The World Bank Economic Review, 38(3), 2024, 514–534 https://doi.org10.1093/wber/lhae001 Article Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Religious Leaders’ Compliance with State Authority: Experimental Evidence from COVID-19 in Pakistan Kate Vyborny Abstract A randomized controlled trial in Pakistan tests whether one-on-one engagement with community religious leaders can encourage them to instruct congregants to follow government regulations. Treated religious leaders are 25 percent more likely to comply with government requirements to tell congregants they should wear a mask to prevent COVID transmission when attending prayers. Treatment effects do not depend on the religious content of the message. Effects are driven by respondents who already understand the mechanics of COVID transmission at baseline, suggesting the treatment does not work by correcting basic knowledge about the disease, but rather through a mechanism of persuasion. JEL classification: I18, Z12 Keywords: State capacity, state authority, religious institutions, imams, RCT, Pakistan Weak state capacity is a central challenge for economic development (Wade 1990; Besley and Persson 2009; Acemoglu, Ticchi, and Vindigni 2011). Among other challenges, states in developing countries often have limited ability to enforce laws and regulations to collect resources, maintain public safety, or protect public health. Regulations promulgated by the state may be challenged by informal or non-state leaders, such as ethnic, tribal, or religious leaders. Thus, states must often engage with non-state leaders, or contend with their opposition. Yet there is surprisingly limited evidence addressing how non-state leaders decide whether to comply with or oppose state regulation. This study investigates whether and how local informal leaders can be persuaded to comply, and en- courage their followers to comply, with state institutions. Specifically, it examines local religious leaders’ compliance with government mandates in the context of the COVID-19 pandemic, a situation which has made painfully clear the limited ability of state capacity to directly enforce regulations such as masking. Kate Vyborny is Associate Director of the Devlab@Duke and Research Scientist in the Department of Economics, Duke University, Durham, NC, USA; her email address is katherine.vyborny@duke.edu. Feedback from Eric Edmonds and three anonymous reviewers substantially improved the paper. The author thanks Austin Davis, Eddy Malesky, Muhammad Meki, Erik Wibbels, and workshop participants at Duke, PACDEV, and the ADE at the Lahore School of Economics for useful feedback. Lala Rukh Khan, Syed Uzair Junaid, and Maniha Aftab provided outstanding research assistance. Ahmad Siddiqi, Arafat Mazhar, Rania Nasir, and Mahroz Haider provided helpful input to the development of the intervention scripts. The author gratefully acknowledges institutional support from the Center for Economic Research in Pakistan and funding support from the International Growth Centre (PAK-20078) and from the UK Foreign, Commonwealth & Development Office, awarded through Innovation for Poverty Action’s Peace & Recovery Program (MIT0019-X16). Duke’s IRB approved this study under protocol number 2020-0432. A supplementary online appendix is available with this article at The World Bank Economic Review website. C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com The World Bank Economic Review 515 A randomized controlled trial is used to test whether local religious leaders can be persuaded to follow a government mandate to ask followers to follow such measures when attending congregational prayer. The intervention tested is low cost and light touch: a one-on-one, interactive persuasive telephone call, focusing Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 on the participants’ importance as community leaders in guiding followers and protecting vulnerable members of the congregation. The outcome of interest is a change in religious leaders’ behavior in asking congregants to comply with government mandates for masking and other preventative measures while attending prayer. In separate “mystery-shopper” calls, a prospective congregant calls the leader to ask about service timings and re- quirements for masking. The religious leaders do not know that the caller is associated with the study. This approach has a substantial advantage over survey-based measures, in that behaviors may correlate only weakly to stated attitudes (Druckman 2021; Lau 2020), and the treatment itself could have induced differential social desirability bias in survey questions on beliefs or behaviors; recent evidence points to substantial over-reporting of COVID-19 prevention measures, including by religious leaders in the context of Pakistan (Grépin, Mueller, and Wu 2023; Pattan Development Organization 2020). The intervention leads to a 25 percent increase in the proportion of respondents who instruct congregants to comply with the government mandates when attending prayer. Variation in baseline knowledge about the basics of COVID transmission allows one to test whether the communication treatment takes effect through an information/education channel or a persuasion channel (Lau 2020). The results are stronger among those who answered baseline questions about COVID transmission correctly. This is not consistent with a pure information updating effect. Rather, this result suggests that the interactive conversation has an effect through a persuasion or commitment mechanism (Lau 2020; Druckman 2021; Dellavigna and Gentzkow 2010), connecting that knowledge to respondents’ sense of identity as leaders and motivation to protect the vulnerable in their community. Testing whether explicit religious framing moderates the effects on leaders’ decisions sheds further light on this mechanism. Religious framing might be particularly important in contexts in which religious institutions and the state are potentially competing sources of authority. However, the results show no differential impact of adding explicit religious content to the script. This study contributes to three main strands of literature. First, an extensive literature in economics and political science has explored influences on decisions made by leaders and officials, in particular bureaucrats and state employees (reviewed by Finan, Olken, and Pande (2015); Besley et al. (2022)) or elected officials (reviewed by Grose (2014)). Within this literature, this study contributes most closely to a nascent literature conducting information and communication experiments with bureaucrats or elected officials (Callen et al. 2020; Vivalt and Coville 2020; Banuri, Dercon, and Gauri 2019; Hjort et al. 2021; Nakajima 2021; Grundler and Potrafke 2022; Rogger and Somani 2018; Mehmood, Naseer, and Chen 2022; Lee 2022). However, these studies do not examine the decisions of non-state or informal leaders, and tend to focus on mechanisms that would not apply to such leaders. Studies of bureaucrats typically focus on extrinsic incentives applicable only to employees. Similarly, studies of elected officials focus on direct or indirect electoral incentives, which are inapplicable to informal leaders. Thus it is unclear to what extent findings from studies of state leaders might generalize to the behavior of non-state or informal leaders. A second literature explores the role of non-state institutions in supporting or hindering the capacity of the state. This literature focuses on understanding the sources of legitimacy and power of institutions, us- ing historical quasi-experiments (Dell, Lane, and Querubin 2018; Acemoglu and Robinson 2017; Banerjee and Iyer 2005) or using natural or field experiments to understand citizens’ decisions on compliance with state versus non-state authorities or trust in state services (Acemoglu et al. 2020; Balán et al. 2022; Khan et al. 2021; Blattman, Hartman, and Blair 2014; Sandefur and Siddiqi 2015; McCauley 2014; Wilke 516 Vyborny 2023; Cavgias et al. 2023; Blair et al. 2022; Martinez-Bravo and Stegmann 2021).1 This literature has highlighted that non-state leaders may either directly influence the political system or act as an alternative source of authority or legitimacy to that of the state, and are thus particularly important in influencing Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 compliance with state authority. However, these studies do not explore the determinants of decisions made by informal or non-state leaders. The third literature examines the behavior of local leaders in interacting with the state. These studies have focused on the role of such leaders as intermediaries in the electoral process (Kitschelt and Wilkinson 2007; Stokes, Dunning, and Brusco 2012; Nichter and Peress 2017; Nichter 2018; Baldwin 2013, 2014; De Kadt and Larreguy 2018; Acemoglu, Reed, and Robinson 2014), or the delivery of state or donor assistance (Basurto, Dupas, and Robinson 2020; Casey et al. 2012; Kilic, Whitney, and Winters 2015; Alatas et al. 2012, 2019; Conning and Kevane 2002; Mansuri and Rao 2004; Baldwin 2018; Casey et al. 2023; Voors et al. 2018). Quantitative researchers have not focused on the question of how informal leaders make decisions on supporting or undermining compliance with the state. In addition, this literature has focused on leaders such as landlords and village chiefs; there has been little work on decisions by religious leaders, despite a wealth of evidence on the influence of religious institutions and leaders on individuals’ economic, social, and political decisions.2 Two recent exceptions use quasi-experiments or survey experiments to study how religious leaders engage in electoral politics and rent seeking (Mehmood and Seror 2023; Smith 2016). This study thus contributes to the existing literature by rigorously investigating how informal leaders decide whether to support or undermine the state, an area that has received limited attention in quantita- tive research. It also innovates by applying experimental methods to study the decision-making of religious leaders; previously such methods have been used primarily with individual citizens and to a more limited extent with politicians. By examining how targeted communication can influence informal and religious leaders in making decisions that support or undermine state capacity, this study bridges a gap between the literatures on the influences on the decisions of leaders and the relationship between the state and informal institutions. A secondary contribution of this study relates to the literature investigating how communication may influence pro-social public-health behavior. Information campaigns often focus on inducing individuals to update their baseline beliefs about the private returns to health behaviors, even when the targeted be- haviors have large positive spillovers: convincing individuals to wash their hands (Bennett, Naqvi, and Schmidt 2018), use a bednet to prevent the spread of malaria (Rhee et al. 2005), have themselves or their children vaccinated (Nyhan et al. 2014; Alsan and Eichmeyer 2023), or avoid risky sexual behavior (de Walque 2007; Kerwin 2020; Dupas 2011; Duflo, Dupas, and Kremer 2015). However, approaches emphasizing private returns may face limitations in situations where the positive spillover of a health 1 Some of these studies examine interventions by the state or non-state actors, and examine citizen behavior in response (Cavgias et al. 2023; Sandefur and Siddiqi 2015; Balán et al. 2022; Martinez-Bravo and Stegmann 2021; Blattman, Hartman, and Blair 2014), while others use lab or survey experiments to elicit citizens’ preferences over state versus non- state institutions (Blair et al. 2022; McCauley 2014). This study is methodologically most closely related to Acemoglu et al. (2020) and Khan et al. (2021); in both these studies a research team member provides information to participants in a survey or lab-in-field experiment about state effectiveness, and examines shifts in survey responses or behavior in incentivized choices. However, this study differs from these studies in that it tests the role of persuasion as distinct from information. 