It’s Not the Flu: Popular perceptions of the impact of COVID-19

Messaging from authorities about COVID-19 has been widely divergent. This research aims to clarify popular perceptions of the threat of COVID-19 and its effects on victims. In four studies with over 4,100 U

Unsurprisingly, given this con icting messaging from authorities, Americans' response to the pandemic was highly variable.While many people de ed, and even protested, requirements for mask wearing, social distancing, and economic shutdowns, many others viewed the pandemic with considerable alarm, even taking precautions before they were required (Bump, 2020;Burnett, 2020;Malone & Bourassa, 2020;Dave, Friedson, Matsuzawa, McNichols, Sabia, 2020).With so much at stake for both health and the economy, it is crucial to understand popular perceptions of the hazards of COVID-19 and the sources of their diversity.In this article, we ask: given mixed messages about the risk and impact of COVID-19, how do people perceive the threat to them, and the severity of its impact on others?

Hypothesis Development
This research involves four studies with more than 4,100 participants within the United States in late April and early May of 2020.First, we examine whether popular perceptions of COVID-19 comport with comparisons with the u and vehicle collisions.The outcome variables we investigate are willingness to help and perceptions of the risk of helping people and communities affected by COVID-19 (versus non-COVID-19 threats), as well as perceptions of victims as contaminated, injured, and responsible for their condition.
We hypothesize that, on average, people perceive victims of COVID-19 very differently from non-COVID-19 victims ( u, car accident, HIV/AIDS, and severe storm) -namely, people view COVID-19 victims as riskier to help, more responsible for their condition, more contaminated, and, people will be less willing to help them.
To the extent that people vary in how they perceive the threat of COVID-19 and its impact on victims, what explains this variability?We expect that it is explained by individual differences in people's moral values, as well as demographic characteristics including political orientation, gender, education, and income level.
According to Moral Foundations Theory (MFT; e.g., Haidt, 2007;Graham, Haidt & Nosek, 2009;Graham et al., 2011), conservative people tend to endorse the group-oriented "binding values" of (1) loyalty, (2) respect for authority, and (3) purity more highly than liberal people, who tend to favor the "individualizing values" that emphasize (4) fairness and (5) care.Unlike individualizing values which stipulate unbiased extension of moral concern, binding values foster group boundaries and "us vs. them" dynamics through (a) reciprocal bonds of loyalty, (b) deference to the authorities in the hierarchies that structure groups, and (c) commitment to preserving purity by rejecting people and behaviors that "contaminate" the integrity of the group.These features suggest that binding values might drive a heightened perception of the threat of COVID-19, despite the relationship of binding values with conservatism, and conservative rhetoric expressing skepticism about COVID-19 dangers.
Alternatively, because binding values promote preserving the integrity of groups, people higher in binding values may be expected to show increased willingness to help people and communities affected by COVID-19.However, we favor the hypothesis that people higher in binding values may actually be less willing to help, given the rationale for expecting people high in binding values to tend to view those affected by COVID-19 as more contaminated and riskier to help.Furthermore, less willingness to help COVID-19 victims may be illustrative of a general tendency for people higher in binding values, regardless of politics, to judge victims with less sensitivity.Prior research shows that people higher in binding values are more likely to stigmatize victims as tainted and contaminated, judge victims as more responsible and blameworthy for their own victimization, and are less likely to defend victims of sexual harassment by confronting and reporting harassment (Goodwin, Graham & Diekmann, 2020; Niemi & Young, 2016; Niemi, Hartshorne, Gerstenberg, Stanley & Young, 2020).By contrast, people higher in individualizing values, which are associated with increased sensitivity to suffering and do not emphasize contamination risks, may be more willing to help those affected by the coronavirus.
In addition to measuring attitudes and individual differences in values, we consider the contribution of political orientation, gender, education, and income level.Finally, in order to illuminate the degree to which the salience of different moral values may be altered, and thereby in uence attitudes, we examine whether increasing the salience of either binding or individualizing values through priming affects the outcome variables.Given prior research showing e cacy in priming moral values, in these studies we anticipated small effects for priming binding and individualizing values (Mooijman et al., 2018;Goenka & Thomas, 2020).This research builds on recent applications of the social, behavioral, and psychological sciences to understand and navigate the myriad challenges posed by COVID-19 (e.g., Kni n et al., 2020; van Bavel et al., 2020).These studies have theoretical utility in illuminating the role of moral values in social perception and decision-making, and also applied utility for policymakers, researchers, and citizens responding to COVID-19, and other public health emergencies.

