Policy Research Working Paper 9469 Building Trust in the State with Information Evidence from Urban Punjab Adnan Khan Sanval Nasim Mahvish Shaukat Andreas Stegmann Development Economics Development Research Group November 2020 Policy Research Working Paper 9469 Abstract Can communication designed to increase support for policy, perceptions of state capacity and trust in the state in government policy and shift perceptions of state capac- Pakistan. This holds true on average and across important ity redress deep-rooted mistrust in state institutions? This dimensions of heterogeneity after accounting for experi- paper finds providing information on past state effective- menter demand. This paper highlights the limits of using ness, highlighting citizens’ cooperation in enabling past information to build trust in state institutions, and the effectiveness or appealing to religious authorities’ support importance of measuring experimenter demand. for government policy have limited impact on support for This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at mshaukat@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Building Trust in the State with Information: Evidence from Urban Punjab∗ Adnan Khan Sanval Nasim Mahvish Shaukat LSE LUMS World Bank Andreas Stegmann University of Warwick JEL: D02, D83, C93, O12, O17, P16. Keywords: Trust, state capacity, information, COVID-19, experimenter demand. ∗ We would like to thank Sam Bazzi, Eli Berman, Migara De Silva, Ben Marx, Chris Roth and Jacob Shapiro for valuable suggestions. We would also like to thank seminar participants in the ESOC Lab at Princeton and the LUMS brownbag seminar for comments. We would like to thank Ahmad Mustafa, Jazib Parvez, Sardar Abdullah Mahmood, Omer Qasim, Haaris Mahmood and CERP Survey Unit for outstanding research assistance in Lahore, and Anna Lane, Iana Gerina and Youpeng Zhang for excellent support in Bonn. We would also like to thank Asim Khwaja and Ben Olken for generously sharing data. We gratefully acknowledge financial support from the briq Institute and the World Bank Development Research Group. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Adnan Khan; email: A.Q.Khan@lse.ac.uk. Sanval Nasim; email: sanval.nasim@lums.edu.pk. Mahvish Shaukat; email: mshaukat@worldbank.org. Andreas Stegmann; email: Andreas.Stegmann@warwick.ac.uk. 1 Introduction Citizens in developing countries often distrust state institutions and their ability to provide public services. When citizens distrust public institutions, they may fail to cooperate with the state – unless the state coerces them to cooperate (Besley, 2020a; Levi, 1988). For example, distrustful citizens might avoid paying taxes for public goods or settling disputes through the judicial system – actions that entail individual, immediate costs in exchange for public, long-term benefits. Increasing public trust becomes ever more critical in times of crises such as pandemics or wars when states must act swiftly and rely on its citizens to cooperate with potentially costly state measures. The Covid-19 pandemic provides a rare opportunity to study the linkages between citizen cooperation, perceptions of state capacity, and trust in the state as a crisis unfolds. Contain- ing the outbreak depends crucially on whether citizens voluntarily comply with government directives that limit civil liberties and thus on whether citizens trust the government and the state. While it is too early to judge whether the Covid-19 pandemic will become a pivotal moment on the scale of a war, Besley (2020b) argues that events similar to the pandemic led to substantial social and economic changes in the past. Importantly, the pandemic un- derscores trust in state institutions as a key outcome at stake. Governments’ management of the crisis will likely carry a lasting impact on this dimension of trust.1 Communication offers an important link between public trust in the state and govern- ments’ pandemic response. In this paper, we provide evidence on the extent to which com- munication designed to (i) encourage support for government policy and (ii) improve per- ceptions of state capacity can redress deep-rooted mistrust in elected governments and state institutions. Estimating the causal link between such communication and attitudes towards the state from observational data poses challenges as governments strategically adjust their communication in response to public opinion. To address our research questions, we leverage randomly assigned informational treat- ments shared with Pakistani citizens through a phone survey. We conducted our experiment during the early stages of the pandemic (in May and June 2020) with 5,771 respondents in Lahore and Faisalabad, the two most-populous cities in Pakistan’s Punjab province. Pak- istan provides an excellent setting to explore our research questions since many Pakistanis distrust the state and increasingly rely on non-state actors for basic services (Cheema et al., 2017, Acemoglu et al., 2020). We designed the informational treatments to increase citizens’ support for the govern- 1 For instance, see Daron Acemoglu’s presentation on “Unkowns, Challenges and Opportunities in the Time of Covid-19” (April 9, 2020) on the Royal Economic Society Covid-19 Hub. 2 ment’s Covid-19 policies and improve how citizens perceive the state’s capacity to implement these policies. We focus on two types of information: (i) information about past state suc- cess in managing crises, and (ii) information about non-state actors’ support for state policy. Governments frequently deploy both strategies to spur citizens to support their policies or build citizens’ trust in the state. All respondents received basic information about the government’s directives as part of the survey. We provided additional information to respondents assigned to a treatment group. Referring to truthful information about how the government successfully managed a large-scale dengue fever outbreak in 2011, the first treatment informs respondents that the government selected the right policies to address a public health emergency in the past, and that it possessed the capacity to successfully implement these policies. The second treatment complements the first treatment. It emphasizes that, according to experts’ assessments, the government successfully tackled the dengue outbreak because, in addition to the government identifying and undertaking appropriate actions, citizens sub- stantially cooperated with the government’s policies. The third treatment represents an appeal to religious authority, a common strategy in Islamic countries’ politics. The treatment provides information that religious authorities support the government’s policy. We informed respondents about a fatwa issued by Al- Azhar University in Egypt – considered one of the highest authorities on Sunni Islamic thought by many Muslims – permitting a ban on congregational prayers and urging citizens to hold confidence in their respective government’s Covid-19 policy. Given concerns about potential experimenter demand effects in phone surveys, we include an additional treatment group that allows us to isolate experimenter demand effects from estimated treatment effects. This condition precludes misleading interpretations of estimated treatment effects. We collected rich demographic data (e.g. on partisanship, belief in conspiracy theories, religiosity and economic exposure to the Covid-19 crisis) before sharing any information with the respondents. After providing respondents the respective informational treatments, we collected data on three sets of outcomes: support for the government’s Covid-19 related directives and policies, perceptions of state capacity, and trust in the state. We specified all outcome measures and the variables used in the heterogeneity analysis in a pre-analysis plan. We find that, on average, the