Published on in Vol 10 (2023)

Preprints (earlier versions) of this paper are available at, first published .
Persuasive Messages for Improving Adherence to COVID-19 Prevention Behaviors: Randomized Online Experiment

Persuasive Messages for Improving Adherence to COVID-19 Prevention Behaviors: Randomized Online Experiment

Persuasive Messages for Improving Adherence to COVID-19 Prevention Behaviors: Randomized Online Experiment

Original Paper

1Haskayne School of Business, University of Calgary, Calgary, AB, Canada

2Department of Medicine, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada

3Department of Community Health Sciences, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada

4Department of Political Science, School of Public Policy, University of Calgary, Calgary, AB, Canada

5Department of Economics, University of Notre Dame, Notre Dame, IN, United States

6Department of Economics, Faculty of Arts, University of Calgary, Calgary, AB, Canada

7Department of Pediatrics, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada

Corresponding Author:

Mehdi Mourali, PhD

Haskayne School of Business

University of Calgary

2500 University Drive NW

Calgary, AB, T2N 1N4


Phone: 1 403 220 6684


Background: Adherence to nonpharmaceutical interventions for COVID-19, including physical distancing, masking, staying home while sick, and avoiding crowded indoor spaces, remains critical for limiting the spread of COVID-19.

Objective: The aim of this study was to test the effectiveness of using various persuasive appeals (deontological moral frame, empathy, identifiable victim, goal proximity, and reciprocity) at improving intentions to adhere to prevention behaviors.

Methods: A randomized online experiment using a representative sample of adult Canadian residents with respect to age, ethnicity, and province of residence was performed from March 3 to March 6, 2021. Participants indicated their intentions to follow public health guidelines, saw one of six flyers featuring a persuasive appeal or no appeal, and then rated their intentions a second time. Known correlates of attitudes toward public health measures were also measured.

Results: Intentions to adhere to public health measures increased in all appeal conditions. The message featuring an empathy appeal resulted in a greater increase in intentions than the control (no appeal) message. Moreover, the effectiveness of persuasive appeals was moderated by baseline intentions. Deontological, empathy, identifiable victim, and reciprocity appeals improved intentions more than the control message, but only for people with lower baseline intentions to adhere to nonpharmaceutical interventions.

Conclusions: Public health marketing campaigns aiming to increase adherence to COVID-19 protective behaviors could achieve modest gains by employing a range of persuasive appeals. However, to maximize impact, it is important that these campaigns be targeted to the right individuals.

Trial Registration: NCT05722106;

JMIR Hum Factors 2023;10:e41328




As of July 2022, over 500 million people worldwide have contracted the SARS-CoV-2 virus, resulting in over 6 million COVID-19–related deaths [1]. Despite the remarkable and ongoing effort to inoculate the world population (over 12 billion vaccine doses have been administered so far), the rapidly evolving virus continues to spread at alarmingly high rates. Even affluent countries like Canada—a G7 member with over 83% of the population fully vaccinated—are struggling to contain the spread, with case and hospitalization numbers reaching all-time highs in the winter of 2022 [2,3]. With governments gradually lifting restrictive measures and reopening borders, it is critical that, in addition to getting vaccinated, individuals continue to follow nonpharmaceutical interventions—including wearing face masks, physical distancing, staying home when ill, and avoiding crowded indoor spaces—to limit the spread of this highly transmissible virus, especially as newer more transmissible variants continue to emerge [4-7].

Mandates and government-imposed restrictions are important policy tools for limiting the spread of COVID-19, but they are insufficient on their own and must be complemented by softer interventions designed to increase compliance with public health guidelines. Convincing citizens to freely adhere to social distancing, masking, and other preventive behaviors requires persuasive communication going beyond providing information on the risks of the pandemic. Public health organizations and governments need to understand how to best frame messages to effectively appeal to different audiences [8].

The primary objective of this study was to empirically test the effectiveness of message framings emphasizing a set of carefully selected persuasive appeals at improving people’s intentions to engage in health protective behaviors. Another aim of the study was to characterize the target audience most susceptible to respond positively to the persuasive appeals. The findings are intended to guide the design and development of public health campaigns in Canada.

Message Framing and Adherence to Public Health Measures

In the past year, numerous studies have investigated the impact of various persuasive appeals on people’s attitudes and intentions around COVID-19–related behaviors. The studies varied in their methods and procedures and produced mixed results. Messages using prosocial, altruistic, other-focused, or community-focused appeals were generally more persuasive than messages using self-interested, self-protective, or threatening appeals [9-16]. Likewise, gain-framed messages were typically more effective than loss-framed messages [17,18], although at least one study found the opposite result [19]. Moreover, messages invoking social norms do not seem to be particularly effective [20,21].

In a comprehensive analysis, Pink and colleagues [21] tested 56 short messages using a wide range of framings, including some of the appeals mentioned above. They found no consistent effects for any of the tested messages. Nevertheless, a message using a reciprocity appeal performed the best in three of their five studies.

The present research adds to this body of work by testing the effectiveness of five appeals (deontological moral frame, empathy, goal proximity, identifiable victim, and reciprocity) at improving people’s intentions to adhere to public health measures. This study differs from prior work in at least two important aspects. First, the pandemic context at the time of our study (early March 2021) is unlike that characterizing the early stages of the pandemic when most previous studies were conducted. At the time of our study, there had been over 880,000 confirmed COVID-19 cases in Canada, including over 22,000 deaths. Vaccine supply was limited with just over 2 million doses administered by March 3, 2021 [2]. Although the daily COVID-19 activity had been declining from mid-January through mid-February, it has leveled off since. The 7-day average was under 3000 new cases a day nationwide, but variants of concern (B.1.1.7 and B1.351) had emerged [22]. Masking in public places was mandated in most jurisdictions, and the public was advised to limit travel and minimize contact with people outside of their household [22]. The difference in context alone may result in notable differences in how people process and respond to various persuasive messages.

Previous experiments have largely neglected the role of baseline attitudes and intentions when testing for differences between messages. In contrast, we expected baseline intentions to have a significant impact on how people respond to persuasive messages. People who are highly compliant to begin with have little room left for improvement. Thus, we expected the effect of persuasive appeals to be stronger among those with relatively lower baseline intentions. This is significant because those who are less compliant with public health measures are a critical target for behavior change.