Methods
The rst set of studies was administered between April 24 -27 (Study 1a & Study 2) in the United States.At this time, most Americans were under shut-down orders and complying with them; however, it was unclear the extent to which people's behavior re ected their attitudes.There was increasing media focus on people's politically sourced disapproval of government COVID-19 policy, centered around coverage of protests against the shutdowns which began on April 15.The second set of studies was administered on May 8-13 (Studies 1b-1c) in the United States.By this point, shutdown orders largely remained, but some communities were already beginning to pursue re-opening.

Procedure (All Studies)
A university Institutional Review Board approved all of the studies here.All studies were implemented using Qualtrics survey software and distributed to participants online via Proli c.Studies were preregistered through AsPredicted.org(see Supplementary Materials) as part of a series focused on understanding how MFT is related to individual attitudes concerning COVID-19.De-identi ed data and study materials are archived in the corresponding authors online data repository on Github (see Supplementary Materials for link).Independent participant pools were recruited for each study.The procedure for each study involved the following materials, described in detail in the next section: (1) a moral values prime,(2) a vignette and series of questions measuring attitudes about individual victims and affected communities, and (3) individual difference measures of moral values, political orientation, and demographics.Attention checks were identical across the studies and participants were excluded if they failed either of the two embedded attention checks, or an attention check at the end of the studies.

Materials and Measures (All Studies)
Participants in all studies rst encountered the binding, individualizing, or control prime.The primes were used effectively in prior research (Mooijman et al., 2018;Goenka & Thomas, 2020).In Studies 1a, 1b, and 2, the primes described a warrior who exempli es loyalty, respectfulness, and concern about purity (binding values), or caring and fairness (individualizing values), or who has good character (control).In Study 1c, we replaced the warrior primes with another set of primes found to be effective in prior research, in which participants read about a scholar's ideas about moral values and wrote a response (Mooijman et al., 2018).
Participants in all studies then read a vignette about a person or community, and subsequently, participants' attitudes, the key outcome variables, were measured with a series of questions.In Studies 1a, 1b, and 1c, participants read about "Dan," who was affected by either COVID-19, the seasonal u, or a car accident: "In March 2020, Dan drove across the country for work, and stopped at many cities and towns in several states.Along the way, he [contracted the coronavirus and became very sick; contracted the seasonal u and became very sick; got into a serious car accident and sustained numerous injuries]."After the vignette, using Likert-scales (1-7), participants rated: "How responsible is Dan for the car accident?"(on one page); "How willing would you be to assist Dan?" and "How risky would it be for you to assist Dan?" (on the next page); and "How injured is Dan?" and "How contaminated is Dan?" (on the next page).In Study 1a, we also examined but do not discuss here whether judgments would be affected by how contagiousness was conveyed (see Supplementary Materials for text of vignettes).
In Study 2, rather than provide assessments of an individual u or car accident victim, participants read about an unnamed community: "Since March of 2020, residents of a large community have been hit hard by [an outbreak of the coronavirus; HIV/AIDS; a severe storm], and the city's infrastructure has been overwhelmed by victims needing care.O cials have called for donations, and volunteers to assist the relief effort in soup kitchens, homeless shelters, and medical facilities."After the vignette, using Likertscales (1-7), participants rated (each on a separate page): "How likely would you be to volunteer at [soup kitchen, homeless shelter, medical facilities]?""How likely would you be to donate to [soup kitchen, homeless shelter, medical facilities]?""How risky to your health do you think volunteering would be?" Next, in all studies we measured participants' endorsement of the ve moral values of Moral Foundations Theory -caring, fairness, loyalty, obedience to authority, and purity values --with the Moral Foundations Questionnaire (MFQ-30, Graham et al., 2011).We averaged caring and fairness values for the individualizing values scores, and loyalty, authority, and purity values for the binding values score.At the end of the Study 1b, we also asked participants to identify a warrior quality from the vignette as an attention check.Finally, participants took a brief demographics survey, and we measured political orientation with the item (Iyer, Koleva, Graham, Ditto, Haidt, 2012): "When it comes to politics, do you usually think of yourself as liberal, moderate, conservative, or something else?" -a drop-down menu contained the choices: (1) Very liberal, (2) Liberal, (3) Slightly liberal, (4) Moderate/middle-of-the-road, (5) Slightly conservative, (6) Conservative, (7) Very conservative, (8) ''Don't know/not political'', (9) ''Libertarian,'' (10) Other.The item included the note: "The terms used in your country may differ."Liberal" is intended to include the Left, progressives, and in some countries, socialists."Conservative" is intended to include the Right, traditionalists, and in some countries Christian Democrats."We used selections 1-7 as a scale variable representing the extent of participants' self-identi cation as politically liberal or conservative.