information treatments have little effect on support for government policy. Informing respondents that the state succeeded in managing a health crisis in the past does not affect their attitudes towards the government’s current directives on hand-washing, social distancing, and avoiding congregational prayers. Stressing that 3 the state succeeded because citizens cooperated with its directives makes no difference to respondents’ attitudes either. While informing respondents that religious leaders endorse the government’s policy appears to matter at first glance, the estimated treatment effects are not robust to accounting for potential experimenter demand effects. Compared to the control group, respondents in this treatment group are more likely to say they intend to avoid congregational prayers in the future and more likely to say they believe doing so can help minimize the spread of coronavirus. Yet, we observe only small (3.5% of the control mean) and marginally significant average treatment effects. Once we compare respondents in this treatment group to respondents that were assigned to the experimenter demand treatment, we find that the small positive treatment effects are not significantly different from the demand effects induced by the experimenter demand condition, despite the fact that the treatment includes an explicit endorsement of the directive.2 These results stand in contrast to Banerjee et al. (2020) which reports relatively large effects on support for Covid-19 related public health directives in India. In fact, our confidence intervals allow us to rule out the effect sizes documented by Banerjee et al. (2020).3 Similarly, we find no treatment effects on perceptions of state capacity – which we measure as the state’s capacity to manage the ongoing pandemic, provide public goods, and enforce regulations. The treatments have little effect on these outcomes on average. If anything, some of the treatments may have caused a small decline in perceived state capacity to enforce regulations. The state positive and religious authority treatments lower perceptions by 3.6% and 5.1% of the control mean, respectively. The first effect suggests that highlighting the state’s competence in the past can undermine perceptions of the state’s capacity in the present. Stressing the role of citizens in increasing state capacity may mitigate this impact since we do not detect a similar effect in the citizen cooperation treatment. The second effect suggests that highlighting the state’s reliance on external actors can undermine perceptions of the state’s own capacity. 2 These results add to the literature on the links between Islam and economic and political performance (Kuran, 2018). A subset of this literature investigates the role of Islam as a source of political legitimacy (see Platteau (2017), Rubin (2017) and Cosgel, Miceli and Rubin (2012) for a historical perspective). Our paper contributes to this literature by providing experimental evidence on the effects of communicating religious authorities’ explicit support for government policies on citizens’ compliance with state directives, perceived state capacity and trust in state institutions. 3 Importantly, there are some key differences between the treatment messages of both studies: Banerjee et al. (2020) share a video in which Abhijit Banerjee, 2019 Nobel laureate in Economics, widely respected as a public intellectual in India, encourages individuals to report symptoms to local health workers and discusses health-preserving behaviors. In contrast, the information in our experiment was provided over the phone, referencing the Pakistani state, but ultimately on behalf of researchers at CERP and LUMS. Moreover, in a context of significant distrust in state institutions, the treatment scripts sought to leverage the Pakistani state’s past successes and contemporaneous support for the Pakistani government by non-state, religious actors. 4 Next, we turn to the estimated effects on trust in the state. Since respondents may assess the elected government differently than the bureaucracy, we elicit trust in these two components separately. We also implement a lab-in-field game to measure trust in the state as elicitations may not fully reflect “real” behavior. In this game, respondents allocate Rs. 200,000 to either a government or non-state charity fund to support Covid-19 relief efforts in Pakistan. We use the share of funds allocated to the government as a behavioral measure of trust in the state. Whatever measure we use, we estimate no increase in trust in the state. If anything, we find the state positive treatment has a small but negative effect on trust. Following our pre-analysis plan, we also investigate whether the treatments had hetero- geneous effects along a large number of dimensions to ensure that these average effects do not mask large responses by certain subgroups. We observe no significant heterogeneity by education, economic exposure to the Covid-19 pandemic, religiosity or belief in conspiracy theories on the Covid-19 pandemic. However, we document significant heterogeneity by partisanship (measured by past vote choice and present media consumption). Respondents who lean more favorably towards the ruling party lend more support to government policy after receiving the information treatments. This effect is particularly strong for the citizen cooperation treatment, where pro-ruling party respondents increase support for government directives on hand-washing, social distancing, and avoiding congregational prayers by 0.1 to 0.15 standard deviations. However, here is where the experimenter demand condition becomes relevant: we find that these heterogeneity patterns are not robust to accounting for potential demand effects when we compare the treatment groups to the experiment demand group. Our results provide a caveat to findings from Acemoglu et al.’s (2019) closely related study which shows that informing citizens of reduced delays in state courts increases expected usage of, allocations in high-stakes lab-in-the-field games to, and trust in state courts. State posi- tive information about a specific policy may increase trust in a specific institution. However, our results suggest that this finding may not necessarily generalize to other policy domains or towards generalized measures of trust in the state. This has important implications for our understanding of the limitations faced by governments engaging in attempts to foster trust in state institutions. Our paper contributes to a body of contemporaneous work studying how different forms of communication shape public perceptions about Covid-19 (Allcott et al., 2020; Ajzenman et al., 2020; Banerjee et al., 2020; Barrios and Hochberg, 2020; Bursztyn et al., 2020; Grossman et al., 2020; Rafkin et al., 2020). These studies focus on how people update different types of beliefs when informed about Covid-19. Most related to our work is a study by Rafkin et al. (2020) which explores the effects of highlighting government inconsistency on trust in the 5 US government. In contrast, within a developing country context, we document the effects of informational treatments that emphasize past government success and support from non- state actors on a wide range of beliefs and attitudes about the state, including perceptions of state capacity and trust in elected and non-elected state officials. Importantly, while applied in the context of the Covid-19 pandemic, our informational treatments represent more general communication strategies deployed by governments in a variety of scenarios. More generally, our results speak to a growing literature on the role of citizens’ trust in state institutions in shaping state capacity.4 Aside from focusing on a high stakes context offering an opportunity to elicit direct measures of citizens’ compliance, we also expand the types of informational treatments used to manipulate attitudes towards the state. The fact that we estimate null treatment effects does not render our results less important, especially since our study design was adequately powered to detect effects on self-reported behavior and post-treatment survey questions indicate that a large share of the respondents in all treatment arms retained the content of the informational treatments. Our findings thus highlight limitations to efforts to shape trust in state institutions through informational campaigns. 