2 Economists and political scientists have demonstrated this in a range of settings across developed and developing countries (Auriol et al. 2020; Bassi and Rasul 2017; Bryan, Choi, and Karlan 2020; Murphy, Nourani, and Lee 2020; Gruber 2005; Moreno-Medina 2021; Torgler 2006; Boyer et al. 2022; McClendon and Riedl 2019, 2015; Mcclendon and Riedl 2016; McClendon 2019; McClendon and Riedl 2021; Blair et al. 2021; Clingingsmith, Khwaja, and Kremer 2009; Rahman 2021; Sharma 2018; Sperber, McClendon, and Kaaba 2021; Chhibber and Sekhon 2014; Condra, Isaqzadeh, and Linardi 2019; Hsiung and Djupe 2019; Freedman 2020; Hong and Paik 2021; Ben-Nun Bloom, Arikan, and Courtemanche 2015; Rink 2018). The World Bank Economic Review 517 behavior is large relative to the private returns, such as mask use by young, healthy people in the case of COVID, or the take-up of childhood vaccinations. Appealing to pro-social motivations may have poten- tial in such situations, particularly given recent evidence on the importance of pro-social motivation in Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 incentivizing health sector workers (Ashraf, Bandiera, and Jack 2014; Deserranno 2018; Khan 2020) and the relevance of social signalling concerns in motivating vaccination take-up (Karing 2021). However, few studies have examined health information treatments that appeal to respondents’ pro-social motivations in changing their behavior, and several recent studies have found no detectable effects of such variations in information delivered to the general public (Guiteras et al. 2016; Banerjee et al. 2020; Khan et al. 2021; Dai et al. 2021). Abaluck et al. (2021) report the results of a package of interventions to reduce COVID transmission in Bangladesh. They find no effect of cross-randomized household-level variation in altruism-based or self-protection-based promotion.3 The message in the intervention in the present study has a strong pro-social focus: respondents are asked in their capacity as leaders to take safety measures in the mosque to protect the elderly and vulnerable in their congregations, and to influence others to do the same. The strong response to this treatment suggests the potential of pro-social messaging in information campaigns for public-health behaviors with large positive spillovers, and those targeted to community leaders, who may be positively selected for pro-social motivation, or feel that communities expect them to take greater responsibility for protecting members as part of their role. The remainder of the paper proceeds as follows. The first section describes the context. The second section details the experimental design. The third section presents results, and the final section concludes. 1. Setting Pakistan is a setting with a low degree of trust in secular authorities; in the 2012 World Values Survey, 60 percent of Pakistanis reported that they feel little or no confidence in the government (World Values Survey 2012). In national polls carried out during the COVID-19 pandemic, 50 percent agreed that COVID-19 is a foreign conspiracy (Gallup Pakistan 2020). Religious institutions, on the other hand, enjoy a high degree of trust. The vast majority of the popu- lation (96 percent) are Muslim (Pakistan Bureau of Statistics 2017). Religiosity is among the highest in the world, with 94 percent saying religion is “very important” in their lives (Pew Research Center 2018). About 80 percent of respondents in the World Values Survey agree that it is an essential characteristic of democracy for religious authorities to interpret the laws (World Values Survey 2012). Attending Friday prayers in congregation at the mosque is generally considered obligatory for men, and this involves close contact: the Sunnah (example of the Prophet) is to stand shoulder to shoulder during the prayer. Islam has no central religious authority; any cleric with a certain level of legal qualification (a mufti) can issue a fatwa, or Islamic legal opinion, and many clerics who are not officially muftis do so as well; thus, there may be many contradictory opinions issued by authoritative leaders on any given issue, and these are not considered binding. Thus, local religious leaders in Islam have substantial discretion to influence their congregants’ practices. In March 2020, with COVID-19 cases rising, the government announced a nationwide lockdown, including the suspension of congregational prayers. However, there was limited compliance with this rule. Nationally influential clerics announced opposition to the rule, and in some cases police attempting to enforce it clashed with worshippers outside mosques. Government and a group of influential clerics at the national level met for a series of negotiations and then announced a joint plan. Mosques would stay open but would follow 20 key rules to reduce the spread of COVID. The most clearly defined of these were as follows: (a) prayer mats should be removed and the floor should be washed with chlorinated water, (b) 3 Abaluck et al. (2021) also added a focused religious and community leader engagement protocol as part of their package of interventions, part way through the experimental rollout; however, they do not test the impact of engagement with religious leaders separate from other approaches, or investigate the responses of religious leaders. 518 Vyborny Table 1. Experimental Design N individuals who responded to Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Baseline Health Government Religious Mystery Baseline + Mystery questions information persuasion persuasion shoppers treatment call shopper Both Secular T Yes Yes Yes No Yes 283 213 213 Secular + religious T Yes Yes Yes Yes Yes 261 185 185 Control Yes No No No Yes 285 511 211 Super control No No No No Yes NA 631 NA Total 829 1,540 609 Note: In the treatment arms, respondents were only attempted for the mystery shopper if they responded to the baseline. For the control and super-control arms, all respondents were attempted for the mystery shopper. Thus, the main analysis compares treatment and control for the sample in column 8 (Sample 1): those who responded to both the baseline and mystery shopper (N = 211 + 213 + 185 = 609). Comparisons between control and super control, use column 7 (sample 2): those who responded to the mystery shopper (N = 511 + 631 = 1, 142). Figure S1.2 provides more details on the sample composition and reasons for exclusion at each stage of the experiment. people over 50 years of age and children should not be allowed to attend, (c) six feet of distance during congregational prayers, (d) people should perform ablution at home, and (e) congregants should wear face masks. However, implementation of these rules was limited, with NGOs reporting in May that 80 percent of mosques were not following these rules (Pattan Development Organization 2020); enforcement by state authorities was practically non-existent. After the first major religious holiday of the year, Eid ul Fitr in May 2020, cases climbed faster (fig. S1.1); many attributed this rise in part to religious and social gatherings on the holiday, includ- ing congregational prayers in mosques (Deutsche Welle 2020). In the period leading up to Eid ul Azha, the second major religious holiday of the year, at the end of July, policymakers were concerned that cases could spike again due to large-scale gatherings over the holiday. Self-reported weekly mosque attendance continued to climb after a low during lockdown, reaching 77 percent by the time of Eid ul Azha (Gallup Pakistan 2020). The randomized control trial (RCT) took place over a three-week period in July 2020, leading up to the Eid ul Azha holiday. The treatments and data collection reference this holiday. Comparing pilot survey data collected in April 2020 to the experiment baseline survey in July 2020 (table S1.1) reveals that by the time of the experiment, respondents were more likely to report asking congregants to wear masks, and less likely to report other steps, such as removing communal prayer mats or encouraging congregants to perform ablution at home. This likely reflects the shift in public-health messaging towards an increased focus on masking given updated knowledge about COVID transmission mechanisms. 2. Experimental Design The RCT sample is a random draw of religious leaders from community mosques across 19 districts of urban and rural Punjab from a government listing of individuals licensed to register marriages, provided by the Government of Punjab as part of a separate experiment (Field and Vyborny 2020). Approximately two-thirds of individuals in this sample frame are local religious leaders (who typically conduct a marriage and then register it); for this RCT, the randomization sample is restricted to this group. These licenses are issued once, with no official qualifications or training; there is no process for expiry or renewal of these licenses, and the respondents do not receive any government compensation. Table 1 gives an overview of the experimental design, and fig. S1.2 shows the steps and sample com- position in more detail. Respondents were randomized individually into one of four experimental con- ditions. Randomization was stratified by subdistrict, rural/urban location, and education level. Baseline phone calls were carried out in a randomized order within a fixed time frame available for field work The World Bank Economic Review 519 (fig. S1.2, box 1); thus about half of the religious leaders originally randomized into each treatment arm were actually attempted for the baseline and become part of the main sample. In the baseline phone call, the enumerator identifies himself as a researcher from the Center for Eco- Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 nomic Research in Pakistan (CERP). He emphasizes that he is not calling from the government, but rather as part of a research study. This framing was used both to avoid deception and also to help elicit truthful responses to the survey questions and avoid concerns that the call might lead to any punishment related to enforcement of COVID regulations. Enumerator debriefing revealed that most respondents seemed as- sured that the caller was not from the government; the handful who seemed unconvinced of this point declined to participate, apparently due to concerns that the call related to enforcement. CERP is well known in policy circles, but is not generally familiar to the public; as a result, respondents were not famil- iar with CERP, and CERP’s own reputation did not play a role. Thus the treatment represents a potentially neutral researcher or non-state actor engaging with the respondent, similar to Acemoglu et al. (2020) and Khan et al. (2021). The caller then confirms the identity of the respondent and his role as a religious leader. Non-response to this call was high, primarily because the government listing was several years old and had many wrong and outdated numbers. Crucially, there was no difference in protocol or information provided to respon- dents between treatment arms until the end of this call (fig. S1.2, box 2), so differential non-response by treatment arm is not possible in this stage of the design. Thus non-response at this stage is a form of selection into the RCT sample, rather than attrition. Notably, this selection into the study is not correlated with the respondents’ education or urban/rural location (table S1.2, column 1). For the religious leaders who are reached and agree to be surveyed, the enumerator carries out the baseline survey, collecting information on baseline knowledge about COVID and existing steps taken to prevent spread in the mosque. For the two treatment groups, the persuasion script follows immediately after the survey (fig. S1.2, box 2). Both treatments (secular persuasion; secular + religious persuasion) emphasize the importance of the respondent’s leadership role in the community and appeal to him in pro- tecting vulnerable community members from COVID. In addition, they both emphasize the key actionable points in the official protocols for mosques, and ask the respondent to follow them and to spread the word to his community through his sermons and mosque loudspeaker announcements. The script is framed as a persuasive appeal by a non-government organization, with no reference to enforcement measures. It is interactive, involving frequent elicitation of the respondent’s reactions and agreement, as well as asking him to commit to action. The interactive nature of the messaging is a key difference from mass-media messages about COVID, which were widely disseminated during this period. In addition to these elements, the secular + religious persuasion treatment arm includes an appeal to religious authority. This includes (a) the fact that the top religious leaders have endorsed the protocols for mosques, (b) hadith (sayings of the Prophet) about avoiding spread of plague, (c) international Sunni and Shia authorities’ pronouncements (fatwas) on the importance of complying with official authorities to prevent spread of COVID, (d) examples of other Muslim countries following strong measures to prevent COVID spread. These components are all designed to convey endorsement from a higher religious au- thority than the respondent. Because the outcome of interest is compliance with government regulations, the RCT design incorporates a combined arm adding religious framing to the script concerning these reg- ulations, rather than using a separate religious-only script. The scripts of the treatments are detailed in supplementary online appendix S2. The main followup data collection uses mystery shoppers to obtain a measure of the religious leaders’ instructions to members of the mosque congregation free of social desirability bias. This approach has been used extensively in recent development economics research to observe behavior of agents in contexts where self-reports are expected to be affected by social desirability or other forms of bias (Gaddis 2018; Banerjee et al. 2021; Anagol, Cole, and Sarkar 2017; Gine and Mazer 2016; Bertrand et al. 2007; Dizon- Ross, Dupas, and Robinson 2017; Hetzel et al. 2008; Mohanan et al. 2015; Currie, Lin, and Zhang 2011; 520 Vyborny Friedman, Woodman, and Chatterji 2015; Fitzpatrick 2021; Bate, Mooney, and Hess 2010; Bate, Jin, and Mathur 2011; Das et al. 2012; Field and Vyborny 2020; Broockman 2013, 2014; McClendon 2016; Kalla and Broockman 2016; Mendez and Grose 2018; Driscoll et al. 2018; Gaikwad and Nellis 2021). Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 A different enumerator from the original surveyor calls each respondent in the days before Eid, pos- ing as a member of the community saying he and his father want to confirm the timing for Eid prayer services at the mosque. Timings for services are usually set one to two days before Eid and differ for each mosque and each year; thus community members must enquire to confirm the time. These calls were credible to respondents; a number of respondents later called back to let the mystery-shopper callers know the confirmed time of prayers. Response to this call is not correlated with treatment (table S1.3). It is also uncorrelated with respondents’ education, although rural respondents are more likely to answer the mystery-shopper call (table S1.2, column 2). Mystery-shopper callers were blinded to the treatment status of the respondent. Enumerators also recorded whether the respondent seemed suspicious of the mystery-shopper call; approximately 10 percent of calls were tagged as suspicious. Treatment does not affect whether respondents seemed suspicious, nor are there heterogeneous effects on suspicion by base- line knowledge of COVID or congregation size (table S1.4). The scripts of the mystery-shopper calls are included in supplementary online appendix S2. Supplementary online appendix S3 discusses ethics aspects of the mystery-shopper call. After asking about the time of prayer services, the caller asks several questions about how he should prepare for attending mosque given COVID conditions, e.g. by wearing a mask, bringing his own prayer mat, or doing ablution at home. This is credible during the Eid prayers as it is a larger gathering than usual, and congregants who do not frequently attend the mosque are likely to attend. The primary outcome is a simple mean index of the following indicators: (a) whether the respondent advises wearing a mask, (b) says wearing a mask is required when the caller says he would prefer not to wear it, (c) tells the caller to bring a prayer mat (i.e. because the mosque mats would have been removed), (d) indicates the caller should do ablution at home, and (e) asks about the caller’s father’s age or indicates that the elderly father should not attend. Results tables also report effects on each of the five individual components of the index. For the primary analysis, enumerators make mystery-shopper calls with respondents who responded to the baseline survey in treatment and control arms (fig. S1.2, box 3, blue and green nodes); this is de- noted as Sample 1. Of the 819 respondents who were surveyed successfully in the treatment and control groups and called by mystery shoppers, 629 (75 percent) answered the mystery-shopper calls, of whom 609 were planning to hold Eid prayers at their mosque and were asked questions about prayers;4 this is not differential by treatment arm (table S1.3, column 1). To quantify the main treatment effect of inter- est, the impact of persuasion, Specification (1) compares mystery-shopper outcomes between treatment and control arms for these 609 respondents reached at baseline (Sample 1, shown in blue and green in fig. S1.2): Yi = β0 + β1 ANYTREATi + i . (1) To test whether religious messaging moderates the treatment effects, an alternative specification breaks down the treatment into its two arms:5 Yi = β0 + β1 SECULARTREATi + β2 COMBINEDTREATi + i . (2) To distinguish information and persuasion, the next analysis divides the sample by whether the re- spondent correctly answered two questions about COVID at baseline: whether it can be transmitted by 4 Smaller mosques do not always hold Eid prayers. 5 Note that the estimation sample here includes the secular treatment, the combined treatment, and the control group. The latter group is the omitted category in equation (2). These categories are mutually exclusive and exhaustive (i.e. the experiment does not use a factorial design)—see table 1—thus the issue of interaction terms for a factorial design does not arise. The World Bank Economic Review 521 people who show no symptoms, and whether it can be transmitted through coughing or sneezing with- out touching. Understanding these two facts is crucial, as it provides the basis for the need for preventive measures even among apparently healthy congregants. Tt baseline,both questions were answered correctly Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 by 69 percent of respondents. Equation (3) interacts the main estimates with the respondent’s baseline knowledge of how COVID is transmitted: Yi = β0 + β1 ANYTREATi + β2 BASELINERIGHTi + β3 ANYTREATi × BASELINERIGHTi + i , (3) where BASELINERIGHT is an indicator for whether the respondent answered both baseline knowledge questions correctly. If information transmission is the key mechanism, this implies β 3 < 0 (there is a smaller impact on respondents who already understood COVID transmission at baseline). In contrast, if persuasion is the key mechanism, this implies β 3 ≥ 0 (there is an equal or greater impact on respondents who understood COVID transmission at baseline). To differentiate between respondents who are uncer- tain versus those who are convinced of misinformation, the sample is divided into three groups based on a Likert scale measure of the respondent’s certainty in his answer: those who are certain and correct, uncertain, or certain and wrong. An additional heterogeneity specification adapts equation (3), dividing respondents into these groups, as an alternative approach to distinguishing these mechanisms. It is possible that simply answering the baseline survey questions plays a role in making COVID salient to religious leaders and increasing their likelihood of recommending preventive measures. To test this, a secondary sample (Sample 2) facilitates a test for the effect of the control group call on behavior. Sam- ple 2 is constructed by calling respondents for the mystery-shopper call from a “super-control” group who were not contacted for baseline calls. To ensure an equivalent sample in both control and super- control groups in Sample 2, enumerators conduct mystery-shopper calls with control-group respondents regardless of baseline response status. Thus this secondary analysis compares mystery-shopper responses between control (N = 511) and super-control respondents (N = 631) without any sample restriction on baseline response (fig. S1.2, box 3, yellow and green nodes). Response rates to the mystery-shopper calls do not differ between the control and super-control arms (table S1.3, column 2). Equation (4) simply compares the control to the super-control arm: Yi = β0 + β1CONT ROLi + i . (4) The scripts of the treatments and mystery-shopper data collection are detailed in supplementary online appendix S2. The study design and outcomes were pre-registered (AEARCTR-0005740, Version 2.0); supplementary online appendix S4 discusses changes in the analysis from the pre-registration. Replication data for this study are available from the Harvard dataverse at https://doi.org/10.7910/DVN/CV1KCJ. 3. Results 3.1. Main Results Table 2 shows descriptive statistics and balance for the main sample from the baseline survey. Respondents lead community mosques with an average of 40 people attending daily evening prayer before COVID. Most respondents have less than 10th grade (Matric) education and some training in madrasa (religious schooling). The majority report that they are receiving frequent messages about COVID. However, the majority do not believe COVID is present in their communities. Respondents reported steps they have taken to prevent spread of COVID in the mosque, without being prompted with any specific step; almost all respondents mentioned at least one step, but only about 25 percent mention masks. Respondents answered two basic questions about COVID transmission: whether it can be spread by people who show no symptoms, and whether it can be spread through coughing even if two people do not touch. About Table 2. Descriptive Statistics and Balance 522 Combined secular + religious Secular only Control Pairwise t-test (1) (2) (3) (1) − (2) (1) − (3) (2) − (3) Mean Mean Variable N Mean/(SE) N Mean/(SE) N Mean/(SE) N difference N difference N Mean difference Ed Matric or above 185 0.465 213 0.493 211 0.431 398 −0.028 396 0.034 424 0.062 (0.037) (0.034) (0.034) Attended madrasa 178 0.831 208 0.841 207 0.758 386 −0.010 385 0.073∗ 415 0.083∗∗ (0.028) (0.025) (0.030) Typical daily congregation 169 39.592 202 36.327 193 38.995 371 3.265 362 0.597 395 −2.668 pre-COVID (3.330) (3.008) (4.453) Urban 185 0.211 213 0.239 211 0.213 398 −0.029 396 −0.002 424 0.026 (0.030) (0.029) (0.028) Governing party constituency 158 0.430 170 0.376 174 0.362 328 0.054 332 0.068 344 0.014 (0.040) (0.037) (0.037) Believes COVID may be present in 173 0.116 202 0.084 199 0.186 375 0.031 372 −0.070∗ 401 −0.102∗∗∗ community (0.024) (0.020) (0.028) BL COVID knowledge: Certain and 162 0.580 191 0.576 188 0.580 353 0.004 350 0.000 379 −0.004 correct (0.039) (0.036) (0.036) BL COVID knowledge: Uncertain 162 0.321 191 0.293 188 0.324 353 0.028 350 −0.003 379 −0.031 (0.037) (0.033) (0.034) BL COVID knowledge: Certain and 162 0.099 191 0.131 188 0.096 353 −0.032 350 0.003 379 0.035 wrong (0.024) (0.024) (0.022) Received few/no COVID messages 168 0.333 200 0.350 195 0.318 368 −0.017 363 0.015 395 0.032 last week (0.036) (0.034) (0.033) Baseline step: Short sermon 131 0.061 181 0.088 187 0.086 312 −0.027 318 −0.024 368 0.003 (0.021) (0.021) (0.021) Baseline step: Clean mosque 131 0.313 181 0.359 187 0.289 312 −0.046 318 0.024 368 0.070 (0.041) (0.036) (0.033) Baseline step: Soap 131 0.328 181 0.260 187 0.246 312 0.069 318 0.082 368 0.014 Vyborny Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Table 2. Continued Combined secular + religious Secular only Control Pairwise t-test (1) (2) (3) (1) − (2) (1) − (3) (2) − (3) Mean Mean Variable N Mean/(SE) N Mean/(SE) N Mean/(SE) N difference N difference N Mean difference (0.041) (0.033) (0.032) Baseline step: Remove mats 131 0.565 181 0.564 187 0.572 312 0.001 318 −0.007 368 −0.009 (0.043) (0.037) (0.036) The World Bank Economic Review Baseline step: Elderly/sick 131 0.099 181 0.155 187 0.166 312 −0.055 318 −0.067∗ 368 −0.011 (0.026) (0.027) (0.027) Baseline step: Distancing 131 0.634 181 0.586 187 0.588 312 0.048 318 0.045 368 −0.003 (0.042) (0.037) (0.036) Baseline step: Announcements 131 0.115 181 0.133 187 0.128 312 −0.018 318 −0.014 368 0.004 (0.028) (0.025) (0.025) Baseline step: Ablution at home 131 0.191 181 0.227 187 0.225 312 −0.036 318 −0.034 368 0.002 (0.034) (0.031) (0.031) Baseline step: Mask 131 0.221 181 0.276 187 0.219 312 −0.055 318 0.002 368 0.057 (0.036) (0.033) (0.030) Baseline step: Other step 131 0.107 181 0.171 187 0.150 312 −0.064∗ 318 −0.043 368 0.022 (0.027) (0.028) (0.026) Baseline step: Number of steps 131 2.634 181 2.818 187 2.668 312 −0.184 318 −0.035 368 0.149 (0.126) (0.126) (0.122) Reports no steps to prevent COVID 131 0.038 181 0.055 187 0.080 312 −0.017 318 −0.042 368 −0.025 in mosque (0.017) (0.017) (0.020) Gov’t trusted source 168 0.214 196 0.173 194 0.180 364 0.041 362 0.034 390 −0.007 (0.032) (0.027) (0.028) Ulema trusted source 167 0.707 195 0.790 193 0.699 362 −0.083∗ 360 0.007 388 0.090∗∗ (0.035) (0.029) (0.033) Note: Example for table source text. BL COVID knowledge correct is an indicator for whether the respondent answered both knowledge questions about COVID transmission correctly at baseline (i.e. COVID can “definitely” or “probably” be transmitted without touching, and a person can “definitely” or “probably” have COVID if he/she does not appear visibly ill. BL COVID knowledge correct and certain is an indicator for whether the respondent answered both knowledge questions about COVID transmission correctly at baseline and indicated certainty (i.e. “definitely” responses to both questions). BL COVID knowledge wrong and certain is an indicator for whether the respondent answered either baseline knowledge question about COVID transmission incorrectly with certainty (i.e. “definitely not” answers on either or both of the questions). All other cases are classified as BL COVID knowledge uncertain. The p-values of F-test for joint significance of all variables in predicting assignment to any treatment:.1; secular only arm:.08. Sample sizes for covariates vary because of non-response to baseline questions due to some respondents hanging up before completing the baseline call. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. 