).

  31. 31.

    Haidt, J. & Joseph, C. in The Innate Mind Vol. 3 (eds Carruthers, P., Laurence, S. & Stich, S.) 367–391 (Oxford Univ. Press, New York, NY, 2007).

  32. 32.

    Haidt, J. The emotional dog and its rational tail: a social intuitionist approach to moral judgment. Psychol. Rev. 108, 814–834 (2001).

  33. 33.

    Hauser, M., Cushman, F., Young, L., Kang-Xing Jin, R. & Mikhail, J. A dissociation between moral judgments and justifications. Mind Lang. 22, 1–21 (2007).

  34. 34.

    Graham, J. et al. Moral foundations theory: the pragmatic validity of moral pluralism. Adv. Exp. Soc. Psychol. 47, 55–130 (2013).

  35. 35.

    Iyer, R., Koleva, S., Graham, J., Ditto, P. & Haidt, J. Understanding libertarian morality: the psychological dispositions of self-identified libertarians. PLoS ONE 7, e42366 (2012).

  36. 36.

    Graham, J., Haidt, J. & Nosek, B. A. Liberals and conservatives rely on different sets of moral foundations. J. Pers. Soc. Psychol. 96, 1029–1046 (2009).

  37. 37.

    Nilsson, A., Erlandsson, A. & Västfjäll, D. The congruency between moral foundations and intentions to donate, self-reported donations, and actual donations to charity. J. Res. Pers. 65, 22–29 (2016).

  38. 38.

    Rottman, J., Kelemen, D. & Young, L. Tainting the soul: purity concerns predict moral judgments of suicide. Cognition 130, 217–226 (2014).

  39. 39.

    Wolsko, C., Ariceaga, H. & Seiden, J. Red, white, and blue enough to be green: effects of moral framing on climate change attitudes and conservation behaviors. J. Exp. Soc. Psychol. 65, 7–19 (2016).

  40. 40.

    Dube, E. et al. “Nature does things well, why should we interfere?”: vaccine hesitancy among mothers. Qual. Health Res. 26, 411–425 (2016).

  41. 41.

    Kata, A. Anti-vaccine activists, Web 2.0, and the postmodern paradigm—an overview of tactics and tropes used online by the anti-vaccination movement. Vaccine 30, 3778–3789 (2012).

  42. 42.

    Kata, A. A postmodern Pandora’s box: anti-vaccination misinformation on the Internet. Vaccine 28, 1709–1716 (2010).

  43. 43.

    Niemi, L. & Young, L. When and why we see victims as responsible. Pers. Soc. Psychol. Bull. 42, 1227–1242 (2016).

  44. 44.

    Kennedy, A., LaVail, K., Nowak, G., Basket, M. & Landry, S. Confidence about vaccines in the United States: understanding parents’ perceptions. Health Aff. (Millwood) 30, 1151–1159 (2011).

  45. 45.

    Clay, R. The behavioral immune system and attitudes about vaccines. Soc. Psychol. Pers. Sci. 8, 162–172 (2016).

  46. 46.

    Clifford, S. & Wendell, D. G. How disgust influences health purity attitudes. Polit. Behav. 38, 155–178 (2016).

  47. 47.

    Betsch, C., Böhm, R., Korn, L. & Holtmann, C. On the benefits of explaining herd immunity in vaccine advocacy. Nat. Hum. Behav. 1, 0056 (2017).

  48. 48.

    Clifford, S., Iyengar, V., Cabeza, R. & Sinnott-Armstrong, W. Moral foundations vignettes: a standardized stimulus database of scenarios based on moral foundations theory. Behav. Res. Methods 47, 1178–1198 (2015).

  49. 49.

    Crone, D., Bode, S., Murawski, C. & Laham, S. The Socio-Moral Image Database (SMID): a novel stimulus set for the study of social, moral and affective processes. Preprint at https://psyarxiv.com/sja3m/ (2017).

  50. 50.

    Hosmer, J. D. W., Lemeshow, S. & Sturdivant, R. X. in Applied Logistic Regression 35–47 (John Wiley & Sons, Hoboken, NJ, 2013).

  51. 51.

    Van Voorhis, C. R. W. & Morgan, B. L. Understanding power and rules of thumb for determining sample sizes. Tutor. Quant. Methods Psychol. 3, 43–50 (2007).

  52. 52.

    LeBlanc, M. & Fitzgerald, S. Logistic regression for school psychologists. Sch. Psychol. Q. 15, 344–358 (2000).

  53. 53.

    Opel, D. Identifying, understanding, and talking with vaccine-hesitant parents. In From Package to Protection: How do we Close Global Coverage Gaps to Optimize the Impact of Vaccination Conference Presentation (Fondation Mérieux, 2014); http://www.globe-network.org/sites/default/files/en/network/resource/4.opel-douglas-identifying-understanding-and-talking-to-vaccine-hesitant-parents.pdf.

  54. 54.

    Kennedy, A. M., Brown, C. J. & Gust, D. A. Vaccine beliefs of parents who oppose compulsory vaccination. Public Health Rep. 120, 252–258 (2005).

  55. 55.

    Wolfe, R. M., Sharp, L. K. & Lipsky, M. S. Content and design attributes of antivaccination web sites. JAMA 287, 3245–3248 (2002).

  56. 56.

    Hayes, A. F. Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach (Guilford Press, New York, NY, 2013).

Download references

Acknowledgements

No external funding source support was used for this work. No funders had any role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Author information

Affiliations

  1. Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, GA, USA

    • Avnika B. Amin
    • , Robert A. Bednarczyk
    •  & Saad B. Omer
  2. Hubert Department of Global Health, Rollins School of Public Health, Emory University, Atlanta, GA, USA

    • Robert A. Bednarczyk
    •  & Saad B. Omer
  3. Department of Psychology, College of Arts and Sciences, Loyola University Chicago, Chicago, IL, USA

    • Cara E. Ray
    •  & Jeffrey R. Huntsinger
  4. Department of Psychology, College of Health and Behavioral Studies, James Madison University, Harrisonburg, VA, USA

    • Kala J. Melchiori
  5. Department of Management, Eccles School of Business, University of Utah, Salt Lake City, UT, USA

    • Jesse Graham
  6. Emory Vaccine Center, Atlanta, GA, USA

    • Saad B. Omer
  7. Department of Pediatrics, School of Medicine, Emory University, Atlanta, GA, USA

    • Saad B. Omer

Authors

  1. Search for Avnika B. Amin in:

  2. Search for Robert A. Bednarczyk in:

  3. Search for Cara E. Ray in:

  4. Search for Kala J. Melchiori in:

  5. Search for Jesse Graham in:

  6. Search for Jeffrey R. Huntsinger in:

  7. Search for Saad B. Omer in:

Contributions

S.B.O. and R.A.B. developed and designed study 1. A.B.A. collected and analysed the data for study 1. A.B.A. interpreted the data from study 1 with input from S.B.O. and R.A.B. J.R.H. developed the initial idea for study 2. C.E.R. and K.J.M. designed study 2 and collected and analysed the associated data. C.E.R., K.J.M. and J.R.H. interpreted the data from study 2. A.B.A. drafted the paper and most of the supplementary materials, with pertinent sections from study 2 drafted by C.E.R. and J.R.H. S.B.O., R.A.B., C.E.R., K.J.M., J.G. and J.R.H. provided critical revision of the paper.

Competing interests

The authors declare no competing interests.

Corresponding author

Correspondence to Avnika B. Amin.

Electronic supplementary material

  1. Supplementary Information

    Supplementary Figures 1–2, Supplementary Tables 1–13

  2. Life Sciences Reporting Summary