2 Experimental Design 2.1 Sample We conducted our survey experiment with 5,771 residents of Lahore and Faisalabad over five weeks in May and June 2020.5 The number of new coronavirus cases in Punjab peaked during this time, rising from approximately 1,500 new cases each day at the start of the experiment to 4,000 at the end.6 We recruited individuals from a pool of 15,000 phone numbers, col- lected in an ongoing study on property tax and public goods provision (Khan et al., 2020). The phone numbers were randomly ordered and assigned to control or treatment groups. Surveyors called each phone number in the assigned order. If the respondent answered and consented, the surveyor initiated the survey. 47% of the calls were not answered and 21% of 4 In economics, this literature builds on the literature studying the role of civic capital in influencing economic development (Guiso et al. (2016, 2011, 2008, 2004), Knack and Kneefer (1997), Putnam et al. (1994)). The importance of citizens’ cooperation for the development of state capacity has been highlighted theoretically in models of states that derive authority from citizens who have the capability to rein them back (Acemoglu and Robinson, 2019). Relatedly, Dell et al. (2018) argue that, historically, the greater state capacity of the North Vietnamese state relative to that of the South Vietnamese state is related to the greater cooperation of citizens with the state which in turn could depend on citizens’ trust in the state. 5 Our study protocols were reviewed and approved by the Institutional Review Board of the Lahore University of Management Sciences (Protocol Number: LUMS-IRB/004202020SN). 6 The number of new cases started to decline after our experiment. 6 respondents (who did answer) did not consent. Neither of these rates are differential across treatment arms.7 Residents were selected from a variety of localities within Lahore and Faisalabad - high density, low density, poor, rich. The sample is therefore representative of urban Punjab, but with two qualifications. First, all individuals in our sample own or rent a taxable property. This restriction is imposed by Khan et al. (2020), which we rely on to build our sample. Our sample therefore excludes individuals who live in informal settlements and individuals who live in private housing societies, the two extremes of the income distribution. Second, the individuals in our sample are almost entirely (96%) male. Men are overrepresented because they were more likely to participate in the Khan et al. (2020) surveys and, conditional on participation, more likely to share their phone numbers. Appendix Table A1 presents additional descriptive statistics. Individuals have an average age of 37.3 years and 72% have completed secondary education. We did not measure income because of respondents’ reluctance to provide this information over the phone. However, there is considerable heterogeneity in respondents’ ability to smooth income while following state directives on minimizing the spread of Covid-19: 57.9% of respondents report com- pliance with state directives would result in at least a 50% loss in weekly income, while 19.1% of respondents report a loss of at most 20% of their weekly income. There is also heterogeneity in compliance with state directives: on average, the average response to past compliance with the directive to avoid congregational prayers is just 0.51 (on a scale from 0 to 1 – meaning the respondent declared not to have followed the directive at all or to have completely followed the directive, respectively). At 13.4%, there is also a sizeable share of the respondents who believe in some version of a conspiracy theory centered around the allegation that the Covid-19 pandemic has non-natural origins. Finally, our sample reflects political divisions in Punjab and Pakistan more generally. 44% of respondents report voting for the ruling Pakistan Tehreek-e-Insaaf (PTI) in recent elections, while 29.5% of respondents report voting for the opposition Pakistan Muslim League - N (PML-N). Political preferences are also reflected in media consumption: 36.5% of the former group watch ARY News, which tends to favor the ruling party, while 51.8% of the latter group watch GEO News, which tends to oppose the ruling party. 2.2 Treatments All respondents received an overview of the Punjab government’s guidelines on minimizing the spread of Covid-19. These guidelines changed over time. In the first half of the ex- 7 Results available upon request. 7 periment, guidelines included washing hands, wearing a mask, maintaining a distance from others when outside, and praying at home instead of at a mosque. In the second half, the government removed the recommendation to avoid congregational prayers, emphasizing so- cial distancing in general instead. The overview reflected whatever guidelines were in place at the time of the survey. The script is shown in section A.1 in the Appendix. Respondents were likely aware of at least some of these guidelines through other channels prior to the survey experiment. We collected data on all Twitter posts made by two major media groups, ARY and GEO, in the 5 weeks prior to and throughout the duration of our experiment.8 This data shows that 31% of tweeted articles focused on the coronavirus pandemic. Of these articles, 23% mentioned at least one of the guidelines. The overview therefore reinforced already available information. Respondents assigned to a treatment group received an additional message designed to increase support for these guidelines and improve assessments of the state’s capacity to manage the pandemic. Each treatment is described in detail below.9 2.2.1 Past state success Reminding respondents of the state’s success in managing a crisis in the past may change their views about the state’s ability to manage a crisis today. Respondents may not otherwise remember past successes or even if they do remember, may not link them to similar events in the present (Acemoglu et al., 2019). In this treatment, we remind respondents of the Punjab government’s response to a dengue outbreak in 2011. Post hoc, this response was considered by several experts to have been instrumental in containing the outbreak (Bhatti et al., 2015). A World Bank case study, for example, argued government measures such as closing public places for fumigation and testing slowed the spread of dengue and prevented a similar epidemic from happening the following year (World Bank, 2018). These reports draw on analysis generated by the Punjab Information Technology Board, an autonomous body set by the Punjab government. We shared an overview of this report with respondents in this treatment group. The script for this and all other treatments are shown in Appendix A. 8 Both media groups jointly published 8766 tweets over this time period. The tweets always link to an article published on the media groups’ main website. We used 3 independent human coders to identify mentions of government directives in a random sample of 55% of the articles linked in a tweet that included the following terms: covid, coronavirus, virus, influenza, flu. 9 We elicited up to three questions about the content of the information treatments at the end of the survey. In Appendix Figure A1, we document that a large share of the respondents assigned to a treatment condition retained the information provided as part of the treatment. 