Five Persuasive Appeals

This study focused on the impact of five persuasive appeals: deontological moral frame, empathy, identifiable victim, goal proximity, and reciprocity. Deontological moral frames are frequently encountered in the current public discourse; they appeal to the sense of duty and responsibilities we have to our families and communities [23]. Prior research suggests that agents making deontological judgments are perceived to be more trustworthy than agents making utilitarian judgments [24,25], even when they are not actually more trustworthy [26]. Moreover, research using machine learning found that moral identity is a strong predictor of adherence to public health measures [27]. Thus, we expect persuasive appeals that use deontological moral frames to help increase adherence to public health measures.

Empathy—understanding and feeling concerned for vulnerable others—has been found to increase altruism and caring, and to motivate helping behavior [28-30]. Thus, inducing empathy by highlighting that the sick, elderly, and immunocompromised need our help is expected to increase adoption of health protective behaviors [13,15].

Goal-proximity appeals emphasize that better days are approaching. This is important because people’s motivation to comply with public health advice has declined since the pandemic’s early days. A Gallup study tracking social distancing behaviors found that the percentage of Americans practicing social distancing dropped steadily over time, from 75% in April 2020 to 38% in March 2021 [31]. A drop in motivation over the course of goal pursuit is not uncommon when pursuing goals with no clear end states or when the tasks required to achieve the goal are difficult [32]. Fortunately, motivational strength tends to increase as the distance to the goal decreases. The goal-gradient hypothesis holds that people apply more effort and persistence as they get closer to a goal’s end state [33-37]. The third message tested in this study relies on this motivational property.

The fourth message relies on the persuasive power of identifiable victims. The identifiable victim effect refers to people’s propensity to offer more help to specific, identifiable victims rather than to anonymous, statistical victims [38-40]. This effect has been attributed to the fact that identifiable victims evoke more powerful emotional responses than statistical victims [38,41]. The identifiable victim effect also arises because people believe their contribution will have a greater impact on an identified victim than on a large group of unidentified victims [39].

Our fifth message relies on the principle of reciprocity. According to Cialdini [42], “all societies subscribe to a norm that obligates individuals to repay in kind what they have received” (page 76). The reciprocity code is not limited to gifts and favors but also includes concessions, whereby people are more likely to make concessions to those who have made concessions to them [43,44]. Accordingly, our reciprocity message emphasizes the sacrifices health care workers are making to help and protect us, and asks that we return the favor by adhering to health protective behaviors.

Individual Differences in Compliance With Public Health Measures

We expect persuasive communication to have a greater impact among individuals who have lower initial intentions to adhere with public health measures. This is because individuals who have high initial intentions have little room left for improvement; that is, they are already persuaded and further exposure to persuasive communication is unlikely to change their intentions. From a campaign planning perspective, it is important to identify who these individuals might be so that the messages can be efficiently targeted.

The existing literature points to significant variability in the levels of adherence to public health measures [45-52]. A recent review of 29 empirical studies concluded that greater adherence to public health measures is reliably associated with being older, identifying as female, trusting governments, perceiving COVID-19 as a threat, and accessing information through traditional news media [50]. Variability in uptake of public health behaviors was also linked to differences in political ideology [51,52] and perceived responsibility for others [53]. In this study, we measured these characteristics and examined their associations with baseline intentions.

Participants and Procedure

A representative sample of adult Canadian residents with respect to age, ethnicity, and province of residence was recruited by the research firm Critical Mass between March 3 and March 6, 2021. A description of the study was posted on Lucid Marketplace, a third-party platform that maintains an online research panel of 15 million verified users. Users from Canada were invited to visit a screening page assessing demographic and geographic variables. Target quotas for province of residence, age, gender, and ethnicity were set to obtain a demographically representative sample based on the 2016 census data (see Table S1 in Multimedia Appendix 1 for details on the quota system).

Upon consenting in writing, participants reported on their intentions to engage in a set of prevention behaviors over the coming weeks (T1). They were then randomly assigned to an active control or one of five persuasive appeal conditions (control vs deontological vs empathy vs goal proximity vs reciprocity vs identifiable victim) and reported on their intentions to engage in the same set of prevention behaviors a second time (T2). This design allowed us to examine whether the effectiveness of persuasive appeals varies as a function of initial prevention intentions. Finally, participants completed a series of questions assessing potential correlates of prevention intentions. These included measures of political orientation, trust in institutions, perceived threat of COVID-19, and perceived responsibility toward others.

Ethics Approval

This study was approved by the University of Calgary Conjoint Research Ethics Board (REB21-0173) and was conducted according to the principles expressed in the Declaration of Helsinki.


Index variables for intentions to engage in prevention behaviors (pre- and posttreatment) were created by averaging across six items: (1) Limit my physical contact with others when possible, (2) Completely avoid any unnecessary physical contact with others (eg, hugging or handshakes), (3) Avoid crowded indoor spaces, (4) Wear a mask when I leave the house, (5) Wash my hands as much as possible, and (6) Stay home when mildly sick. These items were measured on 100-point sliding scales (0=strongly disagree, 50=neither agree nor disagree, 100=strongly agree).

Persuasive appeals were manipulated using promotional flyers ostensibly distributed by the Public Health Agency of Canada. In the control condition, the flyer contained a simple list of what participants can do to help prevent the spread of COVID-19. In each of the five persuasion conditions, the flyer contained the same basic information and a unique persuasive appeal (see Figure 1 for an example and Figures S1-S5 in Multimedia Appendix 1 for the remaining flyers). The wording of the messages is shown in Textbox 1.

Trust in various institutions (politicians, civil servants, public health officials, physicians, other health care providers [eg, nurses, pharmacists], scientists, journalists, and pharmaceutical companies) was measured using eight items (α=.91) on 100-point sliding scales (0=do not trust at all, 100=trust completely).

Perceived COVID-19 threat was measured using four items (α=.89) adapted from previous research [11]. A sample item is: “To what extent are you afraid of contracting COVID-19 because of the consequences for you personally/your community?” (0=not at all, 50=to a moderate extent, 100=to an enormous extent).

