key: cord-0973654-kzgw7qx3 authors: Kao, Kristen; Lust, Ellen; Dulani, Boniface; Ferree, Karen; Harris, Adam; Metheney, Erica title: The ABCs of Covid-19 Prevention in Malawi: Authority, Benefits and, Costs of Compliance date: 2020-08-28 journal: World Dev DOI: 10.1016/j.worlddev.2020.105167 sha: b568fbdcc4fa376cc2fcf1a3de992f5efd61a70c doc_id: 973654 cord_uid: kzgw7qx3 A wide array of authorities--from religious leaders to government ministers—call upon citizens to take preventative measures against Covid-19. Which authorities can most effectively gain public compliance, and which measures will the public take up? Moreover, do people comply with authorities out of respect for their legitimacy, due to their expertise, or for fear of sanctioning? Answers to these questions are important for development practitioners, who need to understand how different partnerships might affect health behavior, and for scholars interested in understanding authority, legitimacy, and compliance. We explore these questions using a conjoint experiment embedded in a telephone survey of nearly 4,000 Malawians. Individuals in our sample are more likely to say that they will comply with precautionary measures when the costs are low and expected benefits are high. Respondents view both TAs and hospital heads as legitimately issuing directives and having the ability to monitor and sanction non-compliance, but appear to comply more with hospital heads and do so out of respect for their expertise. These results emphasize how who issues directives affects whether individuals comply and provides insights as to why they do so. The findings also reflect individuals’ cost-benefit calculations when considering precautionary measures, highlighting the importance of steps that can reduce costs (e.g., food security or income measures) or accurately reflect risks (e.g., information signaling the prevalence of Covid-19). The study not only helps to address the Coronavirus crisis but also has important implications for broader questions of authority and compliance. The failed attempt to institute a lockdown in Malawi and subsequent retooling of the Covid-19 taskforce raises important questions about the ability of authorities to enact public health measures in times of a pandemic. Which authorities can most effectively gain public compliance with difficult measures? Do people comply with authorities out of respect for their expertise, or fear of sanctioning? Does compliance depend on the type and associated cost of the measure, or the prevalence of the illness and therefore risk of catching it in the community? We draw on a survey experiment embedded in a telephone survey implemented in Malawi in May 2020 to explore these questions. We focus on district-level officials and find that Malawians are more likely to report compliance with directives from the head of the district hospital than with religious authorities. Our data suggests that compliance with the hospital head is out of respect for expertise. Malawians in our sample are more likely to report that the head of the district hospital and traditional authorities (TAs) i have the right to ask for compliance and would monitor, as well as quite possibly sanction, compliance compared to religious authorities. We also find that individuals are more likely to state that they would observe less costly measures, such as frequent handwashing. However, if the benefits of engaging in a costly action increase, as they do when many in the local area have symptoms, then our respondents report greater willingness to avoid large gatherings and even to stay at home. Finally, with respect to engagement in more costly actions, our results suggest that authorities who are seen as having the most expertise in the area of health gain the most compliance. In short, Malawians value appropriate expertise, prefer less costly measures, and are more likely to engage in costly ones when a situation appears dire. These findings not only lend policy-relevant insights into the rationale driving Malawians' uptake of precautionary measures, but they also contribute to a literature aimed at understanding pandemics and public health responses in the Global South. They highlight how the drivers of compliance with public health directives may vary according to different authorities, with some based upon respect for expertise. We also demonstrate how the widely cited Health Belief Model (Rosenstock, 1966) , developed in the West, travels to the Global South. A range of authorities can be incorporated in the fight against Covid-19. In Malawi, an increasingly broad set of actors is engaged in formulating policy responses and communicating directives, including local-and national-level actors, non-state and government authorities. Initially, President Mutharika appointed a Special Cabinet Committee on Covid-19, composed entirely of government ministers, to oversee the government response. However, as the number of cases grew and the cabinet committee faced growing criticism over its handling of Covid-19, Mutharika dissolved it and selected a more inclusive 21-member Presidential Task Force. Membership of the Taskforce, co-chaired by a public health expert from the Malawi College of Medicine, included the nine cabinet ministers from the Special Cabinet Committee, alongside other key stakeholders -the influential