8 2.2.2 Past state success due to citizen cooperation A state’s ability to implement policy often requires active citizen cooperation. This is es- pecially true for public health policy. In this treatment arm, we provide respondents with the information that citizens cooperated in the past. In response to this information, re- spondents may change their perceptions of the associated public benefit of cooperation and update their beliefs about others’ likelihood of cooperating today. In particular, respondents in this group receive the same information detailed above about the Punjab government’s success in handling the dengue outbreak – but with an emphasis on the importance of citizen cooperation in achieving this success. We share details from Rehman et al. (2016) showing how citizen hotlines helped the government identify dengue hotspots and target resources more effectively. 2.2.3 Religious authority Respondents may be more likely to support government guidelines if a credible and respected non-state actor also supports government guidelines. This may be particularly true when these guidelines are perceived to conflict with long-held practices, such as congregational prayers. In this case, a religious authority’s endorsement of the guideline to avoid congre- gational prayers may be enough to convince a respondent of its legitimacy. The Pakistan government, perhaps following this line of thinking, tried first to secure the support of re- ligious authorities within the country. Failing to do so, they requested an edict from the Grand Imam of of Al-Azhar University in Egypt, considered by some Muslims to be the highest authority in Sunni Islamic thought.10 This edict provided a religious justification for the suspension of Friday congregational prayers and, furthermore, declared unlawful any action undermining confidence in state protective measures. Respondents in this treatment group received an overview of the edict. Note that – while the edict explicitly encourages support for the government’s policy – the edict is less focused on information pertaining to perceptions of state capacity. 10 According to numerous news articles, the President of Pakistan requested this edict after leading religious authorities within the country refused to support the government’s restrictions on congregational prayers. For more information, please refer to the following news articles: The Dawn (2020). “Egypt’s Al-Azhar issues fatwa permitting Juma prayers’ suspension in Pakistan” The Dawn, March 26. https://www.dawn.com/news/1543801. (last accessed on Sep 1, 2020) The Nation (2020). “President Alvi’s request, Egypt’s Al-Azhar issues Fatwa permitting suspension of Friday prayers”, March 25. https://nation.com.pk/25-Mar-2020/president-alvi-s-request-egypt-s-al-azhar- issues-fatwa-permitting-suspension-of-friday-prayers. (last accessed on Sep 1, 2020). The fact that the President of Pakistan explicitly sought out the support of Al-Azhar University is a strong indicator that Pakistani state officials believe in the relevance of this type of messaging. More generally, similar actions by government actors around the world suggest a widely held belief that appeals to religious authorities are an effective strategy to achieve compliance and support for government policies. 9 2.2.4 Experimenter demand One concern with the treatments described above is that respondents may feel obligated to change their survey responses because of experimenter demand, rather than any real change in attitudes or intended behavior.11 We address this concern by priming respondents to react exactly in this way in a separate treatment group. After sharing the overview of government guidelines, the enumerator tells respondents in this treatment group that he or she thinks the guidelines are “a really good idea”. By comparing outcomes in the information treatments with this experimenter demand group, we can “net-out” any potential experimenter demand effects from estimated treatment effects. 2.3 Outcomes At the end of the survey experiment, we asked respondents three sets of questions. First, we asked respondents their intended compliance with state guidelines. Respondents were asked how likely they are to i) wash hands frequently, (ii) social distance, and (iii) avoid congregational prayers on a 5-point Likert scale. We also asked whether each guideline is beneficial, and if the respondent believes others should comply with the guideline. These questions are measured on a 5-point Likert scale. Second, we measured respondents’ support for possible policy responses such as shutting down public spaces and suspending Friday prayers. This allows us to explore whether the information treatments directly affect policy preferences. Finally, we asked respondents their perception of state capacity and how much they trust the state. We elicited perceptions of the state’s capacity to manage the Covid-19 crisis and, more generally, to provide public services and enforce regulations. Because respondents may assess the elected government differently than the bureaucracy, we measured trust in these two components separately. As elicitations may not fully reflect “real” behavior, we also implemented a lab-in-field game to measure trust. In this game, respondents are asked to allocate Rs. 200,000 to either a government or non-state charity fund to support Covid-19 relief efforts in Pakistan12 . Respondents were told their allocations will be averaged, and the Rs. 200,000 donated to each fund according to this average. We use share of funds allocated to the government as an additional measure of trust in the state. Finally, we also elicited respondents’ demand for information on the government mandated behaviors by offering 11 Haaland, Roth and Wohlfart (2020) emphasize the importance of accounting for potential demand effects. 12 The non-state charity fund was established by the Edhi Foundation. EDHI is a secular and non-political foundation and is widely known as a provider of social welfare services in Pakistan. This implies that the share of funds allocated to the government measures trust in the state relative to a prominent, private provider of public services. 10 them to subscribe to a text-message service summarizing the latest government directives and official recommendations. 3 Results We estimate the average impact of the information treatments on respondents’ support for government policy, perceptions of state capacity and trust in the state using the following specification: Yi = β0 + β1 Si + β2 Ci + β3 Ri + δc(i) + θj (i) + ωt(i) + γXi + i , (1) where Yi is an outcome variable, Si , Ci , and Ri are dummy variables indicating whether respondent i was assigned to one of the three treatment groups (past state effectiveness, citizen cooperation, religious authority) or the control group, δc(i) are city fixed effects, θj (i) are enumerator fixed effects, and ωt(i) is a dummy variable indicating the period after the change in the wording of the public service announcement. Xi is a measure of self-reported, past behavior, an additional control variable that we include only when estimating treatment effects on attitudes towards the government’s Covid-19 related directives on hand-washing, social distancing and forgoing congregational prayers.13 Standard errors are adjusted to account for heteroscedasticity. The coefficients of interest, β1 , β2 , and β3 , measure the average difference in outcomes between treatment and control groups. We use a similar specification to estimate experimenter demand effects: Yi = γ0 + γ1 Si + γ2 Ci + γ3 Ri + δc(i) + θj (i) + ωt(i) + γXi + i , (2) where Si , Ci , and Ri are dummy variables indicating whether respondent i was assigned to one of the three treatment groups or the experimenter demand group, and all other variables are defined as before. The coefficients γ1 , γ2 and γ3 capture the extent to which respondents in each treatment group update their beliefs above and beyond respondents in the experimenter demand group, who received a clearly worded favorable opinion of the government. We use these estimates to net-out any potential experimenter demand effects from estimated treatment effects. 