Study 1a
Participants.Study 1a included 1,627 participants (836 female, 765 male, 26 other) with 72 exclusions based on failure of attention checks.The sample size was calculated to yield at least 50 participants per condition, plus ten additional participants in each condition to account for typical rates of exclusion.The average age of the participants was 36.4 (SD=13.0)years old; 90% of participants were not Hispanic or Latino, 10% were Hispanic or Latino, 74% White or European-American, 7% Black or African-American, 12% Asian or Asian-American, less than 1% Native American or Paci c Islander, 4% Multiracial, and 1.5% selected other.Combined annual income was: 22% less than $30,000; 20% between 30,000-49,999; 18% between 50,000-69,999, 19% between 70,000-99,999, and 19% 100,000+.The majority, 43%, of participants were liberal or very liberal; a similar percentage, 39%, was slightly liberal, middle-of-the-road, or slightly conservative; 12% were conservative or very conservative; 3% did not know or were not political; 1% selected libertarian; 1% selected other.Participants were from all four regions of the US: West (24%), Midwest (20%), Northeast (22%) and South (34%).Study 1a Results.We conducted analyses of variance to investigate the in uence of the "damage type" -COVID-19, seasonal u, or car accident -on whether participants considered the protagonist responsible, contaminated, and injured, their willingness to help him, and how risky they considered helping him; as well as whether priming with binding values increased perception of Dan as responsible, contaminated, and risky to help, and decreased willingness to help.
For contamination, injury, riskiness, and willingness to help, we found signi cant main effects of damage type only (see means in Figure 1).Contamination ratings were highest in the case of COVID-19, higher than the u (p<.000), and higher than the car accident (p<.000;F(2,4.003)= 448.00,p < .000,partial eta2= .996).Dan was rated most injured by the car accident, followed by COVID-19 (p<.000), and the u (p<.000;F(2,4.04)= 2487.68,p < .000,partial eta 2 =.999).People considered helping Dan to be riskiest when he contracted COVID-19, compared to the u, or a car accident (p<.000;F(2,4.040)= 2693.42,p < .000,partial eta 2 = .999).People were signi cantly less willing to help the COVID-19 victim compared to the u victim and the car accident victim (p<.000;F(2,4.004)= 68.96,p = .001,partial eta 2 = .972).Study 1a Summary.Result suggest that the perceived threat to health and safety from COVID-19 was elevated compared to the u or a car accident: participants were less willing to help COVID-19 victims, and considered them riskier to help, more responsible, and more contaminated than u and car accident victims.