Perceived responsibility toward others was assessed using four items (α=.94) adapted from previous research [18]. A sample item is: “I owe it to my family to do whatever I can to stop the spread of COVID-19” (1=strongly disagree, 7=strongly agree).

Finally, political orientation was measured using the following item: “If you think about your own political views, where would you classify your views on this scale?” (1=very liberal, 7=very conservative).

Figure 1. Sample flyer: empathy appeal.
View this figure
Messages across appeal conditions.


The virus spreads mainly between people who are in close contact with one another. You can help prevent the spread of COVID-19. We can all do our part:

  • Avoid social gatherings.
  • Wear a mask when you go out.
  • Stay at least six feet away from people outside your household.
  • Wash your hands often with soap and water.

These actions prevent the spread of COVID-19.


The virus spreads mainly between people who are in close contact with one another. You can help prevent the spread of COVID-19. We can all do our part:

  • Avoid social gatherings.
  • Wear a mask when you go out.
  • Stay at least six feet away from people outside your household.
  • Wash your hands often with soap and water.

We all need to do this, however difficult, because it is the right thing to do: it is our duty and responsibility to protect our families, friends, and fellow citizens.


The sick, elderly, and immunocompromised need our help. We all have a choice. If we don’t take the right actions, we risk the lives of others. But we can protect those most likely to be harmed. We can protect those who are vulnerable by taking simple steps:

  • Avoid social gatherings.
  • Wear a mask when you go out.
  • Stay at least six feet away from people outside your household.
  • Wash your hands often with soap and water.

Take action to protect those who are vulnerable!

Identifiable victim

A few weeks ago, Sam was a healthy 26-year-old with no medical complications. Then he suddenly came down with a bad cough and a feeling like he could not breathe. He tested positive for COVID-19 and is now hospitalized, receiving oxygen from a ventilator, and fighting for his life. This could be any of us. Reduce the risk to yourself and others:

  • Avoid social gatherings.
  • Wear a mask when you go out.
  • Stay at least six feet away from people outside your household.
  • Wash your hands often with soap and water.

If we take these actions, we can prevent more people from suffering the way Sam has.

Goal proximity

The recent development of safe and effective vaccines gives us great hope. We see the light at the end of the tunnel, but we are not quite there yet. Until a large proportion of the population is immunized, we must remain vigilant and double our efforts to prevent the spread of COVID-19.

  • Avoid social gatherings.
  • Wear a mask when you go out.
  • Stay at least six feet away from people outside your household.
  • Wash your hands often with soap and water.

These actions prevent the spread of COVID-19.


Doctors, nurses, and other health care workers are working around the clock, often risking their lives to care for patients with the coronavirus. Working long hours in highly infectious environments, many of them are falling ill. As our health care workers put their lives on the line, we can do our part:

  • Avoid social gatherings.
  • Wear a mask when you go out.
  • Stay at least six feet away from people outside your household.
  • Wash your hands often with soap and water.

Our brave health care workers have sacrificed to help others. We should take action too.

Textbox 1. Messages across appeal conditions.

Data Analysis

First, we sought to address the broad question: does exposure to messages using persuasive appeals improve intentions to engage in prevention behaviors more than exposure to the control message? Given the structure in our data (each participant provided two sets of ratings), we fitted a linear mixed effects model (estimated using maximum likelihood) with intention to engage in prevention behaviors as the outcome variable; random intercepts for participants (id); and fixed effects for appeal condition, time of rating, and their interaction. In this analysis, the P values were estimated via t-tests using the Satterthwaite approximation to degrees of freedom. Effect sizes for the fixed effects are indicated by the standardized regression coefficients (β) and their 95% CIs.

We performed a series of moderated regressions (estimated using ordinary least squares [OLS]) to investigate whether the effectiveness of persuasive appeals varies as a function of baseline prevention intentions. We used change in intentions as the outcome variable, persuasion appeal as a binary predictor, and baseline intentions as a continuous moderator.

To help characterize the target audience, we examined the association of baseline intentions with demographic variables, including age, gender, ethnic background, education, and geographic region, as well as attitudinal variables such as perceived COVID-19 threat, perceived responsibility toward others, trust in institutions, and political orientation.

We fitted a linear model (estimated using OLS) using all predictors. The continuous predictors (age, threat, responsibility, trust, and political orientation) were mean-centered and the categorical predictors were dummy-coded. The ethnic background variable was constructed by recoding the original ethnicity variable into a binary variable (0=ethnic majority, 1=ethnic minority). Education was modified by combining the “less than high school” and “high school” categories into a single “high school or less” category, which served as the baseline group in the analysis. The region variable was constructed by collapsing the Newfoundland and Labrador, Nova Scotia, New Brunswick, and Territories categories in the province variable into a single “Maritimes and Territories” category. Ontario was set as the baseline category for the five-level region variable and female was set as the baseline category for the three-level gender variable.

Data analysis was performed using the statistical program R version 4.0.2 [54], and the level of statistical significance was set at α=.05.

Participant Characteristics

A total of 7079 respondents visited the screening page. Of those, 3746 qualified for the main study based on the quota requirements. Of the qualified respondents, 78 failed to complete the survey, resulting in a final sample of 3668 participants (see Table 1 for sample characteristics). Those who failed to complete the survey were demographically similar to those who completed the survey, but were predominantly from the provinces of Quebec (40%) and Nova Scotia (19%) (see Table S1 in Multimedia Appendix 1).