13 The empirical specifications were pre-specified in the AEA RCT registry (RCT-ID: AEARCTR-0005744). 11 3.1 Impact on attitudes towards state directives We first test the hypothesis that our information treatments increase support for the gov- ernment’s Covid-19-related policies and directives by estimating the average impact of the information treatments on respondents’ attitudes towards different state directives to mini- mize the spread of coronavirus. We focus on the following three state directives: frequent hand-washing, social distancing, and forgoing congregational prayers at the mosque. We measure attitudes by asking respon- dents their intended compliance with the directive, beliefs about whether others should com- ply with the directive, and beliefs about whether the directive is beneficial. Each outcome is measured on a 5-point Likert scale ranging from 0 to 1, with higher values indicating a more favorable attitude. Our primary outcomes are the z-score indices of the three measures, one for each recommended behavior. We use equation (1) to estimate average treatment effects, including respondents’ self-reported behavior in the last week as an additional control. Table 1 shows the information treatments have little to no impact on attitudes towards recommended behaviors on average. In Panel A, we find no statistically significant effects on compliance with frequent hand-washing (column 4). The point estimates for each treatment group are small and precise, ruling out positive effects larger than 0.05 standard deviations. Given high compliance at baseline (the control mean for intended compliance with the hand- washing directive is 0.93 (measured on a 0 to 1 scale), possibly due to social desirability bias), this result is not surprising. In Panel B, we find similarly small and insignificant effects on social distancing, though baseline compliance is lower at 0.83. Finally, Panel C shows the religious authority treatment had small, but marginally significant effects on respondents’ attitudes towards forgoing congregational prayers at the mosque. In the upper part of this panel we compare respondents in the religious authority treatment to respondents in the control group. Baseline compliance with this recommended behavior is only 0.54 on average, reflecting the state’s difficulty in imposing recommendations perceived to conflict with religious practices. The comparison suggests that the religious authority treatment improves attitudes by 0.045 standard deviations (column 4). This effect is not particularly large, however, and statistically indistinguishable from that of the state positive and citizen cooperation treatments. The confidence intervals for the state positive, citizen cooperation, and religious authority treatments rule out positive effects larger than 0.08, 0.06, and 0.1 standard deviations, respectively. In the bottom part of Panel C, we present the comparison to the experimenter demand group which accounts for potential demand effects. In this setup, the effect of the religious authority is reduced to 0.027 standard deviations and is no 12 longer statistically significant.14 In Appendix Table A2, we investigate whether the information treatments affected addi- tional, related beliefs. We document that the information treatments did not change beliefs about the effectiveness of different government policies such as the shutdown of public places (Column 1) and the suspension of Friday prayers (Column 2). In addition, the informational treatments did not change beliefs about social sanctions faced by individuals who decide not to comply with the state directives (Column 3). All of these outcomes are also mea- sured on a 5-point Likert scale ranging from 0 to 1, with higher values indicating a higher perceived effectiveness or probability to face social sanctions, respectively. All estimated average treatment effects are small and statistically insignificant. 3.2 Impact on perceptions of state capacity Next, we estimate the average impact of each information treatment on perceptions of state capacity. Table 2, Panel A reports estimated effects on perceptions of state capacity to pro- vide public goods, enforce regulations, and manage the coronavirus pandemic competently. Conceptually, we view perceptions of the capacity to enforce regulations as a measure of citizens’ perceptions of legal capacity, the reach of the state in establishing the rule of law.15 Similarly, we regard perceptions of the capacity to provide public services as a measure of perceptions of collective capacity, the extent to which states have made investments in the structures that are needed to provide public services. We thus elicit perceptions on two of the three main forms of state capacity identified by Besley and Persson (2009, 2011). Each outcome variable is measured on a 5-point Likert scale ranging from 0 to 1, with higher values indicating higher perceived state capacity. Our primary outcome is a z-score index of perceived state capacity to provide public goods and enforce regulations.16 This capacity index measures respondents’ perceptions of general state capacity. All of the treatments have a negligible impact on perceptions of state capacity to man- age the coronavirus epidemic (column 1). These null effects are precisely estimated: the confidence interval for each treatment rules out effects larger than 0.04 points on the Likert 14 More precisely, this specification benchmarks the informational treatment conditions against a short treatment designed to induce experimenter demand effects. The informational treatment conditions may have induced demand effects of a different magnitude (including no demand effects). Strictly speaking, this comparison is thus not “netting-out” potential demand effects, but provides a useful benchmarks on the estimated treatment effects. 15 Relatedly, we also elicited respondents’ beliefs about the likelihood that a citizen not complying with the Covid-19 related state directives would face sanctions from the state, an additional measure of perceived legal capacity. The information treatments do not alter these beliefs relative to the control group. These results are presented in Appendix Table A2, Column 4. 16 Primary outcomes are specified in the pre-analysis plan. 13 scale (8% of the control mean). The average impact of each treatment group on the capacity index is small, and for the most part, statistically insignificant (column 4). The one excep- tion is the religious authority treatment, which has a marginally significant negative effect on perceptions of state capacity. The point estimate is small, however, and the confidence interval rules out a negative effect larger than 0.13 standard deviations. 