Christian Health Association of Malawi (CHAM) and Chiefs Council among them. ii These authorities may differ in their ability to mobilize citizens' compliance with preventative measures. Research suggests that compliance is higher when people view authorities as legitimate (e.g. Sunshine & Tyler, 2003) , which is often measured in terms of perceptions of it being right and proper for an authority to do something (Tyler, 2006) or in terms of generalized trust in them (Tyler & Jackson, 2013) , although most of the health literature focuses on the latter measure. Individuals in the Democratic Republic of Congo were more likely to take preventative measures when they trusted local officials (Vinck et al., 2019) , a result echoed in a survey in Liberia (Blair et al., 2017) . Arriola and Grossman (2020) , studying HIV/AIDS testing, find that Guineans listen to the president if they share his ethnicity and argue that this, too, is driven by trust. Citizens with higher institutional trust were more likely to take preventative measures against Covid-19 in China (Wong et al., 2020) . Other findings point to leaders' ability to reward compliance, or sanction dissent. Studying the Ebola crisis in Sierra Leone, Van der Windt and Voors (2020) find that chiefs facing less competition for their positions were associated with lower death rates, leading them to conclude that "strong leaders" have both greater desire and higher capacity for implementing policies that counter the spread of the disease. This resonates with arguments that restrictive measures, often associated with authoritarian regimes, may best contain the Covid-19 pandemic, and thus, that Covid-19 may foster authoritarianism (Wang, 2020) . These insights are instructive, but questions remain. Which authorities garner the most trust? For instance, individuals trust local authorities more than national leaders in DR Congo (Vinck et al., 2019) , co-ethnic more than non-co-ethnic leaders in Guinea (Arriola & Grossman, 2020) , and those whom they know personally in Canada (DiGiovanni et al., 2004, p. 270) . How do people gauge legitimacy: is it based on specific expertise (e.g., medical knowledge) or a more encompassing role in society (e.g., a TA)? Or, is compliance driven more by carrots and sticks than conviction? And if so, which authorities are most likely to be effective? Answering these questions provides important insights that can underpin public health initiatives in Malawi and, for practitioners and scholars of other countries, turn attention to the importance of basing public health initiatives on a more nuanced understanding of the drivers of compliance with public health measures. Of course, compliance may be more difficult to achieve when individuals view the actions prescribed as more costly or less beneficial. The Health Belief Model (HBM), first described by Rosenstock (1966 Rosenstock ( , 2005 , predicts that people are more likely to take up appropriate health behavior if they believe they are at risk, recognize the severity of the health problem, feel that the behavior will reduce the likelihood of negative health outcomes, and do not face high barriers to adopting the measures. iii This model is consistent with evidence that cost-and-benefit assessments influence Australians' preparedness to comply with measures combatting an influenza pandemic (Barr et al., 2008) , Canadians' compliance with quarantines (Cava et al., 2005) , and Nigerians' willingness to vaccinate (Onyeneho et al., 2015) . Uptake is particularly likely if there are cues to action and, as a meta-study of the HBM (Carpenter, 2010) found, when actions are aimed at prevention, as in the measures we study here. We explore questions about compliance with preventative measures through a singleprofile conjoint experiment. The experiment was embedded within a broader phone survey on the Covid-19 pandemic conducted in May 2020 in Malawi. Our sampling frame was derived from telephone numbers collected from respondents to surveys the team conducted in Malawi in 2019 (N~10,000) (Author et al., 2020) and 2016 (N~8,000) (Author et al., 2016) . The final sample included 4,641 respondents. (See Appendix for details.) The experiment presents each respondent with a hypothetical scenario that describes the extent of the pandemic and guidance to combat it from various authorities. Treatments were aimed at assessing the extent to which the degree of risk, the cost of action, and the nature of the authority affect compliance. (See Table 1 .) The extent to which others in the area were sick with Covid-19 proxies the degree of risk, and therefore benefit of compliance, with "many people" reflecting high risk and "no one" low risk. The actions represented different costs of compliance, with handwashing being the least costly action, not gathering in groups of 50 or more the second least costly, and staying at home except for essentials the costliest measure. The nature of authority included three types, all chosen to be roughly at the district level and thus about equidistant from the respondents: the head of the respondent's district hospital, the respondent's TA, and the respondent's religious leader. These authorities differ with regard to expertise: the district hospital head has relevant medical expertise, the religious authority has expertise in spiritual matters (potentially