Table 1. Sample characteristics.
CharacteristicOverall (N=3668), n (%)Control (n=582), n (%)Deontological (n= 622), n (%)Empathy (n=624), n (%)Proximity (n= 603), n (%)Reciprocity (n=623), n (%)Victim (n=614), n (%)P valuea
Gender (n=3668).27

Female2202 (60.03)334 (57.4)380 (61.1)353 (56.6)359 (59.5)386 (62.0)390 (63.5)

Male1450 (39.53)245 (42.1)238 (38.3)267 (42.8)243 (40.3)234 (37.6)223 (36.3)

Other16 (0.44)3 (0.5)4 (0.6)4 (0.6)1 (0.2)3 (0.5)1 (0.2)
Age group (years) (n=3667).73

18-24345 (9.41)54 (9.3)65 (10.5)60 (9.6)52 (8.6)54 (8.7)60 (9.8)

25-34690 (18.82)118 (20.3)118 (19.0)125 (20.0)115 (19.1)113 (18.1)101 (16.4)

35-44785 (21.41)119 (20.4)125 (20.1)146 (23.4)145 (24.0)131 (21.0)119 (19.4)

45-54599 (16.33)97 (16.7)106 (17.0)101 (16.2)86 (14.3)103 (16.5)106 (17.3)

55-64595 (16.23)100 (17.2)103 (16.6)91 (14.6)102 (16.9)100 (16.1)99 (16.1)

65-99653 (17.81)94 (16.2)105 (16.9)101 (16.2)103 (17.1)121 (19.4)129 (21.0)
Ethnicity (n=3650).28

White2840 (77.81)448 (77.0)479 (77.0)478 (76.6)465 (77.1)488 (78.3)482 (78.5)

Black110 (3.01)21 (3.6)21 (3.4)18 (2.9)16 (2.7)15 (2.4)19 (3.1)

East Asian297 (8.14)48 (8.3)49 (7.9)63 (10.1)44 (7.3)47 (7.6)46 (7.6)

South Asian193 (5.29)29 (5.0)27 (4.4)34 (5.5)30 (5.0)42 (6.8)31 (5.1)

Indigenous63 (1.73)12 (2.1)13 (2.1)10 (1.6)6 (1.0)11 (1.8)11 (1.8)

Other147 (4.03)21 (3.6)29 (4.7)18 (2.9)40 (6.7)19 (3.1)20 (3.3)
Education (n=3667).23

Less than high school86 (2.35)12 (2.1)11 (1.8)12 (1.9)14 (2.3)26 (4.2)11 (1.8)

High school718 (19.58)112 (19.2)108 (17.4)127 (20%)107 (17.7)121 (19.4)143 (23.3)

Some college631 (17.21)98 (16.8)111 (17.8)101 (16%)113 (18.7)100 (16.1)108 (17.6)

College834 (22.74)128 (22.0)155 (24.9)125 (20.4)143 (23.7)149 (23.9)134 (21.8)

University1007 (27.46)169 (29.0)174 (28.0)185 (29.6)156 (25.9)165 (26.5)158 (25.7)

Graduate degree391 (10.66)63 (10.8)63 (10.1)74 (11.9)69 (11.4)62 (10.0)60 (9.8)
Province (n=3668).79

Newfoundland and Labrador74 (2.02)6 (1.0)18 (2.9)13 (2.1)9 (1.5)14 (2.2)14 (2.3)

Prince Edward Island19 (0.52)3 (0.5)5 (0.8)4 (0.6)4 (0.7)2 (0.3)1 (0.2)

New Brunswick96 (2.62)11 (1.9)20 (3.2)21 (3.4)15 (2.5)10 (1.6)19 (3.1)

Nova Scotia122 (3.33)22 (3.8)23 (3.7)24 (3.8)16 (2.7)22 (3.5)15 (2.4)

Quebec472 (12.87)72 (12.4)87 (14.0)83 (13.3)76 (12.6)66 (10.6)88 (14)

Ontario1555 (42.39)256 (44.0)259 (41.6)251 (40.2)267 (44.3)273 (43.8)249 (40.6)

Manitoba155 (4.23)26 (4.5)26 (4.2)26 (4.2)26 (4.3)30 (4.8)21 (3.4)

Saskatchewan128 (3.49)18 (3.1)21 (3.4)22 (3.5)18 (3.0)20 (3.2)29 (4.7)

Alberta470 (12.81)76 (13.1)77 (12.4)84 (13.5)65 (10.8)86 (13.8)82 (13.4)

British Columbia569 (15.51)89 (15.3)85 (13.7)95 (15.2)105 (17.4)100 (16.1)95 (15.5)

Territoriesb8 (0.22)3 (0.5)1 (0.2)1 (0.2)2 (0.3)0 (0)1 (0.2)

aPearson χ2 test.

bTerritories=Yukon, Northwest Territories, and Nunavut.

Intentions to Engage in Prevention Behaviors

The results of the fixed factors in the mixed effects model are summarized in Table 2 (random effects: σ2=18.90, τ00id=282.54, intraclass correlation coefficient=0.94, Nid=3668, observations=7331, marginal R2=0.006, conditional R2= 0.938). Prior to exposure to the persuasive appeals, participants in all conditions reported similarly high intentions to engage in prevention behaviors. Prevention scores at T1 did not differ significantly between any appeal condition and the control condition, as shown in Table 2 (P values for deontological, empathy, goal proximity, reciprocity, and victim are all greater than .05). This confirmed that random assignment produced groups with equivalent baselines. Furthermore, exposure to a reminder message about prevention behaviors (ie, control condition) increased participants’ intentions to engage in prevention behaviors (see Time [T2] variable in Table 2). Additionally, exposure to a persuasive message using an empathy appeal resulted in a larger increase in intentions to engage in prevention behaviors relative to the control message (Table 2).

Exposure to messages using other types of appeals (deontological, goal proximity, reciprocity, and victim) produced positive changes in intentions to engage in prevention behaviors (see Table 3), but these changes did not differ in magnitude from those produced by exposure to a simple reminder message (all P>.05). Figure 2 shows the estimated marginal means for each group and their 95% CIs.

Table 2. Mixed effects regression results for intentions to engage in prevention behaviors.
PredictorsEstimate, b (SE)t statisticdfP valueβ (95% CI)
(Intercept)87.11 (0.72)121.043905.10<.001–.08 (–.16 to .00)
Time [T2a]2.12 (0.25)8.323663.20<.001.12 (.09 to .15)
Deontological0.37 (1.00)0.373905.10.71.02 (–.09 to .13)
Empathy–0.61 (1.00)–0.613905.10.54–.03 (–.15 to .08)
Proximity–0.52 (1.01)–0.513905.10.61–.03 (–.14 to .08)
Reciprocity0.71 (1.00)0.703905.10.48.04 (–.07 to .15)
Victim0.44 (1.00)0.443905.10.66.03 (–.09 to .14)
T2×Deontological0.47 (0.35)1.333663.38.19.03 (–.01 to .07)
T2×Empathy1.04 (0.35)2.933663.38.003.06 (.02 to .10)
T2×Proximity0.06 (0.36)0.173663.57.87.00 (–.04 to .04)
T2×Reciprocity0.60 (0.35)1.693663.38.09.03 (–.01 to .07)
T2×Victim0.53 (0.36)1.483663.20.14.03 (–.01 to .07)

aT2: posttest time point.