3.3 Impact on attitudes towards the state Table 2, Panel B reports estimated average treatment effects on attitudes towards the state. We measure attitudes towards the provincial government in three ways: trust in elected representatives, trust in civil servants and other government officials, and belief that the provincial government has helped the respondent in the last year. Each of these outcomes is measured on a 5-point Likert scale ranging from 0 to 1, with higher values indicating higher trust and perceived benevolence. Our first primary outcome in this panel is a z-score index of these variables. We also elicit the share of funds respondents prefer to allocate to the government’s coronavirus relief fund rather than a well-known charity. This behavioral measure of trust in the state is the second primary outcome in this panel. Finally, we offer respondents’ the option to subscribe to a newsletter summarizing the latest directives issued by the provincial government. We analyze respondent’s demand for this type of information as our second behavioral measure of interest.17 The average impacts of each treatment group on the trust index are all small and statis- tically insignificant (column 8). The point estimates range from −0.028 standard deviations (religious authority treatment) to 0.028 standard deviations (citizen cooperation treatment), though all three estimates are statistically indistinguishable at the 5% level. The confidence intervals of these estimates rule out effects larger than 0.09 standard deviations. We also find limited impact on the behavioral measure of trust in the state (column 9).18 The state positive treatment has a marginally significant negative effect on the share of funds allocated to the government, but the effect size is small (2.6 percentage points or 7% of the control mean) and the confidence interval rules out negative effects larger than 6 percentage points. Lastly, the treatments do not affect respondents’ demand for information about government- issued directives. The confidence intervals rule out positive effects larger than 4.9 percentage points (12.5% of the control mean). Overall, Table 2 shows that the information treatments neither improved perceptions 17 We view this measure as conceptually interesting, yet acknowledge that it is not only a measure of trust in the state. 18 We also document that respondents assigned to different treatment conditions are not differentially likely to refuse engaging in the fund allocation exercise. These results are available upon request. 14 of state capacity nor increased trust in the state. These results suggest that respondents’ perceptions of state inefficiency and distrust towards the state are deeply engrained and difficult to manipulate, even with strong appeals to credible state-positive information and religious authority. 3.4 Heterogeneous effects We also investigate whether the treatments had heterogeneous effects along a large number of dimensions to ensure that these average effects do not mask large responses by certain subgroups. In Figure 1, we study whether partisanship (measured using past vote choice) affect who responds to the treatment and who does not. We plot the point estimates and the corresponding 95% confidence intervals for all treatment indicators and the respective interaction terms. The point estimates in blue correspond to the estimates derived from a comparison to the control group, while the point estimates in red are obtained from a setup that compares respondents in the different treatment groups to respondents in the experimenter demand condition. The different subfigures repeat this exercise for each of the primary outcomes pre-specified in the pre-analysis plan. While the comparison to the control group (estimates in blue) in subfigures a) to c) seems to suggest that supporters of the ruling PTI party update their beliefs about government policy positively and opponents either retain their original beliefs or update negatively, the comparison to the experimenter demand group (estimates in red) confirms that this pattern is not robust to accounting for potential demand effects. We also explored education, economic exposure to the Covid-19 pandemic, religiosity, be- lief in conspiracy theories related to the Covid-19 pandemic or partisanship based on present media consumption as additional dimensions. None of these other investigated dimensions provide evidence of heterogeneous treatment effects.19 4 Conclusion This paper provides experimental evidence on the effects of informational treatments de- signed to increase support for Covid-19-related state directives on citizens’ attitudes towards these directives, perceptions of state capacity and trust in state institutions. To study these effects, we contacted 5,771 citizens living in Lahore and Faisalabad, the two most popu- lous urban centers in Pakistan’s eastern province of Punjab. We find that, on average, the 19 We specified these dimensions in our pre-analysis plan. For brevity, we do not include a full depiction of these results here. However, the corresponding figures for these additional dimensions of heterogeneity are available upon request. 15 information treatments have little effect on support for government policy, perceptions of state capacity or trust in state institutions. Moreover, we rule out heterogeneous treatment effects along a series of important dimensions. These findings have important implications for our understanding of the limitations faced by governments seeking to build trust in state institutions. 16 REFERENCES Acemoglu, Daron, Ali Cheema, Asim I. Khwaja and James A. Robinson (2020). “Trust in State and Nonstate Actors: Evidence from Dispute Resolution in Pak- istan.” Journal of Political Economy, 128 (8), 3090-3147. Acemoglu, Daron and James A. Robinson (2017). “The emergence of weak, despotic and inclusive states.” NBER working paper, 23657. Allcott, Hunt, Levi Boxell, Jacob Conway, Matthew Gentzkow, Michael Thaler and David Yang (2020). “Polarization and Public Health: Partisan Differences in Social Distancing during the Coronavirus Pandemic.” Journal of Public Economics, forthcoming. Ajzenman, Nicolas, Tiago Cavalcanti and Daniel Da Mata (2020. “More Than Words: Leaders’ Speech and Risky Behavior during a Pandemic.” SSRN working paper, 3582908. Banerjee, Abhijit, Marcella Alsan, Emily Breza, Arun G. Chandrasekhar, Abhijit Chowdhury, Esther Duflo, Paul Goldsmith-Pinkham and Benjamin A. Olken (2020). “Messages on COVID-19 Prevention in India Increased Symptoms Reporting and Adherence to Preventive Behaviors Among 25 Million Recipients with Similar Effects on Non-recipient Members of Their Communities.” NBER working paper, 27496. Barrios, John M. and Yael Hochberg (2020). “Risk Perception through the Lens of Politics in the Time of the COVID-19 Pandemic.” NBER working paper, 27008. Besley, Timothy (2020a). “State Capacity, Reciprocity, and the Social Contract.” Econometrica, 88, 4, 1307-1335. Besley, Timothy (2020b). “The Covid-19 Pandemic through the Lens of Political Economy.” mimeo. Besley, Timothy and Torsten Persson (2009). “The Origins of State Capacity: Property Rights, Taxation and Politics.” American Economic Review, 99, 1218-1244. Besley, Timothy and Torsten Persson (2011). “Pillars of Prosperity: The Political Economics of Development Clusters.” Princeton: Princeton University Press. Bhatti, Zubair, Jody Kusek and Tony Verheijen (2015). “Logged On. Smart Government Solutions from South Asia.” Washington: World Bank Group. Bursztyn, Leonardo, Aakaash Rao, Christopher Roth and David Yanagizawa- Drott (2020). “Misinformation during a Pandemic.” Working paper. Cheema, Ali, Zulfiqar Hameed and Jacob N. Shapiro (2017). “Victimization, Citizen Engagement, And Policing in Lahore.” Institute of Development and Economic Alternatives, Policy Report, Lahore, Pakistan. 