relevant regarding large, church gatherings), and the TA has no medical expertise but is seen as an important traditional leader, concerned with community welfare. They also differ with regard to their ability to sanction respondents: the head of the district hospital has low ability to monitor and sanction, while the TA has a high ability to monitor and sanction, drawing on the network of village heads for monitoring and enforcement. We anticipate the religious leader's ability to monitor and sanction is located between that of the hospital head and TA. iv The experimental prompt read as follows: "If (many people in your area are/a few people in your area are/no one in your area is) sick with Covid-19 and (the head of your district hospital/your Traditional Authority/your {religious leader}) v asked everyone to (stay at home except for essential needs/not gather in groups of more than 50 people including religious services, weddings, and funerals/frequently wash their hands with soap and water)." We investigate a series of hypotheses on authority, risks, and costs and benefits. All hypotheses were pre-registered, with a few exceptions that we note below. (See details on Open Science Framework pre-registration in the appendix.) We test three bases of compliance with authorities: legitimacy, sanctioning, and expertise. H1a. Legitimacy drives compliance. If legitimacy drives compliance, on average, we expect the authority to gain higher scores on the question of whether "it is right and proper that this authority asks for compliance with his/her directives." We did not have a pre-registered hypothesis as to which authority would gain the most compliance, on average, but we are able to use the experiment to see which ones do. H1b. Fear of sanctioning drives compliance. If fear drives compliance, on average, people will say they will comply with, and expect others in their community will comply with, requests from their TA (B2), who has a greater ability to monitor citizens via their appointed village heads and has a more direct impact on respondents' daily lives. We verify this using the outcome question on monitoring, expecting that individuals are more likely to think the TA knows whether or not they comply. H1c. Expertise drives compliance. If expertise drives compliance, we hypothesize that, on average, people will express greater willingness to comply with the requests of their district hospital head (B1) because s/he has more experience with, and knowledge of, health issues. We verify this using the outcome question on whether the respondents think they will get Covid-19, expecting this to reflect the quality of advice. We expect respondents will think compliance is more likely to be beneficial if it is in response to the district hospital head's advice. People comply when more people are sick, as risks of infection then appear higher and there are increased benefits to compliance. As people become more scared, they are more likely to comply with advice from officials on preventing the spread of the virus. Respondents are therefore more likely to state that they will comply with any request when there are many people (A1) or a few people (A2) who are sick compared to no one (A3). People are less likely to comply with more costly actions. As stated above, we assume that the cost of actions increases from handwashing, to avoiding large gatherings, to staying at home. We predict respondents are significantly more likely to state that they would more frequently wash their hands (C1) with soap and water than not gather in groups of more than 50 people (C2) or stay at home except for essentials (C3). The higher the prevalence of the virus, and therefore greater the benefits of compliance, the more likely people will express willingness to comply with costly actions. We test this with an interaction between the prevalence of the virus (Treatment A) and action (Treatment C). We predicted that authorities' ability to gain compliance would vary across actions because they would have more legitimacy in making some recommendations than others. We hypothesized that religious leaders would be seen as having the most legitimacy in asking for avoidance gatherings of 50 or more people on the grounds that this action affects church services. Further, TAs are likely to have the most legitimacy in asking people to stay home. We test this with an interaction between the prevalence of the virus (Treatment B) and type of request (Treatment C). We ran a standard causal conjoint analysis (Hainmueller et al., 2014) , Note. Robust standard errors in parentheses. ***p < 0.001 **p < 0.01 *p < 0.05 ^ p <0.1 Authority. Legitimacy. Compared to religious leaders, TAs and district hospital heads are about 6.5 percentage points (pp) (p <0.001) more likely to be seen as having the right to issue directives, on average (H1a). Sanctioning. On average, TAs and district hospital heads are seen as being 5 pp (p <0.01) and 6 pp (p <0.001) more likely to monitor (and have the ability to sanction) compliance than religious leaders, on average (H1b). Expertise. Looking at compliance, we see that district hospital heads gain 4 pp (p <0.01) more compliance than religious leaders (H1a), on average, while there is no evidence that TAs have any such effect. Moreover, in models with the TA as the baseline comparison authority, we find that