523 Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 524 Vyborny Table 3. Impact of Persuasion Treatment on Religious Leaders’ Instructions to Mystery Shopper Panel A: Effect of persuasion treatment (pooled) Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Index components: Index—advice Bring own Ablution at Mask to MS prayer mat home recommended Mask required Asks father’s age (1) (2) (3) (4) (5) (6) Any treatment 0.066∗∗ 0.040 0.084∗∗ 0.109∗∗ 0.097∗∗ −0.001 (0.026) (0.042) (0.042) (0.042) (0.041) (0.013) Observations 609 609 609 609 609 609 Control group mean 0.375 0.531 0.526 0.436 0.360 0.024 Panel B: Disaggregated by individual persuasion scripts Index components: Index—advice Bring own Ablution at Mask Mask required Asks father’s to MS prayer mat home recommended age (1) (2) (3) (4) (5) (6) Secular + religious treatment 0.054∗ 0.010 0.085∗ 0.094∗ 0.072 0.009 (0.030) (0.050) (0.050) (0.050) (0.049) (0.017) Secular treatment 0.076∗∗ 0.065 0.084∗ 0.123∗∗ 0.119∗∗ −0.010 (0.030) (0.048) (0.048) (0.048) (0.048) (0.013) Observations 609 609 609 609 609 609 Covariates No No No No No No p-value, treatment effects equal 0.464 0.264 0.992 0.564 0.354 0.233 Note: Sample 1 (treatment and control, respondents who answered both the baseline survey and the mystery-shopper call). “Bring own prayer mat” was included in the instrument but excluded from the pre-analysis plan in error; index results are robust to the exclusion of this variable (table S4.1). Robust standard errors. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. 60 percent of respondents answered both questions correctly in the affirmative and expressed confidence in their answers, a third were unsure, and 10 percent gave a definite no to one or both questions. Overall, the randomization is well balanced. Of 72 tests, 8 are significant at the 10 percent level or greater; the p-value on the joint F-test of the significance of all variables in predicting any treatment is 0.1, and on predicting the secular-only treatment 0.08. Respondents in the control group appear to be slightly more likely to believe that COVID is present in their community, and more likely to report discouraging the elderly or sick from attending mosque during the pandemic. Both of these apparent imbalances should not drive the treatment effects of interest; if anything, they should bias the estimates towards zero. Table 3 shows the main results of the experiment. In the control group, respondents recommended on average 38 percent of the counter-COVID measures to callers; about half recommended the caller bring his own prayer mat and do ablution at home, 44 percent recommended a mask, and only 36 percent said a mask was required. Only 2 percent of respondents asked the caller (unprompted) about his father’s age. Respondents may not have imagined the callers’ fathers to be elderly (most enumerators are in their 20s and 30s). Perhaps more importantly, the religious importance of congregational prayer for men may be sufficient that religious leaders disregard this component of the official rules about COVID; in contrast, wearing a mask or doing ablution before attending still allows members of the congregation to completely fulfill their religious obligations. Panel A shows the main treatment effects, comparing treatment and control groups (equation (1) esti- mated on Sample 1). Overall, the treatment increased the index of COVID compliance instructions by 18 percent (7 percentage points). The effects are driven by an increase in recommendations to do ablution at home and to wear a mask; the proportion who tell callers they are required to wear a mask increases by The World Bank Economic Review 525 25 percent over the control group mean (10 percentage points over a mean of 0.36). Overall, the treatment had a large and significant impact on respondents’ observed behavior. Panel B uses the same sample to break down the results by treatment arm, to investigate whether Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 religious persuasion has any additional effect with religious leaders. The two treatment effects on the index ˆ 2 = 0.076 and statistically indistinguishable, p = 0.46). The treatment is ˆ 1 = 0.054; β are similar in size (β equally effective with or without religious content. 3.2. Information as a Mechanism Table 4 estimates equation (3) to investigate whether the effect of the treatment works through a mech- anism of new information about COVID transmission, rather than persuasion. The sample is divided by whether the respondent correctly answered two questions about COVID at baseline: whether it can be transmitted by people who show no symptoms, and whether it can be transmitted through coughing or sneezing without touching. In panel A, respondents who gave the correct answers but were uncertain are classified as having correct knowledge at baseline. These groups are interacted with the treatment indicator. The effects are completely driven by respondents who gave the correct answers at baseline; treatment effects on the two groups are statistically significantly different at the 5 percent level. To investigate the role of the treatment in resolving uncertainty, panel B splits respondents into three groups: those who answered the knowledge questions correctly at baseline, those who were uncertain, and those who gave the wrong answer but said they were certain about it. The effects are driven completely by those who are correct and certain in their beliefs at baseline.6 As in any heterogeneous treatment effects analysis, it is possible that other factors, such as education, could be correlated with respondents’ knowledge about COVID and could themselves mediate respon- siveness to treatment. Table S1.5 tests for robustness of the results in table 4 to this possibility. Panel A shows the correlation between respondents’ characteristics and correct answers to the baseline COVID questions. More-educated respondents are more likely to answer these questions correctly; respondents who report receiving fewer messages about COVID are less likely to answer correctly. Panel B shows the results of equation (3), in which the dependent variable is the index of advice to the mystery shop- per; column 1 repeats the original specification, while columns 2–4 also include these respondent baseline characteristics and their interaction with treatment as additional control variables. If the heterogeneity re- sults in table 4 were driven by differences in these other characteristics, the coefficient on treatment would be expected to change. Instead, it is similar in size across specifications, suggesting that these differences do not drive the heterogeneity results. 3.3. Testing for Survey or Reminder Effects Table S1.6 uses Sample 2 to investigate whether simply receiving the baseline survey, which has no infor- mational content but makes COVID salient to respondents or may remind them of existing knowledge about government rules, drives the results (equation (4)). This specification compares the responses be- tween the control (baseline survey + mystery shopper) and super-control group (mystery shopper only). Administering a survey alone has no detectable effect on the index of instructions to the mystery shopper. As an additional test for the possibility of reminder effects as an alternative mechanism, fig. S1.3 tests whether treatment effects decay in the short run. The experiment was carried out over a period of several 6 Treatment group respondents were also asked as a part of the treatment script whether they were already aware of the 20-point plan for mosque COVID compliance measures (see supplementary online appendix S2.1); 97 percent indicated they were aware of the plan. This supports the results of the formal heterogeneity analysis, suggesting that information about the COVID compliance rules themselves is an unlikely channel for treatment effects; however, this question could be subject to social desirability bias effects, and did not test respondents’ knowledge, as it was designed primarily to encourage engagement with the treatment script. 526 Table 4. Heterogeneous Effects by Baseline Knowledge about COVID Transmission Panel A Index—advice to Bring own prayer Ablution at Mask MS mat home recommended Mask required Asks father’s age (1) (2) (3) (4) (5) (6) Any treatment = 1 −0.022 −0.056 0.015 0.014 −0.041 −0.044 (0.050) (0.080) (0.079) (0.080) (0.078) (0.034) BL COVID knowledge correct = 1 −0.044 −0.040 −0.048 −0.040 −0.040 −0.056∗ (0.049) (0.077) (0.077) (0.077) (0.075) (0.032) Any treatment = 1 × BL COVID knowledge correct = 1 0.131∗∗ 0.149 0.106 0.139 0.198∗∗ 0.064∗ (0.060) (0.097) (0.096) (0.097) (0.094) (0.036) Observations 544 544 544 544 544 544 Control mean Y | baseline correct 0.370 0.520 0.520 0.440 0.360 0.010 Control mean Y | baseline wrong 0.410 0.560 0.570 0.480 0.400 0.060 Proportion of sample correct 0.690 0.690 0.690 0.690 0.690 0.690 p-value: Total effect on subgroup = 0 0.001 0.085 0.025 0.005 0.003 0.129 Vyborny Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Table 4. Continued Panel B Index—advice to Bring own prayer Ablution at Mask Mask required Asks father’s MS mat home recommended age (1) (2) (3) (4) (5) (6) Any treatment = 1 0.012 0.033 −0.007 0.050 0.003 −0.021 (0.050) (0.080) (0.079) (0.080) (0.078) (0.032) The World Bank Economic Review BL COVID knowledge correct and certain −0.033 −0.027 −0.086 −0.000 −0.010 −0.040 (0.050) (0.080) (0.080) (0.080) (0.078) (0.029) BL COVID knowledge wrong and certain −0.003 0.070 −0.035 −0.070 0.012 0.006 (0.089) (0.132) (0.134) (0.132) (0.131) (0.061) Any treatment = 1 × BL COVID knowledge correct and 0.105∗ 0.051 0.149 0.104 0.179∗ 0.042 certain (0.061) (0.100) (0.099) (0.100) (0.097) (0.035) Any treatment = 1 × BL COVID knowledge wrong and −0.026 −0.156 0.110 0.024 −0.074 −0.034 certain (0.105) (0.161) (0.160) (0.161) (0.157) (0.063) Observations 541 541 541 541 541 541 Control mean Y | certain and correct 0.371 0.514 0.505 0.459 0.367 0.009 Control mean Y | uncertain 0.403 0.541 0.590 0.459 0.377 0.049 Control mean Y | certain and wrong 0.400 0.611 0.556 0.389 0.389 0.056 Proportion of sample certain and correct 0.579 0.579 0.579 0.579 0.579 0.579 Proportion of sample uncertain 0.312 0.312 0.312 0.312 0.312 0.312 Proportion of sample certain and wrong 0.109 0.109 0.109 0.109 0.109 0.109 p-value: Total effect on subgroup BL correct and certain = 0.001 0.155 0.016 0.009 0.002 0.179 0 p-value: Total effect on subgroup certain and wrong = 0 0.875 0.378 0.460 0.594 0.600 0.307 Note: Sample 1 (treatment and control arms, respondents who answered baseline survey and mystery-shopper call). Sample size varies from table 3 because of non-response to baseline questions due to some respondents hanging up before completing the baseline call. BL COVID knowledge correct is an indicator for whether the respondent answered both knowledge questions about COVID transmission correctly at baseline (i.e. COVID can “definitely” or “probably” be transmitted without touching, and a person can “definitely” or “probably” have COVID if he/she does not appear visibly ill. BL COVID knowledge correct and certain is an indicator for whether the respondent answered both knowledge questions about COVID transmission correctly at baseline and indicated certainty (i.e. “definitely” responses to both questions). BL COVID knowledge wrong and certain is an indicator for whether the respondent answered either baseline knowledge question about COVID transmission incorrectly with certainty (i.e. “definitely not” answers on either or both of the questions). Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. 527 Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 528 Vyborny weeks, and call order was randomized; this allows us to test for a decay in effects over time. The estimates are imprecise, but do not suggest a decay in treatment effects. Simply providing basic information about COVID transmission or reminding respondents about gov- Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 ernment regulations does not appear to be the mechanism for the results. Rather, the one-on-one persua- sion, which made salient the importance of the mosque as a venue for transmission and emphasized the respondents’ key leadership role in protecting the vulnerable in their community, mobilizes respondents who already believe that there is a risk of COVID spread to respond. 3.4. Heterogeneity by Relationship with Secular and Religious Authorities Table S1.7 uses baseline survey information as well as geographic variation in the sample to test whether there is heterogeneity in responses by the respondent’s relationship with secular and religious authorities. The treatment dummy is interacted with a series of indicators of these relationships: Yi = β0 + β1 TREATi + β2 Xi + β3 TREATi × Xi + i . There is no pattern of stronger response among respondents with a closer relationship with secular or religious authorities, whether measured by respondents’ survey responses on sources of information they trust, the political alignment of the constituencies where they live, or madrasa training. This result contrasts with recent evidence from the United States and Brazil on the politicization of responses to COVID-19 public-health advice (Grossman et al. 2020; Alcott et al. 2020; Bursztyn et al. 2020; Painter and Qiu 2020; Milosh et al. 2021; Gadarian, Goodman, and Pepinsky 2021). This could be because these patterns are particular to the high degree of polarization in those countries (Barari et al. (2020) find no relationship between reported COVID compliance and trust in government across Italian respondents). Alternatively, it may be because personal interaction is more effective than mass communications in cross- ing party lines, as consistently shown in the political science literature on voter mobilization (Gerber and Green (2019) review this literature in detail). 