Table 3. Intention to engage in prevention behaviors before (T1) and after (T2) exposure to various appeals.
AppealIntention_T1Intention_T2T2–T1t statisticdfP value
Figure 2. Intention to engage in prevention behaviors across appeal conditions and measurement. Data are presented as marginal means with 95% CIs.
View this figure

Moderating Effect of Baseline Intentions

The preceding analysis suggested that, apart from empathy, the use of persuasive appeals does not improve intentions to engage in prevention behaviors beyond a simple reminder message. However, we expected the effectiveness of persuasive appeals to vary according to people’s initial dispositions. Persuasive appeals are likely effective when baseline intentions are relatively low, but may have a limited impact when baseline intentions are so high that there is little room for improvement. Results from the moderated regressions were consistent with our expectations (see Table 4). The appeal×baseline intentions interaction was statistically significant for all but the goal-proximity appeal, suggesting that the effectiveness of the deontological, empathy, reciprocity, and identifiable victim appeals indeed depends on the level of initial intentions.

We followed up with floodlight analyses [55] of each significant interaction. As shown in Figure 3, the conditional effect of seeing a deontological appeal was significant only among participants who had a score of 85.5 or below on the initial intentions measure (30.2% of participants; mean 66.4). In other words, people with lower baseline intentions increased their intentions to engage in prevention behaviors more after seeing a message featuring a deontological appeal than after seeing a message featuring a simple reminder. In contrast, those with high baseline intentions (higher than 85.5; 69.8% of participants; mean 96.2) did not differ significantly in how much they changed their intentions when they saw a message featuring a deontological appeal or a message featuring a reminder.

We observed similar patterns with the other appeals. The conditional effect of empathy was significant only among participants scoring 90.1 or lower on initial intentions (39.5% of participants; mean 71.5), the conditional effect of reciprocity was significant only for those scoring 87.8 or lower on initial intentions (44.1% of participants; mean 68.7), and the conditional effect of identifiable victim was only significant for those scoring 84.8% or lower on initial intentions (29.3% of participants; mean 65.7).

Table 4. Effect of appeal×initial intentions interaction on change in intentions to engage in prevention behavior.
Appeal×baseline intentionsEstimate, b (SE)t statisticdfP valueβ (95% CI)
Deontological–0.08 (0.02)–4.291199<.001–.12 (–.17 to –.06)
Empathy–0.09 (0.02)–4.601201<.001–.13 (–.18 to –.07)
Proximity–0.02 (0.02)–1.141179.26–.03 (–.09 to –.02)
Reciprocity–0.08 (0.02)–4.381200<.001–.12 (–.18 to –.07)
Victim–0.05 (0.02)–2.751192.006–.08 (–.13 to –.02)
Figure 3. Floodlight analysis of the interactive effects of appeal and baseline intentions. n.s: not significant (P>.05).
View this figure

Predictors of Baseline Intentions

The moderation analysis implied that a public health campaign using persuasive appeals would be most effective when targeting individuals with lower baseline intentions: but who might these individuals be?

The regression model using all demographic and attitudinal predictors explained a statistically significant and substantial proportion of the variance (R2=0.51, F16, 3415=224.2, P<.001, adjusted R2=0.51). As shown in Table 5, baseline intentions increased with age, perception of COVID-19 threat, perceived responsibility, and trust in institutions. Conversely, baseline intentions decreased with political conservatism, were lower for males relative to females, and were lower in the Prairies compared to Ontario. Neither education level nor ethnic background was significantly uniquely associated with baseline intentions to engage in prevention behaviors.

Table 5. Multivariable regression model of initial intentions.
PredictorsEstimate, b (SE)t (df=3415)P valueβ (95% CI)
(Intercept)88.44 (0.57)155.99<.001.08 (.01 to .14)
Age0.06 (0.01)4.41<.001.06 (.03 to .08)
Gender [Male]–1.74 (0.44)–3.94<.001–.10 (–.15 to –.05)
Gender [Other]–3.01 (3.19)–0.95.34–.17 (–.53 to .18)
Ethnic [Minority]–0.07 (0.54)–0.13.90–.00 (–.06 to .06)
Education [Some college]–0.23 (0.68)–0.35.73–.01 (–.09 to .06)
Education [College]–0.40 (0.63)–0.63.53–.02 (–.09 to .05)
Education [University]0.38 (0.61) (–.05 to .09)
Education [Graduate degree]–0.24 (0.79)–0.30.76–.01 (–.10 to .07)
Region [Maritimes]–1.16 (0.79)–1.46.14–.07 (–.16 to .02)
Region [Quebec]–0.22 (0.67)–0.33.74–.01 (–.09 to .06)
Region [Prairies]–1.53 (0.57)–2.68.007–.09 (–.05 to –.02)
Region [British Columbia]–1.02 (0.63)–1.64.10–.06 (–.13 to .01)
Political orientation–0.39 (0.14)–2.76.006–.03 (–.06 to –.01)
COVID-19 threat0.15 (0.01)13.53<.0010.21 (.18 to .24)
Responsibility7.66 (0.24)31.90<.001.50 (.47 to .53)
Trust0.08 (0.01)6.04<.001.09 (.06 to .12)

At the time of writing, Canada was entering the fourth wave of COVID-19, with case and hospitalization numbers projected to spike in the coming weeks [2,22]. Maximizing vaccination coverage is paramount, but support for public health measures, including physical distancing, masking, staying home while sick, and avoiding crowded indoor spaces, is also critical for limiting the spread of the virus. This is particularly important since some jurisdictions have moved away from mandatory to recommended measures, relying on the public to make adherence decisions. There is an urgent need for effective messaging to increase adherence to public health measures.