district hospital heads are still significantly more likely to gain compliance. In postexperiment questions we find that 46% of the sample agreed that only the head of district hospital understands the Covid-19 virus best, compared to just 2% for their TA and 8% for their religious leader. ix When asked about trust, 78% of the sample reported a lot of trust in religious leaders compared to 60% in TAs and 64% in heads of district hospitals. This lends observational support that expertise drives compliance (H1c), more than generalized trust. The results suggest that people are on average 3 pp (p <0.05) more likely to comply when many people are sick in their area. We further find no evidence that people are more likely to comply when only a few people are sick. Thus, we find some support for the hypothesis that perceived risk increases compliance (H2). Costs. We find evidence that the costs associated with preventative actions affect compliance. Compared to a directive to wash hands frequently, the least costly action in our set, not gathering in groups begets 8 pp (p <0.001) less compliance and stay at home orders lose 19 pp (p <0.001) of compliance on average (H3). We estimate the ACIEs between the treatments of type of action and prevalence of the disease on compliance, finding evidence that people are more likely to comply with more costly actions when prevalence increases as shown in Table 3 . Compared to the least costly action, handwashing, and no one in the area diagnosed with Covid-19 as baselines, we find an increase in reported compliance with staying at home except for essentials (by 5 pp, p <0.10) when just a few people in the respondent's area are sick; moreover, this condition is associated with an increase in the right of an authority to ask for this action (by 6 pp, p <0.05). When the risks become even higher, with many people in the area becoming sick, this costly action gains significantly more compliance by 6 pp (p <0.05). As the threat of contracting the virus increases, respondents are more willing to comply with more costly actions to combat its spread. (H4). Note. Robust standard errors in parentheses. ***p < 0.001 **p < 0.01 *p < 0.05 ^ p <0.1 Considering the ACIEs between types of authority and action, we find mixed results concerning our pre-registered hypotheses (H5). As shown in Table 4 , using the religious authority and frequent hand washing as the baseline comparisons, it is significantly more right and proper for heads of district hospitals and TAs to ask people to not gather in groups (by 5 pp, p <0.05, and 8 pp, p <0.001 respectively) and to ask people to engage in the most costly action of staying at home (by 6 pp, p<0.05, and 9 pp, p <0.001 respectively). Yet, legitimacy does not seem to be all that matters to our respondents as only the head of the district hospital gains significantly more compliance for asking people not to gather in large groups, in contrast to our expectations. Note. Robust standard errors in parentheses. ***p < 0.001 **p < 0.01 *p < 0.05 ^ p <0.1 We find that perceived cost of directives has the greatest impact on compliance. Type of authority and perceived benefits of directives were also found to impact compliance, but to a lesser degree. Malawians in our sample view both TAs and hospital heads as legitimate in issuing directives and being about equally likely to monitor them, but our findings suggest that hospital heads are seen as having appropriate expertise. Notably, citizens are more likely to state they will comply with heads of hospitals than religious leaders or TAs. Although experimental design does not require the inclusion of controls for causal inference, we re-ran all of our AMCEs and ACIEs with variables to capture differences across respondent genders, ages, levels of education, and regions of residence. 1 The results show that at times, ages and regions of 1 We do not have a measure for income in this dataset. resident are significant. On average across all conditions, those who are 35-54 years of age and those who are older than 50, as compared to those who are aged 18-30, are more likely to believe that the authority has the right to ask for compliance. We also find that in accordance with other research on sub-national variations in Malawian politics (see e.g., Dulani & Dionne 2014), the Southern region in particular is almost always significantly different from the Central and Northern regions in its inhabitants' responses to our outcomes averaged across all treatment arms. Such a finding suggests that another study could be conducted to explore these variations. We examine who Malawians fear sanctioning from as it pertains to contributing to community initiatives. This is based on the Actions and Sanctions sections in a 2019 survey, conducted in Malawi (Author et al., 2019) . In particular we analyze the answers to questions regarding education, health, electricity and water and sanitation, starting with "Would you say that you contributed, at least partly, because…" • others will think poorly of you or your household if you don't contribute? If yes, whom? • you or your household will have to pay fines, lose property or suffer other material loss if you don't contribute? If yes, to/from whom? • you will be physically punished if you don't contribute? If yes, by whom? Since the same battery of questions was