17 Cosgel, Metin M., Thomas J. Miceli, and Jared Rubin (2012). “The Political Economy of Mass Printing: Legitimacy and Technological Change in the Ottoman Empire.” Journal of Comparative Economics, 40 (August): 357-71. Dell, Melissa, Nathan Lane, and Pablo Querubin (2018). “The Historical State, Local Collective Action, and Economic Development in Vietnam.” Econometrica 86, no. 6: 2083-2121. Grossmann, Guy, Soojong Kim, Jonah M. Rexer and Harsha Thirumurthy (2020). “Political partisanship influences behavioral responses to governors? recommenda- tions for COVID-19 prevention in the United States.” Proceedings of the National Academy of Sciences, forthcoming. Guiso, Luigi, Paola Sapienza and Luigi Zingales (2004). “The Role of Social Capital in Financial Development.” The American Economic Review, 94, 526-556. Guiso, Luigi, Paola Sapienza and Luigi Zingales (2008). “Alfred Marshall Lec- ture: Social Capital as Good Culture.” Journal of the European Economic Association, 6, 295-320. Guiso, Luigi, Paola Sapienza and Luigi Zingales (2011). “Civic Capital as the Missing Link.” Handbook of Social Economics, 1, 417-480. Guiso, Luigi, Paola Sapienza and Luigi Zingales (2016). “Long-Term Persis- tence.” Journal of the European Economic Association, 14: 1401-1436. Haaland, Ingar, Chris Roth and Johannes Wohlfart (2020). “Designing Infor- mation Provision Experiments.” CAGE working paper, 484. Khan, Adnan, Asim Khawja, Benjamin Olken and Mahvish Shaukat (2020). “Rebuilding the Social Compact: Urban Service Delivery and Property Taxes in Pakistan.” AEA RCT Registry (RCT-ID: AEARCTR-0003270) Knack, Stephen. and Philip Keefer (1997). “Does Social Capital Have an Economic Payoff? A Cross-Country Investigation.” The Quarterly Journal of Economics, 112, 1251- 1288. Kuran, Timur (2018). “Islam and Economic Performance: Historical and Contempo- rary Links.” Journal of Economic Literature, 56 (4): 1292-1359. Levi, Margaret (1988). “Of Rule and Revenue”, Berkeley: University of California Press Platteau, Jean-Philippe (2017). “Islam Instrumentalized: Religion and Politics in Historical Perspective.” New York: Cambridge University Press. Putnam, Robert D., Robert Leonardi, and Raffaela Y. Nanetti (1994). “Mak- ing democracy work: Civic traditions in modern Italy”, Princeton: Princeton University Press. 18 Rafkin, Charlie, Advik Shreekumar and Pierre-Luc Vautrey (2020). “When Guidance Changes: Government Inconsistency and Public Beliefs.” Workig paper. Rehman, Nabeel Abdur, Shankar Kalyanaraman, Talal Ahmad, Fahad Per- vaiz, Umar Saif and Lakshminarayanan Subramanian (2016). “Fine-grained dengue forecasting using telephone triage services.” Science Advances, 2 (7), e1501215. Rubin, Jared (2017). “Rulers, Religion, and Riches: Why the West Got Rich and the Middle East Did Not.” New York: Cambridge University Press. World Bank (2018). “Using Smartphones to Improve Public Service Delivery in Pun- jab, Pakistan.” in Improving Public Sector Performance Through Innovation and Inter- Agency Coordination, World Bank Group, Global Knowledge & Research Hub in Malaysia. 19 5 Figures 20 21 Figure 1: Treatment Effects by Vote Choice (PTI Party) Notes : Each panel shows heterogeneity in treatment effects by vote choice for the PTI party on different outcome variables. Bars denote 95% confidence intervals. Table 1: Attitudes towards State Directives I intend to... I believe others should... I believe ... is beneficial. Attitudes Index (1) (2) (3) (4) Panel A: Frequent hand-washing Treatment effects Past state success 0.006 -0.004 -0.004 -0.013 (0.004) (0.004) (0.005) (0.026) Citizen cooperation -0.003 -0.000 -0.003 -0.019 (0.005) (0.004) (0.005) (0.027) Religious authority 0.001 0.000 -0.003 -0.006 (0.005) (0.004) (0.005) (0.026) N 4646 4628 4644 4611 Mean of control group 0.933 0.952 0.950 0.026 Experimenter demand effects Past state success 0.004 -0.001 0.001 0.010 (0.005) (0.004) (0.005) (0.026) Citizen cooperation -0.005 0.002 0.002 0.004 (0.005) (0.004) (0.005) (0.027) Religious authority 0.000 0.003 0.001 0.019 (0.005) (0.004) (0.005) (0.027) N 4592 4575 4588 4563 Mean of demand group 0.932 0.948 0.943 -0.013 Panel B: Social distancing Treatment effects Past state success 0.008 -0.004 -0.002 0.001 (0.007) (0.006) (0.007) (0.027) Citizen cooperation 0.007 -0.004 0.004 0.011 (0.007) (0.006) (0.007) (0.026) Religious authority 0.003 -0.003 -0.005 -0.005 (0.007) (0.006) (0.007) (0.026) N 4586 4604 4641 4518 Mean of control group 0.825 0.881 0.880 0.012 Experimenter demand effects Past state success 0.013* 0.000 -0.001 0.018 (0.007) (0.007) (0.007) (0.027) Citizen cooperation 0.011 -0.001 0.005 0.027 (0.007) (0.007) (0.007) (0.027) Religious authority 0.008 0.001 -0.004 0.012 (0.007) (0.007) (0.007) (0.027) N 4533 4547 4579 4465 Mean of demand group 0.818 0.875 0.877 -0.018 Panel C: Avoiding mosques Treatment effects Past state success 0.014 0.002 0.020* 0.027 (0.011) (0.011) (0.012) (0.025) Citizen cooperation -0.005 0.010 0.009 0.010 (0.011) (0.012) (0.012) (0.025) Religious authority 0.019* 0.016 0.020* 0.045* (0.011) (0.011) (0.012) (0.025) N 4500 4433 4522 4291 Mean of control group 0.544 0.596 0.581 0.004 Experimenter demand effects Past state success 0.006 -0.006 0.011 0.007 (0.011) (0.012) (0.012) (0.026) Citizen cooperation -0.013 0.002 0.001 -0.009 (0.011) (0.012) (0.012) (0.026) Religious authority 0.011 0.008 0.012 0.027 (0.011) (0.012) (0.012) (0.026) N 4442 4375 4462 4235 Mean of demand group 0.539 0.597 0.581 -0.010 Notes : OLS regressions of intended behavior, norms, and benefits on treatment. The unit of observation is the individual. In each panel, the first set of regressions estimates the effect of each treatment arm. The second set of regressions estimates the effect of each treatment arm relative to the experimenter demand group, thereby netting out potential experimenter demand effects. I intend to... measures how likely the respondent is to follow a behavior (wash hands more frequently, social distance, or avoid praying at mosque) on a 5-point Likert scale with higher values indicating a higher likelihood. I believe others should... measures how much the respondent believes others should follow the behavior on a 5-point Likert scale. I believe... is beneficial. measures how beneficial the respondent believes the behavior to be on a 5-point Likert scale. The attitudes index is the average of the z-scores of these three outcome variables. The index is set to missing if any of the included outcome variables is missing. The highlighted columns are the treatment effect on the attitudes index. All specifications include stratum fixed effects, enumerator fixed effects, post dummy, and past behavior (measured at baseline). Robust standard errors are in parentheses. * p<0.10, ** p<0.05, *** p<0.01 Table 2: Attitudes towards the State Panel A: State Capacity Panel B: General Attitudes Manage pandemic Provide public goods Enforce regulations Capacity Index Trust in Elected Officials Trust in State Perceived Benevolence Trust Index Govt Share Demand for Govt Info (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Treatment effects Past state success 0.007 -0.007 -0.022 -0.038 -0.014 -0.027** 0.011 -0.019 -0.026* 0.014 (0.013) (0.014) (0.014) (0.036) (0.013) (0.013) (0.011) (0.032) (0.016) (0.018) Citizen cooperation 0.008 0.004 0.002 0.011 0.017 -0.002 0.006 0.028 -0.011 0.008 (0.014) (0.014) (0.014) (0.036) (0.014) (0.013) (0.011) (0.033) (0.016) (0.018) Religious authority -0.003 -0.011 -0.031** -0.058* -0.009 -0.012 -0.009 -0.028 -0.020 0.007 (0.014) (0.014) (0.014) (0.035) (0.013) (0.013) (0.011) (0.032) (0.016) (0.018) N 4414 4510 4467 4404 4289 4267 4580 4105 3758 4661 Mean of control group 0.495 0.459 0.603 0.010 0.351 0.413 0.229 0.005 0.345 0.393 Experimenter demand effects Past state success -0.002 0.002 -0.025* -0.034 -0.028** -0.020 0.008 -0.040 -0.013 -0.002 (0.014) (0.014) (0.014) (0.036) (0.014) (0.013) (0.012) (0.032) (0.016) (0.019) Citizen cooperation 0.000 0.014 -0.000 0.017 0.003 0.005 0.003 0.012 0.001 -0.008 (0.014) (0.014) (0.014) (0.036) (0.014) (0.013) (0.012) (0.033) (0.016) (0.019) Religious authority -0.012 -0.002 -0.035** -0.055 -0.024* -0.005 -0.012 -0.048 -0.007 -0.009 (0.014) (0.014) (0.014) (0.035) (0.014) (0.013) (0.011) (0.032) (0.017) (0.019) N 4371 4451 4422 4351 4243 4225 4522 4066 3694 4602 Mean of demand group 0.508 0.454 0.611 0.017 0.367 0.405 0.234 0.026 0.336 0.417 Notes : OLS regressions of attitudes towards the state on treatment. The unit of observation is the individual. The first set of regressions estimates the effect of each treatment arm. The second set of regressions estimates the effect of each treatment arm