Through a randomized online experiment, we tested the effectiveness of five messages featuring different persuasive appeals (deontological vs empathy vs goal proximity vs reciprocity vs identifiable victim) relative to a control message that simply listed the actions participants could take to help prevent the spread of COVID-19. A pretest-posttest design allowed us to assess and compare the change in intentions after exposure to the various messages. The study produced notable insights. First, baseline intentions across all conditions were relatively high (mean 87.18, SD 17.70 on a 100-point scale). Despite our effort to recruit a demographically representative sample, our pool of respondents may have been skewed toward higher compliance. High baseline intentions could also reflect a degree of social desirability bias in the responses. It is worth noting that similarly high levels of self-reported intentions have been observed in prior research [13,21].

Second, exposure to all messages, including the control message, resulted in a small but statistically significant increase in behavioral intentions. Moreover, the message featuring an empathy appeal increased behavioral intentions to a greater extent than the control message. Given how high intentions were to begin with, a small increase should be considered a significant win.

Third, the impact of persuasive appeals on change in intentions depended on how compliant people were in the first place. For those with lower baseline intentions, messages featuring empathy, deontological, reciprocity, and identifiable victim appeals resulted in greater change than the control message. These results are encouraging, as the intended persuasion targets are precisely those who are less compliant with public health measures.

Finally, the study confirmed much of what prior research had found regarding the correlates of public health compliance. Lower baseline intentions were associated with being male, younger, more politically conservative, residing in the Prairies, perceiving lower levels of COVID-19 threat, accepting less responsibility for the well-being of others, and lacking trust in public institutions [49-53]. These results provide a clear and actionable profile of the audiences that need to be targeted to maximize the efficiency of public health campaigns.

While the findings are reasonably informative, it is important to keep the study’s limitations in mind. For instance, the main outcome consisted of self-reported behavioral intentions. Since a gap often exists between intentions and behavior [56], the observed outcomes may not track perfectly with actual behavior. Moreover, as is the case for all studies of this kind, the results are likely context-dependent. The same appeals may produce vastly different responses in different countries and at different times, depending on cultural values and the COVID-19 situation on the ground. Thus, it is important not to overgeneralize when interpreting the results.

Importantly, the study used a single brief exposure to the messages, offering a conservative test of the messages’ persuasive power. Future research could investigate whether more frequent exposure or a prolonged exposure period would have a stronger impact. Future research could also test the impact of varying the message format (eg, video vs audio vs print), medium (eg, social media vs traditional media), and source. While the Public Health Agency of Canada is generally a trusted source [53], some groups may respond more positively to other sources (eg, trusted religious and community leaders). Although the focus of this study has been squarely on persuasive appeals, public health campaigns would do well to customize not only the content of the message but also its source, format, and media to maximize its impact across different audiences.


We would like to thank the team members at Critical Mass Inc who contributed to participant recruitment. This study was supported by an ImplementAB.digH Program Grant from Alberta Innovates (Grant 202101302).

Authors' Contributions

MM, JLB, RL, MMF, JCB, KC, RJO, CC, TT, DAM, and JH conceived and designed the study. MM performed data analysis and wrote the first draft of the manuscript. All authors participated in critical revision of the manuscript and approved the final version. MM is the guarantor of the work and takes responsibility for the integrity of the data.

Conflicts of Interest

DAM reports non-financial support from ISPOR, grants from Canadian Institutes of Health Research (CIHR), Genome Canada, Arthritis Society, and Alberta Innovates; personal fees from Analytica, Illumina, and Novartis. The grants and fees were received during the the timeline of this study but were unrelated to it.

Editorial Notice

This randomized study was only retrospectively registered, explained by authors with the formative nature of the study. The editor granted an exception from ICMJE rules mandating prospective registration of randomized trials because the risk of bias appears low and the study was considered formative. However, readers are advised to carefully assess the validity of any potential explicit or implicit claims related to primary outcomes or effectiveness, as retrospective registration does not prevent authors from changing their outcome measures retrospectively.

Multimedia Appendix 1

Quota system details (Table S1) and flyers for the persuasion conditions (Figures S1-S5).

DOCX File , 4205 KB

Multimedia Appendix 2

CONSORT checklist.