4. Discussion In this study, a randomized controlled trial establishes that one-on-one interactive communication can be effective in influencing community religious leaders to instruct congregants to follow government- mandated public-health measures in their mosques. The evidence is consistent with a persuasion channel, rather than an information or education channel, for these effects. Perhaps surprisingly, there is no ev- idence that an extensive script linking the persuasive argument to Islamic scripture and the advice of internationally respected religious authorities has any additional effect compared to the basic persuasion script designed to motivate respondents based on their identity as community leaders. Given the limitations on the effectiveness of generic mass messaging and the infeasibility of direct citizen engagement at a population scale, engaging community and religious leaders may be necessary for many low-capacity states to effectively carry out policy. Although the intervention tested in this RCT requires time input at an individual level, its scripted nature and phone-based delivery means it is still low cost. The call center costs (call center agents, managers, and connection costs) amount to approximately $0.75 (in 2022 USD) per religious leader targeted in the list. The average of 230 congregants at Friday prayer reported by respondents in the pilot data suggests there are approximately 200,000 such community mosques nationwide. Thus, carrying out such a campaign at a national scale would involve outreach to 200,000 religious leaders for a cost of approximately $150,000 USD. This is equivalent to the cost estimate from a Pakistan-based advertising agency for a recent mass-media campaign on polio, another high-priority public-health issue, for 150 one-minute advertising spots each week for three weeks. Thus the cost of expanding this intervention at a national level would be comparable to a typical mass-media campaign approach. The World Bank Economic Review 529 Beyond this specific, standardized intervention, governments could consider establishing a mechanism for community-level engagement with local community leaders, such as district-level outreach teams. In this study setting, Pakistan, the government engages with prominent religious leaders at the national level. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 It is this engagement that led to the 20-point plan for mosques. Yet the lack of compliance with this plan at the community level illustrates the need for engagement at the community level. In Pakistan, there is no strategy or government body that has direct communication or engagement with religious or informal leaders. The results of this study demonstrate that such engagement can be effective. Such approaches may be promising to explore not only for the case of public-health campaigns (such as encouraging trust in vaccination), but also for a much broader set of policies where establishing public trust is key—ranging from encouraging households to educate daughters to discouraging participation in violent conflict. Anecdotally, the caller’s identity as part of a non-government institution may have been important not only to elicit truthful information in the baseline survey, but also potentially to engage effectively in persuasion given that respondents might be concerned about speaking to a government caller about regulations with which they might not be compliant. This study cannot test this directly; future work could investigate this issue systematically by varying the framing of the identity of the caller. Data Availability Statement Replication data for this study are available at the https://doi.org/10.7910/DVN/CV1KCJ, the Harvard dataverse. REFERENCES Abaluck, J., L. H. Kwong, A. Styczynski, A. Haque, A. Kabir, E. Bates-Jeffries, E. Crawford et al., 2022. “The Impact of Community Masking on COVID-19: A Cluster-Randomized Trial in Bangladesh.” Science 375: 6577. Acemoglu, D., A. Cheema, A. I. Khwaja, and J. A. Robinson, 2020. “Trust in State and Nonstate Actors: Evidence from Dispute Resolution in Pakistan.” Journal of Political Economy 128(8): 3090–147. Acemoglu, D., T. Reed, and J. A. Robinson, 2014. “Chiefs: Economic Development and Elite Control of Civil Society in Sierra Leone.” Journal of Political Economy 122(2): 319–68. Acemoglu, D., and J. Robinson, 2023. “Weak, Despotic, or Inclusive? How State Type Emerges from State versus Civil Society Competition.” American Political Science Review 117(2): 407–420. Acemoglu, D., D. Ticchi, and A. Vindigni, 2011. “Emergency and Persistence of Inefficient States.” Journal of the European Economic Association 9(2): 177–208.. Alatas, V., A. Banerjee, R. Hanna, B. A. Olken, R. Purnamasari, and M. Wai-Poi, 2019. “Does Elite Capture Matter? Local Elites and Targeted Welfare Programs in Indonesia.” AEA Papers and Proceedings 109: 334–39. Alatas, V., A. Banerjee, R. Hanna, B. A. Olken, and J. Tobias, 2012. “Targeting the Poor: Evidence from a Field Experiment in Indonesia.” American Economic Review 102(4): 1206–40. Alcott, H., L. Boxell, J. Conway, M. Gentzkow, M. Thaler, and D. Yang, 2020. “Polarization and Public Health: Partisan Differences in Social Distancing during the Coronavirus Pandemic.” Journal of Public Economics 191: 104254. Alsan, M., and S. Eichmeyer, 2023. “Experimental Evidence on the Effectiveness of Non-experts for Improving Vaccine Demand.” American Economic Journal: Economic Policy 16 (1): 394–414. Anagol, S., S. Cole, and S. Sarkar, 2017. “Understanding the Advice of Commissions-Motivated Agents: Evidence from the Indian Life Insurance Market.” Review of Economics and Statistics 99(1): 1–15. Ashraf, N., O. Bandiera, and B. K. Jack, 2014. “No Margin, No Mission? A Field Experiment on Incentives for Public Service Delivery.” Journal of Public Economics 120: 1–17. Auriol, E., J. Lassebie, A. Panin, E. Raiber, and P. Seabright, 2020. “God Insures Those Who Pay? Formal Insurance and Religious Offerings in Ghana.” Quarterly Journal of Economics 135(4): 1799–1848. 530 Vyborny Balán, P., A. Bergeron, G. Tourek, and J. L. Weigel, 2022. “Local Elites as State Capacity: How City Chiefs Use Local Information to Increase Tax Compliance in the Democratic Republic of the Congo.” American Economic Review 112(3): 762–97. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Baldwin, K., 2013. “Why Vote with the Chief? Political Connections and Public Goods Provision in Zambia.” Amer- ican Journal of Political Science 57(4): 794–809. ———. 2014. “When Politicans Cede Control of Resources: Land, Chiefs and Coalition-Building in Africa.” Com- parative Politics: 46(3): 253–271. ———. 2019. “Elected MPs, Traditional Chiefs, and Local Public Goods: Evidence on the Role of Leaders in Co- production from Rural Zambia.” Comparative Political Studies 52(12): 1925–1956. Banerjee, A., M. Alsan, E. Breza, A. Chandrasekhar, A. Chowdhury, E. Duflo, P. Goldsmith-Pinkham et al., 2020. “Messages on COVID-19 Prevention in India Increased Symptoms Reporting and Adherence to Preventive Behav- iors among 25 Million Recipients with Similar Effects on Non-recipient Members of Their Communities.” NBER Working Paper 27496. Banerjee, A., R. Chattopadhyay, E. Duflo, D. Keniston, and N. Singh, 2021. “Improving Police Performance in Ra- jasthan, India: Experimental Evidence on Incentives, Managerial Autonomy, and Training.” American Economic Journal: Economic Policy 13(1): 36–66. Banerjee, A., and L. Iyer, 2005. “History, Institutions, and Economic Performance: The Legacy of Colonial Land Tenure Systems in India.” American Economic Review 95(4): 1190–213. Banuri, S., S. Dercon, and V. Gauri, 2019. “Biased Policy Professionals.” World Bank Economic Review 33(2): 310–27. Barari, S., S. Caria, A. Davola, P. Falco, T. Fetzer, S. Fiorin, L. Hensel et al., 2020. “Evaluating COVID-19 Public Health Messaging in Italy: Self-Reported Compliance and Growing Mental Health Concerns.” medRxiv: 1–19. Bassi, V., and I. Rasul, 2017. “Persuasion: A Case Study of Papal Influences on Fertility-Related Beliefs and Behavior.” American Economic Journal: Applied Economics 9(4): 250–302. Basurto, M. P., P. Dupas, and J. Robinson, 2020. “Decentralization and Efficiency of Subsidy Targeting: Evidence from Chiefs in Rural Malawi.” Journal of Public Economics 185: 104047. Bate, R., G. Z. Jin, and A. Mathur, 2011. “Does Price Reveal Poor-Quality Drugs? Evidence from 17 Countries.” Journal of Health Economics 30(6): 1150–63. Bate, R., L. Mooney, and K. Hess, 2010. “Medicine Registration and Medicine Quality: A Preliminary Analysis of Key Cities in Emerging Markets.” Research and Reports in Tropical Medicine: 1: 89–93. Ben-Nun, Bloom P., G. Arikan, and M. Courtemanche, 2015. “Religious Social Identity, Religious Belief, and Anti- immigration Sentiment.” American Political Science Review 109(2): 203–21. Bennett, D., A. Naqvi, and W.-P. Schmidt, 2018. “Learning, Hygiene and Traditional Medicine.” Economic Journal 128(612): 545–74. Bertrand, M., S. Djankov, R. Hanna, and S. Mullainathan, 2007. “Obtaining a Driver’s License in India: An Experi- mental Approach to Studying Corruption.” Quarterly Journal of Economics 122(4): 1639–1676. Besley, T. J., R. Burgess, A. Khan, and G. Xu, 2022. “Bureaucracy and Development.” Annual Review of Economics 14: 397–424. Besley, T., and T. Persson, 2009. “The Origins of State Capacity: Property Rights, Taxation, and Politics.” American Economic Review 99(4): 1218–1244. Blair, G., R. Littman, E. R. Nugent, R. Wolfe, M. Bukar, B. Crisman, A. Etim et al., 2021. “Trusted Authorities Can Change Minds and Shift Norms during Conflict.” Proceedings of the National Academy of Sciences of the United States of America 118(42): e2105570118. Blair, R. A., T. Curtice, D. Dow, and G. Grossman, 2022. “Public Trust, Policing, and the COVID-19 Pandemic: Evidence from an Electoral Authoritarian Regime.” 305: 115045. Social Science & Medicine. Blattman, C., A. Hartman, and R. Blair, 2014. “How to Promote Order and Property Rights under Weak Rule of Law? An Experiment in Changing Dispute Resolution Behavior.” American Political Science Review 108(1): 100–120. Boyer, C., E. Levy Paluck, J. Annan, T. Nevatia, J. Cooper, J. Namubiru, L. Heise et al., 2022. “Religious Leaders Can Motivate Men to Cede Power and Reduce Intimate Partner Violence: Experimental Evidence from Uganda.” Proceedings of the National Academy of Sciences 119 (31): e2200262119. Broockman, D. E., 2013. “Black Politicians Are More Intrinsically Motivated to Advance Blacks’ Interests: A Field Experiment Manipulating Political Incentives.” American Journal of Political Science 57(3): 521–36. The World Bank Economic Review 531 ———. 