Study 1b
Participants.Study 1b included 1009 participants (510 female, 494 male, 5 other) with 73 exclusions based on failure of attention checks.The majority of participants in each prime condition identi ed the correct word from the vignette they read (individualizing 81%; binding 66%, control 71%).The sample size was calculated to yield at least 50 participants per condition, plus ten additional participants to each condition to account typical rates of exclusion.The average age of the participants was 38.9 (SD = 13.7);93% of participants were not Hispanic or Latino, 7% were Hispanic or Latino, 79% White or European-American, 7% Black or African-American, 10% Asian or Asian-American, less than 1% Native American or Paci c Islander, 3% Multiracial and 1% selected other.Combined annual income was: 18% less than $30,000; 20% between 30,000 -49,999; 18% between 50,000 -69,999, 22% between 70,000 -99,999, and 23% 100,000+.The majority of participants, 42%, were liberal or very liberal; a similar percentage, 41%, was slightly liberal, middle-of-the-road, or slightly conservative; 13% were conservative or very conservative; 2% did not know or were not political; ~2% selected libertarian; <1% selected other.Participants were from all four regions of the US: West (22%), Midwest (23%), Northeast (19%) and South (36%).

Study 1b
Results.We conducted identical analyses as in Study 1a.Again (see Figure 1), there were signi cant main effects of damage type on all outcome variables: responsibility (F(2,1000) = 64.75,p < .001,partial eta 2 = .97),contamination (F(2,1000) = 339.76,p < .001,partial eta 2 = .994),injury (F(2,1000) = 281.46,p < .001,partial eta 2 =.993), willingness to help (F(2,1000) = 126.0,p < .000,partial eta2= .98),and risk (F(2,1000) = 177.12,p < .001,partial eta 2 = .989),there was a signi cant effect of damage type, only.As in Study 1a, these results indicated that ratings of contamination and perceived risk of helping were highest for COVID-19, followed by the u and the car accident.Ratings of injury were highest for the car accident, followed by COVID-19, and the u.Ratings of responsibility were highest in the case of COVID-19, followed by the car accident and the u.Finally, willingness to help was highest for the car accident victim, followed by the u, and COVID-19.
Study 1b Summary.Relative to the u and car accident victim, participants were again less willing to help the COVID-19 victim, who was considered signi cantly riskier to help, more contaminated, and more responsible.

Study 1c
In Study 1c, we replaced the "warrior" primes from Studies 1a-b with another set of primes for binding and individualizing values that have been found to be effective in prior research (Mooijman et al., 2018).These primes involved reading a short passage about morality by a purported "morality scholar," arguing that either the well-being of the group (binding values condition) or the well-being of individuals (individualizing values condition) is central to morality.Participants then wrote a brief response essay discussing their perspective on the scholar's ideas about morality.In the control condition, participants did not read or write a passage.
Study 1c Summary.Once again, people were less willing to help the COVID-19 victim, who was considered signi cantly riskier to help, more contaminated, and more responsible than a u or car accident victim.