used for all initiatives, we will begin by pooling the results of all initiatives. Then we will investigate each initiative separately. Here we combine the results for the three types of sanctions (think poorly, cause material loss, cause physical harm) across the 4 topics (education, health, electricity, water and sanitation). See Table B1 . below. Table B1 . Who will think poorly of you? These experiments were embedded in a phone survey conducted in May 2020 in Malawi. Respondents were drawn from three pools: 1) We recontacted participants from the Malawi 2019 survey (details below) who had provided phone numbers at the conclusion of that survey and expressed willingness to participate in future surveys. We obtained 5,100 phone numbers through this process; 2) In some instances, we could not locate the original respondent through the phone number(s) provided, but we found a new participant willing to take the survey and administered it to him/her; 3) As the Malawi 2019 survey did not sample from the south of Malawi, we drew in additional participants by re-visiting villages from the Malawi 2016 survey (details below) and collecting phone numbers for the phone survey. The sample included 5100 phone numbers that had been collected from participants in the 2019. (Regarding the sampling strategy implemented in the 2019 survey, see below.) At the end of that survey, in preparation for a panel study, we had asked individuals if they would be willing to participate in a follow-up survey. We created a dataset that included the individual's name, telephone numbers, how long the individual had lived in the area, gender, age, and education. These questions were used to verify whether the individual answering the phone was the same respondent from 2019. Where the respondent existed but was not available, enumerators set a call-back time and recontacted the individual. Where the respondent was not available but the individual was over 18 years of age, the individual was asked if s/he wanted to participate in the study. Where the individual was under 18 years of age and the initial respondent was not available, the enumerator asked if an adult was available. That adult was then given the chance to participate in the survey. Replacement individuals were asked at the end of the survey if they are willing to participate in future studies. The phone numbers collected in the 2019 survey included respondents only in districts from the north and central regions; thus, to include the south, we sent teams to the southern regions and to two southern central region districts that had not been included in the 2019 . They were given and instructed to wear masks, use hand sanitizers, and maintain social distancing measures and were sent to the same villages that were included in the 2016 survey. For each village, they were given the lists of the first names of the adults who were in the household in 2016 and their ages (drawn from the kish grid), and the name of the original respondent chosen. They met with the village head who then helped them to contact and hire a person from the village. This person went to village houses to ask previous respondents if they would be willing to be contacted. The telephone numbers were collected from those who were willing. Where an individual was not willing or available, another adult in the household was asked to participate and, if s/he agreed, demographic information and the phone number was collected. If no one existed in the original household (e.g,, the family had moved or passed away) or if no one agreed to be contacted, the village contact was asked to find another household in the village willing to be contacted. Telephone numbers and demographics were entered into a database for use in the survey. The Samples were stratified. Border regions were divided into strata that were 0-50 km from the border and 50-100 km from the border, and each of these areas was divided into five subareas. Urban areas were divided into two concentric circles: 0-25 km from the urban center and 25-50 km from the urban center, and each was divided into four areas. The goal was to ensure that the respondents were distributed across the region and to include more and less urban and border areas. We aimed to divide the samples evenly across these regions and strata. Satellite imagery data was employed for selecting sampling units. To do so, we divided the regions/bins into 1 square kilometer areas, and selected these areas using a randomized, probability proportionate to size (PPS) method based on WorldPop estimates of population density. We then divided chosen areas into hectares. The hectares were randomly numbered, and enumerators were asked to begin interviewing in the 1 square kilometer areas in the hectares, moving from those with the lowest to highest numbers. They were asked to complete not more than 5 interviews in the hectare before moving onto the next one, and to complete 30 interviews in each square kilometer. The aim of this strategy was to ensure that enumerators spread out across the 1 square kilometer unit. Enumerators were instructed to enter sampling units using tablets to track their locations and confirm they were in the correct area. They were asked to go to the center of each hectare and then to move outward, in separate directions to additional houses. Within each household, one