2014. “Distorted Communication, Unequal Representation: Constituents Communicate Less to Representa- tives Not of Their Race.” American Journal of Political Science 58(2): 307–21. Bryan, G., J. Choi, and D. Karlan, 2020. “Randomizing Religion: The Impact of Protestant Evangelism on Economic Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Outcomes.” Quarterly Journal of Economics 136(1): 293–380. Bursztyn, L., A. Rao, C. Roth, and D. Yanagizawa-Drott, 2020. “Misinformation during a Pandemic.” NBER Working Paper 27417. Callen, M., S. Gulzar, A. Hasanain, M. Y. Khan, and A. Rezaee, 2020. “Data and Policy Decisions: Experimental Evidence from Pakistan.” Journal of Development Economics 146: 102523. Casey, K., R. Glennerster, and E. Miguel, 2012. “Reshaping Institutions: Evidence on Aid Impacts Using a Preanalysis Plan.” Quarterly Journal of Economics 127(4): 1755–812. Casey, K., R. Glennerster, E. Miguel, and M. Voors, 2023. “Skill versus Voice in Local Development.” Review of Economics and Statistics 105(2): 311–326. Cavgias, A., R. Bruce, and L. Meloni, 2023. “Policy Enforcement in the Presence of Organized Crime: Evidence from Rio de Janeiro.” Journal of Development Economics 162: 103071. Chhibber, P., and J. S. Sekhon, 2014. “The Asymmetric Role of Religious Appeals in India.” Mimeo, Berkeley Univer- sity. Clingingsmith, D., A. Khwaja, and M. Kremer, 2009. “Estimating the Impact of the Hajj: Religion and Tolerance in Islam’s Global Gathering.” Quarterly Journal of Economics 124(3): 1133–70. Condra, L. N., M. Isaqzadeh, and S. Linardi, 2019. “Clerics and Scriptures: Experimentally Disentangling the Influence of Religious Authority in Afghanistan.” British Journal of Political Science 49(2): 401–19. Conning, J., and M. Kevane, 2002. “Community Based Targeting Mechanisms for Social Safety Nets: A Critical Review.” World Development 30(3): 375–394. Currie, J., W. Lin, and W. Zhang, 2011. “Patient Knowledge and Antibiotic Abuse: Evidence from an Audit Study in China.” Journal of Health Economics 30(5): 933–49. Dai, H., S. Saccardo, M. A. Han, L. Roh, N. Raja, S. Vangala, H. Modi et al., 2021. “Behavioral Nudges Increase COVID-19 Vaccinations.” Nature 597: 404–409. Das, J., A. Holla, V. Das, M. Mohanan, D. Tabak, and B. Chan, 2012. “In Urban and Rural India, A Standardized Patient Study Showed Low Levels of Provider Training and Huge Quality Gaps.” Health Affairs 31(12): 2774–84. De Kadt, D., and H. A. Larreguy, 2018. “Agents of the Regime? Traditional Leaders and Electoral Politics in South Africa.” Journal of Politics 80(2): 382–99. De Walque, D., 2007. “How Does the Impact of an HIV/AIDS Information Campaign Vary with Educational Attain- ment? Evidence from Rural Uganda.” Journal of Development Economics 84(2): 686–714. Dell, M., N. Lane, and P. Querubin, 2018. “The Historical State, Local Collective Action, and Economic Development in Vietnam.” Econometrica 86(6): 2083–121. Dellavigna, S., and M. Gentzkow, 2010. “Persuasion: Empirical Evidence.” Annual Review of Economics 2: 643–69. Deserranno, E., 2018. “Financial Incentives as Signals: Experimental Evidence from the Recruitment of Village Pro- moters in Uganda.” American Economic Journal: Applied Economics 11(1): 277–317. Desposato, S., 2015. Ethics and Experiments Problems and Solutions for Social Scientists and Policy Professionals, Routledge. New York NY. Deutsche Welle, 2020. “Eid Festivities Raise Coronavirus Surge Fears in South Asia.”Online news report. Dizon-Ross, R., P. Dupas, and J. Robinson, 2017. “Governance and the Effectiveness of Public Health Subsidies: Evidence from Ghana, Kenya and Uganda.” Journal of Public Economics 156: 150–69. Driscoll, A., G. Cepaluni, F. D. S. Guimarães, and P. Spada, 2018. “Prejudice, Strategic Discrimination, and the Electoral Connection: Evidence from a Pair of Field Experiments in Brazil.” American Journal of Political Science 62(4): 781– 95. Druckman, J., 2021. “A Framework for the Study of Persuasion.” Annual Review of Political Science 25: 65–88. Duflo, B. E., P. Dupas, and M. Kremer, 2015. “Education, HIV and Early Fertility: Experimental Evidence from Kenya.” American Economic Review 105(9): 2757–97. Dupas, P., 2011. “Do Teenagers Respond to HIV Risk Information? Evidence from a Field Experiment in Kenya.” American Economic Journal: Applied Economics 3(1): 1–34. Field, E., and K. Vyborny, 2020. “Information Gaps and de Jure Legal Rights: Evidence from Pakistan.” EDI Working Paper Series. 532 Vyborny Finan, F., B. A. Olken, and R. Pande, 2017. “The Personnel Economics of the Developing State.” Handbook of Eco- nomic Field Experiments 2: 467–514. Fitzpatrick, A., 2022. “The Impact of Public Health Sector Stockouts on Private Sector Prices and Access to Healthcare: Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Evidence from the Anti-malarial Drug Market.” Journal of Health Economics 81(102544): 1–41. Freedman, M., 2020. “Vote with Your Rabbi: The Electoral Effects of Religious Institutions in Israel.” Electoral Studies 68: 102241. Friedman, W., B. Woodman, and M. Chatterji, 2015. “Can Mobile Phone Messages to Drug Sellers Improve Treatment of Childhood Diarrhoea? - A Randomized Controlled Trial in Ghana.” Health Policy and Planning 30(Supplemental 1): i82–i92. Gadarian, S. K., S. W. Goodman, and T. Pepinsky, 2021. “Partisanship, Health Behavior, and Policy Attitudes in the Early Stages of the COVID-19 Pandemic.” PLOS One 16(4): e0249596. Gaddis, S., 2018. “An Introduction to Audit Studies in the Social Sciences.”In Audit Studies: Behind the Scenes with Theory, Method, and Nuance. Methodos Series, edited by S. Gaddis. Springer. New York, NY. Gaikwad, N., and G. Nellis, 2021. “Do Politicians Discriminate against Internal Migrants? Evidence from Nationwide Field Experiments in India.” American Journal of Political Science 65(4): 790–806. Gallup Pakistan, 2020. “Attitude Tracker Survey Pakistan Wave 8 Results.”Online report. Gerber, D., and A. Green, 2019. Get Out the Vote: How to Increase Voter Turnout. Brookings Institution Press. Washington, DC. Gine, X., and R. K. Mazer, 2022. “Financial (Dis-)Information: Evidence from a Multi-Country Audit Study.” Journal of Public Economics 208: 104618. Grépin, K., V. Mueller, and N. Wu, 2023. “Unmasking the Truth: Experimental Evidence of Facemask Compliance in Bangladesh, Kenya, and Nigeria during the COVID-19 Pandemic.” PLOS Global Public Health 3(3): e0001086. Grose, C. R., 2014. “Field Experimental Work on Political Institutions.” Annual Review of Political Science 17: 355– 70. Grossman, G., S. Kim, J. Rexer, and H. Thirumurthy, 2020. “Political Partisanship Influences Behavioral Responses to Governors – Recommendations for COVID-19 Prevention in the United States.” Proceedings of the National Academy of Sciences.117 (39): 24144–24153. Gruber, J. H., 2005. “Religious Market Structure, Religious Participation, and Outcomes: Is Religion Good for You?” Advances in Economic Analysis & Policy 5(1). Grundler, K., and N. Potrafke, 2022. “Policy Advice of Economic Experts: Facts or Preferences?” Mimeo, ifo Institute Munich. Guiteras, R. P., D. I. Levine, S. P. Luby, T. H. Polley, K. Khatun-E-Jannat, and L. Unicomb, 2016. “Disgust, Shame, and Soapy Water: Tests of Novel Interventions to Promote Safe Water and Hygiene.” Journal of the Association of Environmental and Resource Economists 3(2): 321–59. Hetzel, M. W., A. Dillip, C. Lengeler, B. Obrist, J. J. Msechu, A. M. Makemba, C. Mshana et al., 2008. “Malaria Treatment in the Retail Sector: Knowledge and Practices of Drug Sellers in Rural Tanzania.” BMC Public Health 8: 1–11. Hjort, J., D. Moreira, G. Rao, and J. F. Santini, 2021. “How Research Affects Policy: Experimental Evidence from 2,150 Brazilian Municipalities.” American Economic Review 111(5). Hong, J. Y., and C. Paik, 2021. “Hate Thy Communist Neighbor: Protestants and Politics in South Korea.” Journal of Economic Behavior and Organization 186: 707–23. Hsiung, B. O., and P. A. Djupe, 2019. “Religion and the Extension of Trust.” Political Behavior 41(3): 609–31. Kalla, J. L., and D. E. Broockman, 2016. “Campaign Contributions Facilitate Access to Congressional Officials: A Randomized Field Experiment.” American Journal of Political Science 60(3): 545–58. Karing, A., 2024. “Social Signaling and Childhood Immunization: A Field Experiment in Sierra Leone.” Quarterly Journal of Economics. Kerwin, J. T., 2020. “Scared Straight or Scared to Death? Fatalism in Response to Disease Risks.” Mimeo, University of Minnesota. Khan, A., S. Nasim, M. Shaukat, and A. Stegmann, 2021. “Building Trust in the State with Information: Evidence from Urban Punjab.” Journal of Public Economics 202: 104494. Khan, M. Y., 2020. “Mission Motivation and Public Sector Performance: Experimental Evidence from Pakistan.” Mimeo, University of Pittsburgh. The World Bank Economic Review 533 Kilic, T., E. Whitney, and P. Winters, 2015. “Decentralised Beneficiary Targeting in Large-Scale Development Pro- grammes: Insights from the Malawi Farm Input Subsidy Programme.” Journal of African Economies 24(1): 26–56. Kitschelt, H., and S. Wilkinson, 2007. Patrons, Clients and Policies: Patterns of Democratic Accountability and Political Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Competition Cambridge University Press, Cambridge, United Kingdom. Lau, R. R., 2020. “Classic Models of Persuasion.” The Oxford Handbook of Electoral Persuasion: 27–50. Oxford University Press, Oxford, United Kingdom. Lee, D. S., 2009. “Training, Wages, and Sample Selection: Estimating Sharp Bounds on Treatment Effects.” Review of Economic Studies 76(3): 1071–102. Lee, N., 2022. “Do Policy Makers Listen to Experts? Evidence from a National Survey of Local and State Policy Makers.” American Political Science Review 116(2): 677–88. Mansuri, G., and V. Rao, 2004. “Community-Based and -Driven Development: A Critical Review.” World Bank Research Observer 19(1): 1–39. Martinez-Bravo, M., and A. Stegmann, 2022. “In Vaccines We Trust? The Effect of the CIA’s Vaccination Ruse on Immunization in Pakistan.” Journal of the European Economic Association 20(1): 150186. McCauley, J. F., 2014. “Pentecostalism as an Informal Political Institution: Experimental Evidence from Ghana.” Politics and Religion 7(4): 761–87. McClendon, G. H., 2012. “Ethics of Using Public Officials as Field Experiment Subjects.” Newsletter of the APSA Experimental Section 3(1): 13–19. ———. 2016. “Race and Responsiveness: An Experiment with South African Politicians.” Journal of Experimental Political Science 3(1): 60–74. ———. 2019. “Religious Communication and the Effects of Priming.” Oxford Research Encyclopedia of Politics. Oxford University Press, Oxford, United Kingdom. McClendon, G., and R. Riedl, 2015. “Religion as a Stimulant of Political Participation: Experimental Evidence from Nairobi, Kenya.” Journal of Politics 77(4): 1045–57. ———. 2016. “Individualism and Empowerment in Pentecostal Sermons: New Evidence from Nairobi, Kenya.” African Affairs 115(458): 119–44. ———. 2019. From Pews to Politics: Religious Sermons and Political Participation in Africa. Cambridge University Press, Cambridge, United Kingdom. ———. 2021. “Using Sermons to Study Religions’ Influence on Political Behavior.” Comparative Political Studies 54(5): 779–822. Mehmood, S., S. Naseer, and D. Chen, 2024. AI Education as State Capacity: Experimental Evidence from Pakistan Mimeo, Oxford. Mehmood, S., and A. Seror, 2023. “Religious Leaders and Rule of Law.” Journal of Development Economics 160: 102974. Mendez, M. S., and C. R. Grose, 2018. “Doubling Down: Inequality in Responsiveness and the Policy Preferences of Elected Officials.” Legislative Studies Quarterly 43(3): 457–91. Milosh, M., M. Painter, D. Van Dijcke, and A. L. Wright, 2021. “Unmasking Partisanship: How Polarization Influences Public Responses to Collective Risk.” Journal of Public Economics 204. Mohanan, M., M. Vera-Hernández, V. Das, S. Giardili, J. D. Goldhaber-Fiebert, T. L. Rabin, S. S. Raj et al., 2015. “The Know-Do Gap in Quality of Health Care for Childhood Diarrhea and Pneumonia in Rural India.” JAMA Pediatrics 169(4): 349–57. Moreno-Medina, J., 2021. “Sinning in the Rain: Weather Shocks, Church Attendance and Crime.” Review of Eco- nomics and Statistics 105 (1): 54–69. Murphy, D. M. A., V. Nourani, and D. R. Lee, 2020. “Chatting at Church: Information Diffusion through Religious Networks.” Review of Economics and Statistics 04(3): 449–464. Nakajima, N., 2021. “Evidence-Based Decisions and Education Policymakers.” Mimeo, Harvard University. Naurin, E., and P. Öhberg, 2021. “Ethics in Elite Experiments: A Perspective of Officials and Voters.” British Journal of Political Science 51(2): 890–98. Nichter, S., 2018. Votes for Survival: Relational Clientelism in Latin America. Cambridge University Press, Cambridge, United Kingdom. Nichter, S., and M. Peress, 2017. “Request Fulfilling: When Citizens Demand Clientelist Benefits.” Comparative Po- litical Studies 50(8): 1086–117. 534 Vyborny Nyhan, B., J. Reifler, S. Richey, and G. L. Freed, 2014. “Effective messages in Vaccine Promotion: A Randomized Trial.” Pediatrics 133(4): 835–842. Painter, M., and T. Qiu, 2021. “Political Beliefs Affect Compliance with COVID-19 Social Distancing Orders.” Journal Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 of Economic Behavior and Organization 185(May): 688–701. Pakistan Bureau of Statistics, 2017. “Pakistan Census 2017 - Population by Sex, Religion and Rural/Urban.”Online report. Pattan Development Organization, 2020. “Overwhelming Pandemic, Overwhelmed by Fatalist Mindset in Pak- istan.”Online report. Pew Research Center, 2018. “The Age Gap in Religion around the World.”Online report. Rahman, K. W., 2022. “International migration and the religious schooling of children in the home country: evidence from Bangladesh.” Journal of Population Economics 36: 1963–2005. Rhee, M., M. Sissoko, S. Perry, W. McFarland, J. Parsonnet, and O. Doumbo, 2005. “Use of Insecticide-Treated Nets (ITNs) following a Malaria Education Intervention in Piron, Mali: A Control Trial with Systematic Allocation of Households.” Malaria Journal 4(35): 1–9. Rink, A., 2018. “Do Protestant Missionaries Undermine Political Authority? Evidence from Peru.” Comparative Po- litical Studies 51(4): 477–513. Rogger, D. O., and R. Somani, 2023. “Hierarchy and Information.” Journal of Public Economics 219: 1–69. Sandefur, J., and B. Siddiqi, 2015. “Delivering Justice to the Poor: Theory and Experimental Evidence from Liberia.” 1–58. Mimeo, Stanford University. Sharma, K., 2018. “Clerical Persuasion and Religious Extremism: An Experiment among Sunni and Shia Muslims in Northern India.” 