DOCX File , 41 KB

  1. WHO Cornavirus (COVID-19) Dashboard. World Health Organization.   URL: [accessed 2022-07-21]
  2. Canada: Coronavirus pandemic country profile. Our World in Data.   URL: [accessed 2022-07-21]
  3. Update on COVID-19 in Canada: epidemiology and preparedness. Public Health Agency of Canada. 2022 Apr 01.   URL: https:/​/www.​​content/​dam/​phac-aspc/​documents/​services/​diseases-maladies/​coronavirus-disease-covid-19/​epidemiological-economic-research-data/​update-covid-19-canada-epidemiology-modelling-20220401-en.​pdf [accessed 2022-07-21]
  4. Bai Y, Yao L, Wei T, Tian F, Jin D, Chen L, et al. Presumed asymptomatic carrier transmission of COVID-19. JAMA 2020 Apr 14;323(14):1406-1407 [FREE Full text] [CrossRef] [Medline]
  5. Bi Q, Wu Y, Mei S, Ye C, Zou X, Zhang Z, et al. Epidemiology and transmission of COVID-19 in 391 cases and 1286 of their close contacts in Shenzhen, China: a retrospective cohort study. Lancet Infect Dis 2020 Aug;20(8):911-919 [FREE Full text] [CrossRef] [Medline]
  6. Ferretti L, Wymant C, Kendall M, Zhao L, Nurtay A, Abeler-Dörner L, et al. Quantifying SARS-CoV-2 transmission suggests epidemic control with digital contact tracing. Science 2020 May 08;368(6491):eabb6936 [FREE Full text] [CrossRef] [Medline]
  7. Rothe C, Schunk M, Sothmann P, Bretzel G, Froeschl G, Wallrauch C, et al. Transmission of 2019-nCoV infection from an asymptomatic contact in Germany. N Engl J Med 2020 Mar 05;382(10):970-971 [FREE Full text] [CrossRef] [Medline]
  8. Van Bavel JJ, Baicker K, Boggio P, Capraro V, Cichocka A, Cikara M, et al. Using social and behavioural science to support COVID-19 pandemic response. Nat Hum Behav 2020 May;4(5):460-471. [CrossRef] [Medline]
  9. Browning A, Moss M, Berkman E. Leveraging evidence-based messaging to prevent the spread of COVID-19. PsyArXiv.. Preprint posted online July 1, 2021.   URL: [accessed 2022-07-21]
  10. Capraro V, Barcelo H. The effect of messaging and gender on intentions to wear a face covering to slow down COVID-19 transmission. PsyArXiv.. Preprint posted online May 11, 2020.   URL: [accessed 2022-07-21]
  11. Ceylan M, Hayran C. Message framing effects on individuals' social distancing and helping behavior during the COVID-19 pandemic. Front Psychol 2021;12:579164 [FREE Full text] [CrossRef] [Medline]
  12. Heffner J, Vives M, FeldmanHall O. Emotional responses to prosocial messages increase willingness to self-isolate during the COVID-19 pandemic. Pers Individ Dif 2021 Mar 15;170:110420 [FREE Full text] [CrossRef] [Medline]
  13. Jordan J, Yoeli E, Rand D. Don't get it or don't spread it: comparing self-interested versus prosocial motivations for COVID-19 prevention behaviors. Sci Rep 2021 Oct 12;11(1):20222. [CrossRef] [Medline]
  14. Luttrell A, Petty RE. Evaluations of self-focused versus other-focused arguments for social distancing: an extension of moral matching effects. Soc Psychol Person Sci 2020 Aug 18;12(6):946-954. [CrossRef]
  15. Pfattheicher S, Nockur L, Böhm R, Sassenrath C, Petersen MB. The emotional path to action: empathy promotes physical distancing and wearing of face masks during the COVID-19 pandemic. Psychol Sci 2020 Nov 29;31(11):1363-1373. [CrossRef] [Medline]
  16. Capraro V, Boggio P, Böhm R, Perc M, Sjåstad H. Cooperationacting for the greater good during the COVID-19 pandemic. PsyArXiv.. Preprint posted online May 31, 2021.   URL: [accessed 2022-07-21]
  17. Gantiva C, Jiménez-Leal W, Urriago-Rayo J. Framing messages to deal with the COVID-19 crisis: the role of loss/gain frames and content. Front Psychol 2021;12:568212 [FREE Full text] [CrossRef] [Medline]
  18. Sasaki S, Kurokawa H, Ohtake F. Effective but fragile? Responses to repeated nudge-based messages for preventing the spread of COVID-19 infection. Jpn Econ Rev 2021;72(3):371-408 [FREE Full text] [CrossRef] [Medline]
  19. Steffen J, Cheng J. The influence of gain-loss framing and its interaction with political ideology on social distancing and mask wearing compliance during the COVID-19 pandemic. Curr Psychol 2021 Jul 29:1-11 [FREE Full text] [CrossRef] [Medline]
  20. Bilancini E, Boncinelli L, Capraro V, Celadin T, Di Paolo R. The effect of norm-based messages on reading and understanding COVID-19 pandemic response governmental rules. J Behav Econ Policy 2020;4(COVID-19 Special Issue):45-55 [FREE Full text] [CrossRef]
  21. Pink S, Stagnaro M, Chu J, Mernyk J, Voelkel J, Willer R. Short messages encouraging compliance with COVID-19 public health guidelines have minimal persuasive effects. PsyArXiv.. Preprint posted online August 10, 2020.   URL: [accessed 2022-07-21]
  22. Statement from the chief public health officer of Canada on March 3, 2021. Public Health Agency of Canada. 2021.   URL: https:/​/www.​​en/​public-health/​news/​2021/​03/​statement-from-the-chief-public-health-officer-of-canada-on-march-3-2021.​html [accessed 2022-07-21]
  23. Everett J, Colombatto C, Chituc V, Brady W, Crockett M. The effectiveness of moral messages on public health behavioral intentions during the COVID-19 pandemic. PsyArXiv.. Preprint posted online March 20, 2020.   URL: [accessed 2022-07-21]
  24. Everett J, Pizarro D, Crockett M. Inference of trustworthiness from intuitive moral judgments. J Exp Psychol Gen 2016 Jun;145(6):772-787. [CrossRef] [Medline]
  25. Rom S, Weiss A, Conway P. Judging those who judge: Perceivers infer the roles of affect and cognition underpinning others' moral dilemma responses. J Exp Soc Psychol 2017 Mar;69:44-58. [CrossRef]
  26. Capraro V, Sippel J, Zhao B, Hornischer L, Savary M, Terzopoulou Z, et al. People making deontological judgments in the Trapdoor dilemma are perceived to be more prosocial in economic games than they actually are. PLoS One 2018;13(10):e0205066 [FREE Full text] [CrossRef] [Medline]
  27. Pavlović T, Azevedo F, De K, Riaño-Moreno JC, Maglić M, Gkinopoulos T, et al. Predicting attitudinal and behavioral responses to COVID-19 pandemic using machine learning. PNAS Nexus 2022 Jul;1(3):pgac093 [FREE Full text] [CrossRef] [Medline]
  28. Batson CD, Duncan BD, Ackerman P, Buckley T, Birch K. Is empathic emotion a source of altruistic motivation? J Person Soc Psychol 1981 Feb;40(2):290-302 [FREE Full text] [CrossRef]