participant was randomly selected using the Kish method. Survey weights were designed to take into account sampling and to correct for imbalances between the sample and census demographics for the area. The survey was conducted in Malawi during March and April 2016. We implemented the survey using tablet computers. This survey seeks to measure and better understand governance and service delivery at the local level. Importantly, this is a highly clustered survey, which facilitates measurement and inference at the local (in this case, village) level. The survey covers the following topics: political participation, social norms and institutions, education, health, security, welfare, corruption, land, and dispute resolution. Table 1 for a list of the districts and TAs included in the sample and Table 2 for a list of the villages. While the sampling procedures were planned as presented, of course in practice this was not always the case. In total the research team had to draw 11 replacement EAs. One replacement EA was drawn because enumerators were chased out of a village and forced to withdraw from the EA. In the remaining 10 cases, EAs were not accessible (e.g. in one instance our team was unable to reach the designated EA because a bridge had washed away during heavy rains). In these instances, backup enumeration areas were randomly selected within the same EAs (excluding already selected and inaccessible zone) and were used as replacements. In total, only 11 of the 99 sampled EAs are replacement EAs. In addition, given that multiple enumerators conducted surveys in the same village, the target number of 22 respondents per village (neighborhood in urban areas) was not always reached precisely. In some instances, more were surveyed and in others slightly fewer than 22 households were surveyed. In addition, the boundaries between villages and neighborhoods were not always clear, which also caused our teams to deviate from the target of 22 per village/neighborhood. In times of crisis, citizens rely on their leaders for guidance on what to do. These elites have the power to influence public opinion and shape behaviors that could be either life-saving or lifethreatening. But a given citizen may have numerous leader s/he turns in times of need, who may be seen as carrying more or less weight in shaping opinion and reactions across varying recommendations to avoid catastrophe. These outcomes may also be shaped by citizens' levels of threat concerning of contracting the disease and/or the amount of effort required to comply with instructions to avoid it. In the context of the Covid-19 pandemic, this raises important questions requiring research. Key questions this experiment seeks to answer: 1. What authorities are best positioned to help manage a crisis in terms of their ability to positively influence citizens' opinions and behaviors? The head of your district hospital None Q78. Since Covid 19 has become a concern, are you more likely, the same, or less likely to: Visit friends and family in their homes Attend religious services Wash your hands Go to work/conduct business We develop hypotheses around individuals' perception of risk, the costliness of preventative measures and the nature of leadership. We pay particular attention to the nature of authorities, testing the extent to which different authorities draw on different sources of power to gain compliance. We test compliance using Q73 and Q74, expecting that any differences in outcomes are driven only by social desirability bias. For simplification, we refer to 'compliance' in hypotheses below. We expect that risk drives compliance with leaders surrounding health precautions. As people become more fearful, they are more likely to comply with health advisories from all leaders. 1A. People are likely to feel that there is greater risk the more people there are in the area affected by Covid. Thus, compliance is more likely when there are a) many people (A1) who are sick vs. no one (A3); and b) few people who are sick (A2) vs. no one (A3). 1B. We will also run observational analyses of people we expect to be more worried about contracting the virus and willingness to comply, using Q84 (worried about contracting covid), Q106 on health problems in the household, Q31, those having elderly living with them at home, and 2019/2016 Chronic illness or illness in last year response hlth_q5, age (demo_q2). People are more likely to comply with advisories when they view the advisories as less costly. 2A. We expect that compliance is less likely when actions are more costly/inconvenient. We assume that washing hands with soap and water is a less costly action compared to not gathering in groups of more than 50 people, and that staying at home except for essentials is most costly. On average, compliance is significantly more likely for handwashing (C1) than for gathering in groups of more than 50 people (C2) or staying at home except for essentials (C3). Observationally we will look at responses from Q78 on activities people are more or less likely to engage with since the Corona 19 pandemic started to understand if some actions are more costly than others, but these will be simple cross-tabs. 2B. The cost of compliance is likely to be overcome in more risky situations. Compliance will be higher when the risk/threat level is high, namely when many people are sick (A1) compared to when no one is sick (A3). Compliance with more costly actions, namely staying at home except for essentials (C3), will be significantly higher when the risk/threat level is high, namely when many people are sick (A1) compared to when no one is sick (A3). This is an interaction between experimental arms. We consider the extent to which different authorities drive compliance. We are particularly interested in whether and under what conditions people are more likely to comply on the basis of expertise, sanctioning, or legitimacy. 