1–49. Mimeo, ETH Zurich. Smith, A. E., 2016. “When Clergy Are Threatened: Catholic and Protestant Leaders and Political Activism in Peru.” 9(3): 431–455. Politics and Religion. Sperber, E., G. McClendon, and O. Kaaba, 2021. “Estimating the Effect of Christian Messages on Civic Engagement: Evidence from a Community-Collaborative Study in Zambia.” University of Notre Dame Kellogg Institute Working Paper 442. Stokes, S. C., T. Dunning, and V. Brusco, 2012. Brokers, Voters, and Clientelism, Cambridge University Press, Cam- bridge, United Kingdom. Tauchmann, H., 2014. “Lee (2009) Treatment-Effect Bounds for Nonrandom Sample Selection.” Stata Journal 14(4): 884–94. Torgler, B., 2006. “The Importance of Faith: Tax Morale and Religiosity.” Journal of Economic Behavior and Orga- nization 61(1): 81–109. Vivalt, E., and A. Coville, 2023. “How Do Policy-Makers Update Their Beliefs?” Journal of Development Economics 165. Voors, M., T. Turley, E. Bulte, A. Kontoleon, and J. A. List, 2018. “Chief for a Day: Elite Capture and Management Performance in a Field Experiment in Sierra Leone.” Management Science 64(12): 5855–76. Wade, R., 1990. Governing the Market: Economic Theory and the Role of Government in East Asian Industrialization. Princeton University Press. Wilke, A. M., 2024. “How the State Discourages Vigilantism – Evidence from a Field Experiment in South Africa.” American Journal of Political Science. Inglehart, R. C. Haerpfer A. Moreno C. Welzel K. Kizilova J. Diez-Medrano M. Lagos P. Norris E. Ponarin, & B. Puranen et al., (eds.), 2018. World Values Survey: Round Six - Country-Pooled Datafile. doi.org/10.14281/18241.8. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Supplementary Online Appendix Religious Leaders’ Compliance with State Authority: Experimental Evidence from COVID-19 in Pakistan Kate Vyborny S1. Additional Figures and Tables Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Table S1.1. Self-Reported Steps to Control COVID during April 2020 Pilot Survey versus July 2020 Experiment Pilot Baseline Pairwise t-test (1) (2) (1) − (2) Variable Mean/(SE) Mean/(SE) Mean difference Keeping sermon short 0.289 0.065 0.224∗∗∗ (0.022) (0.009) Cleaning mosque 0.781 0.290 0.490∗∗∗ (0.020) (0.016) Providing soap 0.520 0.251 0.270∗∗∗ (0.025) (0.015) Remove prayer mats 0.639 0.477 0.161∗∗∗ (0.024) (0.017) Encourage own prayer mat 0.263 0.124 0.139∗∗∗ (0.022) (0.011) Discourage elderly/sick from 0.388 0.130 0.258∗∗∗ attending (0.024) (0.012) Distancing 0.542 0.510 0.033 (0.024) (0.017) No handshakes/hugs 0.369 0.189 0.180∗∗∗ (0.024) (0.014) Health announcements in 0.205 0.086 0.119∗∗∗ sermon (0.020) (0.010) Health announcements on 0.169 0.047 0.122∗∗∗ loudspeaker (0.018) (0.007) Ablution at home 0.277 0.182 0.095∗∗∗ (0.022) (0.013) Requesting/requiring masking 0.104 0.240 −0.136∗∗∗ (0.015) (0.015) Number of observations 415 830 1,245 Table S1.2. Correlates of Response to Baseline Survey and Mystery-Shopper Call Responded to baseline survey Mystery-shopper call Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 (1) (2) Rural −0.013 0.050∗∗ (0.022) (0.023) Primary −0.016 0.024 (0.057) (0.066) Middle 0.010 0.015 (0.057) (0.065) Matric −0.033 −0.022 (0.056) (0.065) Intermediate −0.039 −0.055 (0.063) (0.072) Graduate −0.012 0.012 (0.063) (0.071) Postgraduate/PhD 0.016 0.070 (0.067) (0.074) Constant 0.330∗∗∗ 0.600∗∗∗ (0.057) (0.064) Observations 2,700 2,489 Note: Column 1: Sample consists of respondents attempted for baseline in treatment and control arms (excludes super control, who were not called in the baseline). Column 2: Sample consists of respondents attempted for mystery shopper. See fig. S1.2 for more details on sample selection for each instrument. Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. Table S1.3. Balance in Response to Mystery-Shopper Calls Responded to mystery shopper (1) (2) Secular + religious treatment −0.025 – (0.038) Secular treatment 0.010 – (0.036) Control group (assigned to baseline – −0.017 survey) (0.022) Constant 0.754∗∗∗ 0.591∗∗∗ (0.026) (0.015) Observations 813 1,956 p-value, treatment effects equal 0.350 – p-value, all treatments = 0 0.640 – Sample Sample 1 Sample 2 Note: Column 1: Treatment and control arms, respondents who answered baseline survey. Column 2: Control and super-control arms, respondents for whom mystery shopper was attempted. Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. Table S1.4. Treatment Effects on Suspicion of Mystery-Shopper Caller Enumerator noted respondent was suspicious Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 (1) (2) (3) (4) Any treatment −0.010 – – – (0.024) Secular + religious – 0.007 – – treatment (0.029) Secular treatment – −0.024 – – (0.026) Any treatment = 1 – – 0.018 −0.009 (0.047) (0.038) BL COVID knowledge – – −0.008 – correct = 1 (0.044) Any treatment = 1 × – – −0.025 – BL COVID knowledge correct = 1 (0.056) Large congregation = 1 – – – −0.025 (0.042) Any treatment = 1 × – – – −0.006 large congregation = 1 (0.050) Observations 629 629 563 583 Control group mean 0.092 – – – Note: BL COVID knowledge correct is an indicator for whether the respondent answered both knowledge questions about COVID transmission correctly at baseline. Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. Figure S1.1. Timeline. 8000 16 Jun: General lockdown eased, 31 Jul: move towards Eid-ul-Azha targeted lockdowns 6000 Daily new COVID-19 cases 14 Apr: 12 May: 21-28 Jul: Clerics oppose Eid-ul-Fitr Baseline ban on Treatment Friday prayers Calls 4000 29-31 Jul: 18 Apr: 20-point Mystery action plan, Shopper mosques allowed Calls to resume prayers 16 March: First Nationwide 2000 Lockdown 0 Mar1 Apr1 May1 Jun1 Jul1 Aug1 Source: COVID case data obtained from ourworldindata.org. Table S1.5. Alternative Mediators of Heterogeneous Treatment Effects by Baseline COVID Knowledge Panel A Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Baseline knowledge correct (1) (2) (3) Ed 6–8 years – 0.021 – – (0.056) Ed 9–10 years – 0.081 – – (0.056) Ed 12+ years – 0.131∗∗ – – (0.058) Urban = 1 – – −0.041 – (0.049) Received few/no COVID messages last – – – −0.145∗∗∗ week = 1 (0.043) Observations – 544 544 544 Panel B Index—Advice to Mystery Shopper (1) (2) (3) (4) Any treatment = 1 −0.022 −0.091 −0.032 −0.006 (0.050) (0.065) (0.054) (0.057) BL COVID knowledge correct = 1 −0.044 −0.053 −0.045 −0.047 (0.049) (0.048) (0.049) (0.050) Any treatment = 1 × BL COVID 0.131∗∗ 0.132∗∗ 0.132∗∗ 0.128∗∗ knowledge correct = 1 (0.060) (0.060) (0.061) (0.061) Ed 6–8 years – −0.010 – – (0.060) Ed 9–10 years – 0.038 – – (0.062) Ed 12+ years – 0.066 – – (0.069) Any treatment = 1 × Ed 6–8 years – 0.096 – – (0.075) Any treatment = 1 × Ed 9–10 years – 0.072 – – (0.075) Any treatment = 1 × Ed 12+ years – 0.088 – – (0.084) Urban = 1 – – −0.001 – (0.051) Any treatment = 1 × Urban = 1 – – 0.045 – (0.063) Received few/no COVID messages last – – – −0.015 week = 1 (0.050) Any treatment = 1 × Received few/no – – – −0.039 COVID messages last week = 1 (0.061) Observations 544 544 544 544 Note: Sample 1 (treatment and control arms, respondents who answered baseline survey and mystery-shopper call). Sample size varies from table 3 because of non- response to baseline questions due to some respondents hanging up before completing the baseline call. Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Figure S1.2. Sampling and Response by Treatment Arm. Table S1.6. Effect of Receiving Baseline Survey on Religious Leaders’ Instructions to Mystery Shoppers: Control versus Super Control Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Index—Advice to MS (1) Control group (assigned to baseline survey) 0.019 (0.018) Observations 1,142 Note: Comparison of control versus super control in Sample 2 (respondents who answered the mystery-shopper call). Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. Figure S1.3. Heterogeneous Treatment Effects by Gap between Treatment and Mystery Shopper. Note: Figure shows coefficients β 1 to β 3 from regressing the mystery-shopper index INDEX = β 0 + β 1 TREAT × SHORTGAP + β 2 TREAT × MEDIUMGAP + β 3 TREAT × LONGGAP + β 4 MEDIUMGAP + β 5 LONGGAP + . Order of phone calls was randomized. Robust standard errors; 95 percent confidence intervals shown. Table S1.7. Heterogeneous Effects by Relationship with Secular and Religious Authorities Panel A: Pooled treatments Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 Index—Advice to mystery shopper (1) (2) (3) (4) Any treatment = 1 0.078∗∗ 0.062∗ 0.050 0.102∗ (0.030) (0.036) (0.059) (0.053) Subgroup = 1 0.103∗ −0.028 0.001 0.033 (0.059) (0.049) (0.052) (0.049) Any treatment = 1 × subgroup = 1 −0.052 0.009 0.022 −0.037 (0.072) (0.060) (0.066) (0.062) Observations 558 502 593 555 Subgroup Gov’t trusted Governing party Attended madrasa Ulema trusted source source constituency p-value: Total effect on subgroup = 0 0.692 0.135 0.016 0.046 Panel B: Individual treatment arms Index—Advice to mystery shopper (1) (2) (3) (4) Secular treatment = 1 0.081∗∗ 0.077∗ 0.076 0.136∗∗ (0.035) (0.042) (0.069) (0.066) Subgroup = 1 0.103∗ −0.028 0.001 0.033 (0.060) (0.049) (0.052) (0.049) Secular treatment = 1 × subgroup = −0.009 −0.023 0.004 −0.067 1 (0.081) (0.070) (0.077) (0.076) Secular + religious treatment = 1 0.074∗∗ 0.045 0.021 0.073 (0.035) (0.042) (0.072) (0.059) Secular + religious treatment = 1 × −0.092 0.043 0.041 −0.014 subgroup = 1 (0.082) (0.068) (0.080) (0.070) Observations 558 502 593 555 Subgroup Gov’t trusted Governing party Attended madrasa Ulema trusted source source constituency p-value: Total effect of secular T on 0.324 0.335 0.020 0.060 subgroup = 0 p-value: Total effect of secular + 0.804 0.103 0.072 0.121 religious T on subgroup = 0 Control mean Y | subgroup = 0 0.360 0.393 0.372 0.352 Control mean Y | subgroup = 1 0.463 0.365 0.373 0.385 Proportion of sample in subgroup 0.188 0.388 0.809 0.733 Note: Sample 1 (treatment and control arms, respondents who answered baseline survey and mystery-shopper call). Sample size varies from table 3 and across columns because of non-response to baseline questions due to some respondents hanging up before completing the baseline call (columns 1, 3, and 4), and insufficient geographical information to map some respondents to their political constituency (column 2). Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. S2. Additional Information on Study Design S2.1. Treatment Scripts The full SurveyCTO version of baseline and treatment call including Urdu is available at https://www.dr opbox.com/s/moatri1ymc437at/Imam%20Calls%20COVID-19_scale_up_first_round.xlsx?dl=0. The full SurveyCTO version of the Mystery Shopper call including Urdu is available at https://www.dropbox.com/s/h2y9bk8fmhfnae4/Imam%20Calls%20COVID-19_mystery_shopper_s caleup_first_round.xlsx?dl=0. S2.1.1. Secular Information: All Treatment Arms r According to leading health organizations like the World Health Organization and Pakistan Medical Association, when someone coughs, sneezes, or speaks they spray small liquid droplets from their nose Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 or mouth which may contain virus. This makes coronavirus much more contagious than the flu since you can catch the disease from an infected person even if you are standing a few steps away from them. r Additionally, approximately half the people with Coronavirus don’t show any sign/symptom, so they can spread it to people around them without knowing it. r Elderly people who catch the Coronavirus have a much higher chance of getting seriously ill and dying as compared to normal flu. r Virus can also be spread when someone touches something—so someone who is sick and doesn’t show it, can pass on the virus by touching the door, the wuzu water tap, and other places in the building. r Because the disease is difficult to detect since many infected people do not show any signs, and there are many elderly and vulnerable people around us who are at high risk of catching it, doctors recommend that one must protect oneself and each other by wearing a mask in public, keeping hands clean, and avoiding close and unnecessary physical contact (such as shaking hands). r Did you know the Government of Pakistan [COMBINED T ONLY: AND leading ulema] have jointly endorsed a statement that to avoid spread of COVID: - Elderly must stay at home - People must stay spread apart 6 feet - No wuzu (religious ablution) in the mosque - People must wear masks - Remove the jah namaz/floor mat - Wash floors between prayers r Do you feel that these points are relevant for your decisions about planning for your mosque activities in the coming weeks? Why or why not? r I would like to send you this information via WhatsApp or SMS. Which medium would you prefer? r As you are an important leader and an example for your community, people look to you as an example and guide on how to face this issue. r I would like to request you to consider helping protect your community by taking the following four steps: - Wear a mask as an example to others. - Announce in the khutbah each day that people should wear masks when they come to prayer. - Announce the importance of wearing masks over the mosque loudspeaker to the community. - Ask people to do wuzu at home and bring their own jah namaz to mosque. r Would you consider doing this? S2.1.2. Additional Text Included in Religious Arm Treatment Script r Islamic scholars have referenced the following hadiths about avoiding the spread of disease, when dis- cussing how the ummah should respond to Coronavirus threat. r Abu Salamah reported: The Messenger of Allah, peace and blessings be upon him, said, “There is no infection (without the decree of Allah). Do not mix the sick with the healthy.” Source: Sahih Muslim 2221 b r Because so many people have the illness without showing any sign, the recommendation is to wear a mask to protect others. r Do you feel this hadith is relevant to the situation with Coronavirus? r The Islamic affairs minister in Saudi Arabia has issued strict guidelines about social distancing, wearing masks and performing ablution at home for masjids. r Do you think these guidelines are helpful for you in planning for congregational prayers at mosque? Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 r Leading muftis at Al Azhar University in Egypt have provided the following guidelines on what mosques should do to prevent the spread of coronavirus: “The real danger lies in the rapid spread of the virus, and that gatherings are a direct cause of infection. The Islamic Shariah is keen on taking care of the interests of people both in this world and in the Hereafter, and on warding any harm off them. The major objective of the Shariah is to preserve people’s lives and protect them from all dangers. The pandemic that has hit the whole of humanity constitutes a necessity that leads to using some juristic concessions in the performance of some Islamic obligations, in order to avoid the danger of the further spread of the virus outbreak.” r Do you think these guidelines are helpful for you in planning for congregational prayers at mosque? r There are similar fatwas from Ayatollah Sistani and Ayatollah Khameini, do you want to hear it?7 Grand Ayatollah Sayyid Ali Sistani in Iraq has issued a fatwa stating that it is the duty of the believers to follow the health guidelines. He also stated that when health guidelines are ignored where there is fear of becoming infected with the virus and a serious possibility of death or significant harm if the person were to catch the disease, then not adhering to guidelines would be sinful. r Do you think these guidelines are helpful for you in planning for congregational prayers at mosque? r Similarly Ayatollah Khameini in Iran announced: “Surely, everything that leads to society’s health and everything that helps prevent the disease from becoming rampant is a good deed and on the contrary, everything that helps the disease become widespread is a sin. Allah the Exalted has made us responsible towards our own health and that of others and the health of the people. So, the first word of advice is that we should consider it our responsibility to completely observe the instructions specified by officials and to act on them r Do you think these guidelines are helpful for you in planning for congregational prayers at mosque? S2.1.3. Mystery-Shopper Script r Assalam-o-alaikum. My father and I wanted to attend the Eid prayers? Could you please tell me when are they being held? r Do we have to bring our own jah namaz? r Do we have to do wuzu at home? r Do we have to wear a mask? r ENUMERATOR: Did the respondent mention it is COMPULSORY to wear a mask in order to come? r Thanks so much for your time, goodbye. S3. Ethics Duke University Campus IRB approved this study under protocol number 2020-0432.8 In this appendix, we discuss ethical issues related to the experiment, building on discussions of ethics in field experiments with elites by McClendon (2012), Desposato (2015), Naurin and Öhberg (2021). Participants in the treatment and control groups were consented at the beginning of the baseline survey and were informed of the duration of the survey, as well as the fact that they would not be compensated for their participation. The protocol included the use of mystery shoppers (also known as an “audit study” approach), an approach used extensively in recent development economics research to observe behavior of agents in 7 Shia religious authorities; Shias are a religious minority in Pakistan, thus respondents are prompted to check their interest in this fatwa. 8 At the time of the inception of this study, the Center for Economic Research in Pakistan did not have a local IRB, and to the researchers’ knowledge other institutions in Pakistan did not accept external applications for IRB review. contexts where self-reports are expected to be affected by social desirability or other forms of bias, par- ticularly in the fields of public services and the behavior of public officials (Gaddis 2018; Banerjee et al. 2021; Anagol, Cole, and Sarkar 2017; Gine and Mazer 2016; Bertrand et al. 2007; Dizon-Ross, Dupas, Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 and Robinson 2017; Hetzel et al. 2008; Mohanan et al. 2015; Currie, Lin, and Zhang 2011; Friedman, Woodman, and Chatterji 2015; Fitzpatrick 2021; Bate, Mooney, and Hess 2010; Bate, Jin, and Mathur 2011; Das et al. 2012; Field and Vyborny 2020; Broockman 2013, 2014; McClendon 2016; Kalla and Broockman 2016; Mendez and Grose 2018; Driscoll et al. 2018; Gaikwad and Nellis 2021). Some de- ception is inherent in this methodological approach. In this scenario, it is necessary to use deception in order to observe their advice to congregants in a real-life situation. Respondents will not be punished or reported for their answers. The mystery-shopper data collected is kept confidential and only used to analyze the impact of the intervention on how they advise the community. Key considerations in the decisions on compensation and deception included the following: r The limited amount of time required from respondents. The mystery-shopper call asks basic, simple information that other congregants would commonly ask an imam, so this call will be well within the normal scope of activities for respondents. The total time for each call is approximately 2 minutes, so this instrument does not represent an undue burden on their time. The total time taken for respondents who participated in the baseline survey and were called in the mystery-shopper exercise was approximately 10 minutes. r The non-sensitive nature of the questions asked. Mosque activities during communal prayers (prayers, use of masks, social distancing, etc.) are public and could be observed by any casual observer in the area by stepping into the mosque. Hence the mystery shoppers are not collecting any sensitive or private information. r The influential role of respondents as community leaders, such that they would not be considered marginalized or vulnerable. r The ex ante most likely anticipated impact of treatment (enhancing masking and other protective mea- sures in the mosque) would have potential benefits to congregants, while the control group would simply continue with the status quo, and neither respondents nor their congregants would experience any harm as a result of the experiment. The mystery-shopper script involves the caller further asking whether a mask is really required even when the imam initially says he should wear it; however, we anticipated no substantive risk of discouraging mask use because of this single query, given widespread non-compliance which was already documented as the norm (Pattan Development Organization 2020). For the super-control group only, participants received only the 2 minute mystery-shopper call and no other instruments. Therefore, they would not be briefed on the study, making an active consent process impossible for this subgroup only. This was deemed to be acceptable based on the following criteria: r Criterion 1: The research involves no more than minimal risk to the subjects. There is no intervention with the control group participants; the only interaction is the collection of a few simple variables on their current practices in the mosque through the mystery-shopper instrument. We do not anticipate any risk to these subjects. r Criterion 2: The waiver or alteration will not adversely affect the rights and welfare of the subjects. Mosque activities during communal prayers (prayers, use of masks, social distancing, etc.) are public and could be observed by any casual observer in the area by stepping into the mosque. Hence we are not collecting any sensitive or private information. The mystery-shopper call is the only research activity with these participants and presents an extremely minimal burden (a 2 minute call with only simple information frequently asked from imams and also readily available to any observer in the area). Thus inclusion in the study without briefing would not lead to any adverse effects or negatively affect the welfare of the subjects. Table S4.1. Robustness to Alternative Construction of Index Dependent variables: Alternative constructions of Mystery Shopper index Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024 (1) (2) (3) Any treatment 0.0659∗∗ 0.0629∗∗ 0.0623∗∗ (0.0263) (0.0269) (0.0244) Observations 609 621 617 Variables excluded from index None Prayer mat Prayer mat; mask required Note: Robust standard errors in parentheses. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. r Criterion 3: The research could not practicably be carried out without the waiver or alteration. The goal of the super-control group is to detect the effect of being surveyed. Thus briefing participants on the study would make it impossible to carry out this component effectively. S4. Variations from Study Pre-registry The study was pre-registered (AEARCTR-0005740, Version 2.0). Registry Version 1.0 was registered during early-stage piloting. Piloting was complete, the design was revised, and then Version 2.0 was registered before the main study fieldwork. Observations collected during piloting, before Version 2.0 was registered, are not included in the analysis presented in the paper. This appendix discusses variations between the study pre-registry and the analysis presented in the paper. Variable Included in Instrument but Omitted from Pre-registration in Error Due to an error in coordination between post-pilot questionnaire revisions and registration, the outcome “asks the respondent to bring his own prayer mat” was left out of the registry entry inadvertently. The full script of the instrument (supplementary online appendix S2) demonstrates that all the mystery- shopper outcome variables collected are presented in the paper. Our results are not dependent on the inclusion of this variable. The index results are robust to the exclusion of this variable (table S4.1). Stated intent questions At the end of the baseline call, after the persuasion script was complete for the treatment group, we also asked respondents in all treatment arms to report their planned steps to reduce the spread of COVID in the mosque. We listed these stated intent variables in the study registry. We consider these variables weaker evidence than the mystery-shopper data, because they are self-reported and therefore subject to response bias. In addition, because of variation in the length of the call between treatment and control, response rates to this module differ between treatment arms, causing potential sample selection bias in the estimates. Table S4.2 shows the results for the pre-specified variables from this module with and without Lee (2009) bounds; the results are consistent with the main results but imprecisely estimated after bounding. Table S4.2. Treatment Effects on Stated Intent Questions Asked at End of Call 1 Panel A: Basic estimates Number of Short Clean Remove Ablution at Answered steps sermon mosque Soap mats Elderly/sick Distancing Announcements home Mask Other step intent Qs (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Any treatment 0.1823 0.0238 0.0021 −0.0108 0.0535 0.0051 −0.0422 0.0421 0.0750∗∗∗ 0.0568∗ −0.0285 −0.1212∗∗∗ (0.1162) (0.0207) (0.0312) (0.0302) (0.0351) (0.0263) (0.0372) (0.0258) (0.0289) (0.0305) (0.0345) (0.0261) Observations 670 670 670 670 670 670 670 670 670 670 670 829 Control group mean Y 2.0949 0.0751 0.2292 0.2016 0.2846 0.1423 0.3874 0.1304 0.1542 0.1660 0.3241 Panel B: Lee bounds Number of Short Clean Soap Remove Distancing Announce- Ablution at Mask Other step All rights reserved. For permissions, please e-mail: journals.permissions@oup.com steps sermon mosque mats Elderly/sick ments home (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Any treatment Lower −0.1325 0.0090 −0.0257 −0.0392 0.0086 −0.0281 −0.1033∗∗ 0.0048 0.0493 0.0356 −0.0780∗ (0.1439) (0.0242) (0.0380) (0.0358) (0.0415) (0.0309) (0.0448) (0.0307) (0.0339) (0.0346) (0.0425) Upper 0.6301∗∗∗ 0.0959∗∗∗ 0.1324∗∗∗ 0.1189∗∗∗ 0.1666∗∗∗ 0.1300∗∗∗ 0.0548 0.1559∗∗∗ 0.2074∗∗∗ 0.1936∗∗∗ 0.0801∗ (0.1458) (0.0144) (0.0466) (0.0457) (0.0480) (0.0438) (0.0480) (0.0178) (0.0456) (0.0457) (0.0477) Observations 829 829 829 829 829 829 829 829 829 829 829 Note: Panel A: Treatment and control arms, respondents who answered baseline survey. Robust standard errors in parentheses. Panel B shows Lee (2009) treatment effect bounds estimated using the Leebounds package developed by Tauchmann (2014). ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01. C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhae001/7678777 by LEGVP Law Library user on 01 August 2024