  29. Sassenrath C, Diefenbacher S, Siegel A, Keller J. A person-oriented approach to hand hygiene behaviour: emotional empathy fosters hand hygiene practice. Psychol Health 2016 Oct 15;31(2):205-227. [CrossRef] [Medline]
  30. Todd AR, Burgmer P. Perspective taking and automatic intergroup evaluation change: testing an associative self-anchoring account. J Pers Soc Psychol 2013 May;104(5):786-802. [CrossRef] [Medline]
  31. Jones JM. Social distancing behaviors drop as US vaccinations rise. Gallup News. 2021 Mar 31.   URL: [accessed 2022-07-21]
  32. Touré‐Tillery M, Fishbach A. The course of motivation. J Consum Psychol 2011 May 17;21(4):414-423. [CrossRef]
  33. Hull CL. The goal-gradient hypothesis and maze learning. Psychol Rev 1932 Jan;39(1):25-43. [CrossRef]
  34. Brown JS. Gradients of approach and avoidance responses and their relation to level of motivation. J Comp Physiol Psychol 1948 Dec;41(6):450-465. [CrossRef] [Medline]
  35. Förster J, Higgins ET, Idson LC. Approach and avoidance strength during goal attainment: regulatory focus and the "goal looms larger" effect. J Person Soc Psychol 1998;75(5):1115-1131. [CrossRef]
  36. Kivetz R, Urminsky O, Zheng Y. The goal-gradient hypothesis resurrected: purchase acceleration, illusionary goal progress, and customer retention. J Market Res 2018 Oct 10;43(1):39-58. [CrossRef]
  37. Nunes J, Drèze X. The endowed progress effect: how artificial advancement increases effort. J Consum Res 2006 Mar;32(4):504-512. [CrossRef]
  38. Jenni K, Loewenstein G. Explaining the identifiable victim effect. J Risk Uncertain 1997;14:235-257. [CrossRef]
  39. Schelling TC. The life you save may be your own. In: Chase Jr SB, editor. Problems in public expenditure analysis. Washington, DC: The Brookings Institute; 1968:162.
  40. Small DA, Loewenstein G, Slovic P. Sympathy and callousness: the impact of deliberative thought on donations to identifiable and statistical victims. Org Behav Hum Decis Process 2007 Mar;102(2):143-153. [CrossRef]
  41. Small DA, Loewenstein G. Helping a victim or helping the victim: Altruism and identifiability. J Risk Uncertain 2003;26:5-16. [CrossRef]
  42. Cialdini R. The science of persuasion. Sci Am 2001 Feb;284(2):76-81 [FREE Full text] [CrossRef]
  43. Axelrod R. The evolution of cooperation. New York, NY: Basic Books; 1984.
  44. Cialdini RB, Vincent JE, Lewis SK, Catalan J, Wheeler D, Darby BL. Reciprocal concessions procedure for inducing compliance: the door-in-the-face technique. J Person Soc Psychol 1975 Feb;31(2):206-215. [CrossRef]
  45. Doogan C, Buntine W, Linger H, Brunt S. Public perceptions and attitudes toward COVID-19 nonpharmaceutical interventions across six countries: a topic modeling analysis of Twitter data. J Med Internet Res 2020 Sep 03;22(9):e21419 [FREE Full text] [CrossRef] [Medline]
  46. Seale H, Heywood AE, Leask J, Sheel M, Thomas S, Durrheim DN, et al. COVID-19 is rapidly changing: examining public perceptions and behaviors in response to this evolving pandemic. PLoS One 2020 Jun 23;15(6):e0235112 [FREE Full text] [CrossRef] [Medline]
  47. Underschultz JG, Barber P, Richard D, Hillier T. What drives resistance to public health measures in Canada's COVID-19 pandemic? A rapid assessment of knowledge, attitudes, and practices. SSRN Journal. 2020.   URL: [accessed 2022-07-21]
  48. Nivette A, Ribeaud D, Murray A, Steinhoff A, Bechtiger L, Hepp U, et al. Non-compliance with COVID-19-related public health measures among young adults in Switzerland: insights from a longitudinal cohort study. Soc Sci Med 2021 Jan;268:113370 [FREE Full text] [CrossRef] [Medline]
  49. Benham JL, Lang R, Kovacs Burns K, MacKean G, Léveillé T, McCormack B, et al. Attitudes, current behaviours and barriers to public health measures that reduce COVID-19 transmission: a qualitative study to inform public health messaging. PLoS One 2021;16(2):e0246941 [FREE Full text] [CrossRef] [Medline]
  50. Moran C, Campbell D, Campbell T, Roach P, Bourassa L, Collins Z, et al. Predictors of attitudes and adherence to COVID-19 public health guidelines in Western countries: a rapid review of the emerging literature. J Public Health 2021 Dec 10;43(4):739-753 [FREE Full text] [CrossRef] [Medline]
  51. Van Bavel JJ, Cichocka A, Capraro V, Sjåstad H, Nezlek JB, Pavlović T, et al. National identity predicts public health support during a global pandemic. Nat Commun 2022 Jan 26;13(1):517. [CrossRef] [Medline]
  52. Gollwitzer A, Martel C, Brady W, Pärnamets P, Freedman IG, Knowles ED, et al. Partisan differences in physical distancing are linked to health outcomes during the COVID-19 pandemic. Nat Hum Behav 2020 Nov;4(11):1186-1197. [CrossRef] [Medline]
  53. Lang R, Atabati O, Oxoby RJ, Mourali M, Shaffer B, Sheikh H, et al. Characterization of non-adopters of COVID-19 non-pharmaceutical interventions through a national cross-sectional survey to assess attitudes and behaviours. Sci Rep 2021 Nov 05;11(1):21751. [CrossRef] [Medline]
  54. R: A language and environment for statistical computing Computer Software. Version 4.0. R Foundation for Statistical Computing.   URL: [accessed 2022-07-21]
  55. Johnson PO, Neyman J. Tests of certain linear hypotheses and their application to some educational problems. Stat Res Mem 1936;1:93.
  56. Gollwitzer A, McLoughlin K, Martel C, Marshall J, Höhs JM, Bargh JA. Linking self-reported social distancing to real-world behavior during the COVID-19 pandemic. Soc Psychol Person Sci 2021 Jun 23;13(2):656-668. [CrossRef] [Medline]

OLS: ordinary least squares
T1: pretest time point
T2: posttest time point

Edited by A Kushniruk; submitted 22.07.22; peer-reviewed by V Capraro, E Shattuck; comments to author 04.10.22; revised version received 16.11.22; accepted 11.12.22; published 13.02.23


©Mehdi Mourali, Jamie L Benham, Raynell Lang, Madison M Fullerton, Jean-Christophe Boucher, Kirsten Cornelson, Robert J Oxoby, Cora Constantinescu, Theresa Tang, Deborah A Marshall, Jia Hu. Originally published in JMIR Human Factors (, 13.02.2023.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Human Factors, is properly cited. The complete bibliographic information, a link to the original publication on, as well as this copyright and license information must be included.