3A1. If people worry about sanctioning for non-compliance drives compliance, we expect that on average, people will comply with the requests of authorities that they believe can monitor them. We expect that they are more likely to comply with requests of their Traditional Authority and Religious Leader given their higher capacity to monitor them and others in the community, using village heads or other congregationalists, respectively, to assist in monitoring, vs. the head of the district hospital. That is, compliance will be higher with requests from their Traditional Authority (B2) or their district religious leader (B3) vs. their head of their district hospital (B1). We will verify these assumptions using Q77 on monitoring, expecting that respondents are more likely to think that the Traditional Authority (B2) and religious leader (B3) can monitor their compliance than the head of their district hospital (B1). 3A2. Given hypothesis 3A1, more costly prevention measures will be followed when the individual expects the authority will monitor the action. Requests by Traditional Authorities (B2) to stay home except for essential needs (C1) will be significantly more likely to be complied with than this same request by religious leaders (B3) or by heads of district hospitals (B1). 3B1. If people care more about expertise, we expect that on average, compliance will be higher with requests of their district hospital head (B1) over the Traditional Authority (B2) or Religious Leader (B3). 3B2. We will observationally verify this mechanism using the follow-up question on who best understands the Novel Corona virus (q94). We will dichotomize this question by whether the respondent thought the authority seen in the experiment had the most knowledge about the virus, and interact this with the leader type to see if in fact expertise drives compliance with the head of district hospital (HTE). We expect that health directors will be seen as having the most expertise. 3B2. We will observationally verify the mechanism for this theory by interacting the randomized authority with a follow-up question (Q98) on whether the respondent has confidence in the public/government health system's ability to handle the Covid 19 crisis. (HTE) 3C1. Realm of competence is likely to play an important role in compliance. Requests by religious leaders (B3) to not gather in groups of more than 50 (C2) will be significantly more likely to be complied with than this same request by Traditional Authorities (B2) or heads of district hospitals B1). (HTE) 3C2. We will use q75 on whether the authority has the right to ask people to comply with his/her instructions to interrogate this. Religious leaders are expected to have the right to tell people not to attend church, which is likely to be a gathering of more than 50 people in many areas, whereas other leaders do not have this same right. When it comes to locking down and stopping social interactions, the traditional authority is most likely to have the right to go this far in telling people what to do. On average, religious leaders (B3) will be significantly more likely to be seen as having the right to ask for compliance when they ask people not to gather in groups of 50 (C2) than other authorities (B*). 3C3. People who are more religious (as measured by the previous Malawi 2019 survey: "How often do you attend church, mosque or other religious services or meetings?") are significantly more likely to comply with their religious leaders (B3) than other leaders (B*). (HTE) 3C4. On average, Traditional Authorities (B2) will be significantly more likely to be seen as having the right to ask for compliance when they ask people to stay home except for essential needs (C1) than other authorities (B*). Note that this is a different mechanism than proposed in 2c. Above in that it suggests TAs have the most legitimacy in this realm, beyond their capacity to monitor compliance. While we cannot definitively say it is one of these mechanisms or the other, we can examine both to see if one seems to have more of an effect than the other. Our outcome questions include: 1) willingness to comply with precautionary measures, 2) expectations that others in the neighborhood would comply, 3) perceptions of whether this authority has the right to request compliance, 4) belief that the behavior would lessen the likelihood of Institutional Review Board approval. At the end of that survey, we asked respondents if they would be willing to be contacted again. We only interviewed respondents who are over 18 years of age, and for whom we have informed consent. If the respondent we reached was not the same as in the original survey, we asked the respondent's age and declined to interview if they were under age. All data collected will be kept anonymous and