Study 2
Binding values re ect group-level moral concerns; as such, Study 2 aimed to shed light on whether binding values would affect judgments of communities (groups) differently than they affected judgments of individuals.We investigated whether perceived risk and willingness to help a community, rather than an individual victim, would vary with moral values and damage type.Study 2 Participants.Study 2 included 571 participants (317 female, 218 male, 4 other) with 23 exclusions based on failure of attention checks.The sample size was calculated to yield at least 50 participants per condition, plus ten additional participants in each condition to account for typical rates of exclusion.The average age of the participants was 35.7 (SD = 12.2); 92% of participants were not Hispanic or Latino, 8% were Hispanic or Latino, 79% White or European-American, 8% Black or African-American, 7% Asian or Asian-American, less than <1% Native American or Paci c Islander, 4% Multiracial and 1.3% selected other.Combined annual income was: 19% less than $30,000; 20% between 30,000-49,999; 20% between 50,000-69,999, 21% between 70,000-99,999, and 19% 100,000+.The majority of participants, 43%, were liberal or very liberal; a similar percentage, 39% was slightly liberal, middle-of-theroad, or slightly conservative (39%); 13% were conservative or very conservative; 2% did not know or were not political; <1% selected libertarian, <1% selected other.Participants were from all four regions of the US: West (18%), Midwest (23%), Northeast (24%) and South (35%).Study 2 Results.We used analyses of variance to investigate whether participants would be less willing to help (donate to or volunteer in) an unnamed community affected by COVID-19, and how they perceived the risk of helping, when binding values were made salient, versus individualizing values or control (no prime).As in the previous studies, we varied damage type such that the community was affected by COVID-19, HIV/AIDS, or a severe storm.We also examined whether effects differed on the targets of donations and volunteering (soup kitchen, homeless shelter, medical facilities).
Consistent with the previous studies, we found a signi cant main effect of damage type for volunteering (F(2,531) = 7.62, p < .001,partial eta 2 = .028):participants were less willing to volunteer in the case of a community affected by COVID-19, compared to HIV/AIDS or a severe storm (p's<.005;see means in Figure 2).There was no effect of the moral values primes.For donation, no effects were signi cant.For we again found a signi cant main effect of damage type (F(2,531) = 85.1, p < .001,partial eta 2 = .24):people considered volunteering riskiest in a community affected by COVID-19, compared to HIV/AIDS or a severe storm (p's<.000,see Figure 2).Finally, merging across volunteering and donating, there was a signi cant effect of helping location (F(2,1062) = 44.00,p < .001,partial eta 2 = .08):people preferred to help a soup kitchen, followed by a homeless shelter, and then medical facilities.
Study 2 Summary.with the previous studies examining an individual victim, participants were less willing to help a community of COVID-19 victims, relative to HIV/AIDS and storm victims, and considered helping the COVID-19 victims to be riskier.
All studies: Stable moral values, politics, and demographics analyses We examined whether people's stable moral values (binding values and individualizing values) predicted attitudes about victims, along with politics, gender, education, and income (politics from 1-7: very liberal to very conservative, gender: male (0) and female (1), income in increments from 1-7: under $30K to $100K and over per year, and education from 1-6: some high school, high school, some university/college, university/college, graduate degree, or professional degree (e.g., M.D., J.D., etc.).We conducted a series of regression analyses on our outcome variables by damage condition (COVID-

General Discussion
The results of these studies, conducted in late April and early-mid May 2020 in the United States, illuminate how people perceive victims of COVID-19 compared to victims of other adversities, as well as the role of moral values and political commitments in these perceptions.On average, participants were signi cantly less willing to help a person or community described as having been affected by COVID-19, compared to another disease (the u or HIV/AIDS) or non-disease damage (car accident or severe storm).
Participants considered a person affected by COVID-19 to be more contaminated and responsible for their adverse circumstances compared to a person with the u or in a car accident; they did not, however, view the COVID- These ndings should help make sense of the myriad downstream problems that have emerged as a result of COVID-19.For example, future research examining how people diagnosed with the disease, or who have been in close contact with someone who was diagnosed, respond to contact tracers; or, how parents make decisions about their children's schooling based on their own assessments of safety and risk, would bene t from awareness of the ndings that people often attribute responsibility to people who contracted COVID-19 and regard them less as suffering people than as contagion vectors.Engaging the prosocial tendencies embodied in people's stable values, rather than their politicized perspectives on the disease, has the potential to reduce stigmatizing characterizations of victims and facilitate a cooperative collective response to the pandemic Note.Beta values represent coe cients with all variables modeled, in a two-step model with binding and individualizing values entered in step one, and demographics (politics, education, gender, and income) entered in step two.The top value in the R 2 change columns represents the binding and individualizing values variables, the bottom value represents the change from the addition of the four variables: politics, education, gender, and income.Variance In ation Factors were computed for each of the models reported in Table 1 and are consistently below 2.00, within acceptable range.For Study 2: Willingness to Help, the rst column represents volunteering, the second column represents donation.Signi cant (p<.05) beta values and R 2 change values indicated in bold.

Figures Figure 1
Figures

Table 1 )
19, u, car accident).We entered moral values (binding values, individualizing values) in step one, and politics, education, gender, and income in step two, to predict (a) responsibility, (b) contamination, (c) injury, (d) perceived risk, and (e) willingness to help.The results for all studies are presented in Table 1, where standardized Beta coe cients and signi cance levels, R 2 change values and signi cance levels, and averaged R 2 change values are indicated; signi cant (p<.05) values indicated in bold.As visible in Table 1, the results of the regression analyses indicate that moral values played a role in people's judgments of both COVID-19 and non-COVID-19 victims; however, moral values consistently played a stronger role in judgments about victims of non-COVID-19 damage.Speci cally, binding or individualizing values signi cantly predicting judgments in cases of non-COVID-19 victims 45% of the time (38/84 tests) and COVID-19 victims 21% of the time (9/42 tests).Averaging R 2 change across all studies to estimate the variability accounted for by the predictors (Johnson & LeBreton, 2004; see , for non-COVID-19 victims, total R 2 change accounted for by binding and individualizing values was:

Table 1 .
Results of regression analyses of judgments of responsibility, contamination, injury, riskiness of helping, and willingness to help for COVID-19 and non-COVID-19 victims in Studies 1a-c and 2.
(Goenka & Thomas, 2020;d than a person in a car accident.Additionally, participants considered the risk of helping a person or a community affected by COVID-19 to be signi cantly greater than the risk involved in helping a person with the u or in a car accident, or a community affected by HIV/AIDS or a severe storm.Political orientation, demographics, and moral values were related participants' responses.Notably, leftleaning politics predicted increased judgments of victims of COVID-19 as responsible and contaminated in two of three studies, whereas for non-COVID-19 victims, responsibility and contamination judgments were reliably predicted by binding values, rather than politics and demographics.Likewise, judgments of victims of COVID-19 as risky to help were predicted by politics (left-leaning), gender (female), and binding values; while judgments of the risk of helping non-COVID-19 victims were reliably and more strongly predicted by, again, binding values, rather than demographics.Finally, individualizing values and increased education, predicted willingness to help victims of COVID-19.Individualizing values played a stronger role than demographics in willingness to help the non-COVID-19 victims.Although we observed relationships between surveyed moral values and attitudes, we did not nd that people's moral values or their responses to victims were reliably altered by our attempts to increase the salience of binding or individualizing values through priming.This may re ect the nature of our outcome variables, which were mainly pertinent to people's health and safety, rather than third-party moral judgments which have been in uenced by values primes in previous work(Goenka & Thomas, 2020;   Mooijiman et al., 2018).The attitudes we measured may be more resistant than other-focused moral judgments to transient exogenous changes in the salience of particular moral values.Consistent with this possibility, the instance in which we observed a priming effect was in participants' third-party judgments of responsibility in Study 1a, where priming binding values increased perceptions of victim responsibility.education,while taking into account individual differences in politics and moral values.Future work should address which interventions, such as public-service messaging or manipulations of the salience of values, are conducive to prosocial and health-related behaviors (e.g., Amin et al., 2017; Batson, 2011; Benish-Weisman, Daniel, Sneddon, & Lee, 2019; Cameron & Payne, 2011; Shariff, Willard, Andersen, & Norenzayan, 2016; Tannenbaum, Hepler, Zimmerman, Saul, Jacobs, Wilson, & Albarracín, 2015; Waytz, Dungan, & Young, 2013).The studies presented in this article help unpack the complex dynamics associated with how people and institutions are responding to COVID-19 and its victims.First, these results make clear that people do not perceive those affected by COVID-19 like they do those with the u; notably, prosocial inclinations toward COVID-19 victims are comparatively diminished.Second, people's differing moral values, politics, and demographic characteristics are associated with their reactions to victims of COVID-19; whereas reactions to non-COVID-19 victims are primarily predicted by moral values.
These associations, particularly the relationships between binding values and judgments of victim responsibility and contamination, and individualizing values and willingness to help, are consistent with previous research.Binding values have been found to predict increased perceptions of victims as blameworthy, responsible, and contaminated; individualizing values have been linked with prosociality (e.g., Iyer et al., 2012; Niemi & Young, 2013, 2016; Niemi et al., 2020; Noser et al., 2015).The nding that demographic factors and politics played a greater role in judgments of COVID-19 victims than moral values indicates that participants' responses may have been based less on stable moral principles related to binding or individualizing values (e.g., help those who are suffering), and more on messages from trusted authorities or political gures.This was not the case for non-COVID-19 victims, where moral values mattered more than demographics and politics.Practical ImplicationsTo the extent that policymakers, charity organizations, and concerned individuals wish to persuade others to assist people affected by COVID-19, these results suggest it will be helpful to address people's safety concerns with