stored in encrypted files. We will not distribute anything with GPS coordinates or names, and all data will be retained on encrypted University servers. We understand that there is always risk when handling confidential data, and we did all in our power to mitigate that risk by ensuring encrypted data storage and enforcing communication regulations. Additionally, all enumerators signed non-disclosure agreements, and were subjected to GDPR guidelines. We asked for new consent from all respondents. Part of this consent section clearly explained that participation is voluntary and that the respondents have the right to seek clarification of their rights or to withdraw participation consent at any time. Each question also allowed for a do not know/refuse to answer response to mitigate discomfort for the participants. The consent scripts were as follows, changing slightly depending on the respondent type reached: -The type of authority issuing directives, perceived benefits, and costs of actions affect citizens' compliance with precautionary measures. -Citizens are more likely to follow directives of authorities whom they view as having appropriate expertise. -Citizens prefer less costly measures, but they report willingness to comply with costly ones when health risks are higher. -Different authorities are important and distinct partners in gaining compliance with pandemic prevention initiatives. -Policymakers should emphasize public health risks, provide expert-backed advice, and take measures to make compliance less costly. i Traditional Authorities (TAs) are leaders within Malawi's traditional authority system, corresponding roughly to the district-level. They stand above village headman and group village headman and below chiefs and paramount chiefs, sit on the District Committee and serve as Chairpersons of Area Development Committees. For more on the relationship between the state and TAs, see Chinsinga (2006) and Power (2020) , and on TAs' ability to influence public opinion, see Muriaas et al. (2018) . ii Malawi Broadcasting Corporation Online, https://www.mbc.mw/news/entertainment/item/9387-presidential-task-force-on-covid-19- vi We included the question of others' compliance because we felt it would reduce potential social desirability bias. However, we found that people had seemingly little problem with admitting that they would not be willing to personally comply with some directives. We thus focus on individuals' compliance here and report results regarding others' expected compliance in the appendix. vii AMCE is the causal effect of an attribute, averaged over the joint distribution of the remaining attributes (Hainmueller at al., 2014) viii ACIE is the difference in the causal effect of an attribute, conditional on the value of another attribute (Hainmueller at al., 2014) . We also estimated Average Marginal Interaction Effects, which do not rely on a baseline attribute (Egami & Imai 2019) . We note here that our results are not robust to this analysis; yet, since our hypotheses are structured around the comparison of our treatments to baselines, we do not believe this detracts from our findings. ix Note that twenty-seven percent of respondents in the sample believe that all of the three authorities understand the Covid-19 virus and 4% believe that none of them does. Ethnic marginalization and (non) compliance in public health emergencies Pandemic influenza in Australia: Using telephone surveys to measure perceptions of threat and willingness to comply Public health and public trust: Survey evidence from the Ebola Virus Disease epidemic in Liberia A meta-analysis of the effectiveness of health belief model variables in predicting behavior Perceived threat in compliance and adherence research Risk perception and compliance with quarantine during the SARS outbreak The interface between tradition and modernity: The struggle for political space at the local level in Malawi Factors influencing compliance with quarantine in Toronto during the 2003 SARS outbreak Causal interaction in factorial experiments: Application to conjoint analysis Causal inference in conjoint analysis: Understanding multidimensional choices via stated preference experiments Malawi Survey Malawi Survey Presidential task force on COVID-19: Mutharika hires 21-member team. Malawi Broadcasting Corporation Online Why the gender of traditional authorities matters: Intersectionality and women's rights advocacy in Malawi Compliance with regimens of existing vaccines in Orumba North Local Government Area of Anambra State Chieftaincy in Malawi: Reinvention, re-emergence or resilience? A Kasungu case study Why people use health services Why people use health services The role of procedural justice and legitimacy in shaping public support for policing Psychological perspectives on legitimacy and legitimation Popular legitimacy and the exercise of legal authority: Motivating compliance, cooperation, and engagement Traditional leaders and the 2014-2015 Ebola epidemic Institutional trust and misinformation in the response to the 2018-19 Ebola outbreak in North Kivu, DR Congo: A population-based survey Authoritarianism in the time of COVID The authors would like to thank Witness Alfonso, Cecilia Ahsan Jansson, Rose Shaber-Twedt, Tove Wikehult, and researchers at the Institute of Public Opinion and Research. We are grateful to the