relative to the experimenter demand group, thereby netting out potential experimenter demand effects. In Panel A, beliefs on state capacity to provide public goods, enforce regulations, and manage the coronavirus pandemic are measured on a 5-point Likert scale with higher values indicating higher perceived state capacity. The capacity index is the average of the z-scores of provide public goods and enforce regulations. The index is set to missing if any of the included outcome variables is missing. The highlighted column is the treatment effect on the capacity index. In Panel B, trust in elected officials, trust in the state, and perceived state benevolence are measured on a 5-point Likert scale with higher values indicating higher trust. The trust index is the average of the z-scores of these three outcome variables. The index is set to missing if any of the included outcome variables is missing. Govt share is the proportion of funds the respondent allocated to government coronavirus relief efforts in a lab-in-the-field game. Demand for information from the government is a dummy variable indicating whether the respondent wants to receive a text message summarizing the latest official recommendations and directives issued by the Government of Pakistan. The highlighted columns are the treatment effect on the trust index and the share of funds allocated to government coronavirus relief efforts. All specifications include stratum fixed effects, enumerator fixed effects, and the post dummy. Robust standard errors are in parentheses. * p<0.10, ** p<0.05, *** p<0.01 A Appendix (For Online Publication Only) A.1 Basic information Prior to the change in guidelines: I would like to take a break from the survey now to convey this public service announcement from the provincial Government of Punjab. The Government of Punjab recommends that you follow the following directives: 1. Wash your hands frequently, and wear a mask when outside if available. 2. Avoid shaking hands, and maintain distance from others when outside or at work. 3. Offer prayers (including taraveh and itekaf during Ramadan) at home in- stead of at the mosque After the change in guidelines: I would like to take a break from the survey now to convey this public service announcement from the provincial Government of Punjab. The Government of Punjab recommends that you follow the following directives: 1. Stay at home as much as possible. If you HAVE to go out, maintain distance from others avoid shaking hands. 2. You must wear a mask when outside. 3. Wash hands frequently. A.2 Past state effectiveness Experts at PITB (Punjab Information Technology Board) have shown that the provincial government’s handling of a past public health crisis was successful. They analyzed the provincial government’s response to the dengue outbreak in 2011. Punjab saw a massive outbreak of dengue fever in 2011 with 21,000 reported cases and 352 deaths. The government then introduced aggressive measures to control the outbreak which included: 24 • Closing down educational institutions to fumigate premises • Identifying and fumigating potential breeding sites • Introducing aggressive testing According to the experts, as a result of effective government measures, the total number of reported cases in Punjab declined to 255 in 2012. A.3 Past state effectiveness due to citizen cooperation {Repeat past state effectiveness message.} While these measures are certainly important, experts also point out that citizen cooperation is necessary to contain these kinds of outbreaks. An instance of such cooperation during the dengue outbreak is how citizens shared information with state institutions via the government hotline about where people showed symptoms of the disease. In the first 3 years of this operation, the system recorded more than 300,000 calls. The experts estimate that this high level of cooperation significantly slowed down the outbreak as it allowed the government to better identify disease hotspots and to then target resources more efficiently. According to the experts, as a result of effective government measures and citizens’ cooperation, the total number of reported cases in Punjab declined to 255 in 2012. A.4 Religious authority Religious authorities across the world are also recommending these measures to be followed. In fact, Supreme Ulema Council of Al Azhar University issued a fatwa at the beginning of the outbreak which disallowed holding Friday prayers in mosques. They state public gatherings, including congregational prayers, can spread the coronavirus. The fatwa further states “it is unlawful to make people lose confi- dence in the measures taken by the governments to protect their homelands and citizens.” A.5 Experimenter demand This is what the provincial government is recommending. I have been thinking about this lately and I think this is a really good idea. 25 B Appendix Figures Figure A1: Knowledge of treatments Notes : This figure shows average retention of information across treatment groups. We asked respondents in each treatment group two questions at the end of the survey. In the past state success and citizen cooperation groups, we asked Question 1: Was the provincial government’s handling of the dengue outbreak successful? and Question 2: Why was it successful?. Question 2 is asked only if the respondent answers Question 1 correctly. The enumerator did not provide answer options for Question 2. The correct answer for Question 2 depends on the treatment: In the past state success group, any response mentioning the government’s introduction of measures to control the outbreak is marked correct. In the citizen cooperation group, any response mentioned BOTH the government’s introduction of measures to control the outbreak AND citizen cooperation with the government is marked correct. In the religious authority group, we asked respondents Question 1: Does the Supreme Council of Al Azhar support the government’s recommendation to offer Friday and congregational prayers at home instead of the mosque? and Question 2: Why?. Question 2 is asked only if the respondent answers Question 1 correctly. The enumerator did not provide answer options for Question 2. In Panel A, any understand measures whether respondents answered any of the questions correctly and complete understanding measures whether respondents answered both questions correctly. In Panel B, we break down respondents’ understanding by question. 26 C Appendix Tables Table A1: Summary Statistics and Balance (1) (2) (3) (4) (5) T-test Basic Information (Control) Past state success Citizen cooperation Religious authority Experimenter demand Difference N Mean/SD N Mean/SD N Mean/SD N Mean/SD N Mean/SD (1)-(2) (1)-(3) (1)-(4) (1)-(5) Demographics Age 1139 40.299 1164 39.671 1121 39.011 1151 39.460 1085 38.731 0.628 1.289** 0.840 1.569*** (13.392) (13.116) (13.041) (13.318) (13.063) Female 1169 0.043 1179 0.044 1144 0.032 1169 0.037 1110 0.044 -0.001 0.010 0.006 -0.001 (0.202) (0.205) (0.177) (0.188) (0.206) Education level 1121 9.106 1129 9.305 1089 9.278 1112 9.188 1056 9.120 -0.199 -0.172 -0.082 -0.014 (3.685) (3.318) (3.419) (3.499) (3.555) Children engage in religious extracurriculur 767 0.366 809 0.376 775 0.335 782 0.377 775 0.361 -0.009 0.031 -0.011 0.005 (0.482) (0.485) (0.472) (0.485) (0.481) Covid-19 Economic cost of compliance, % 1146 0.614 1159 0.628 1118 0.610 1139 0.605 1095 0.632 -0.014 0.003 0.009 -0.018 (0.383) (0.381) (0.376) (0.371) (0.376) Do you know someone who was infected with the coronavirus? 1159 0.121 1164 0.143 1130 0.143 1147 0.139 1092 0.139 -0.023 -0.023 -0.018 -0.018 (0.326) (0.351) (0.351) (0.346) (0.346) Do you think the pandemic has natural origins? 916 0.872 896 0.863 860 0.853 906 0.879 854 0.862 0.010 0.019 -0.006 0.010 (0.334) (0.344) (0.354) (0.327) (0.345) Past behavior (0-1 Likert scale) Washed hands more frequently 1167 0.922 1176 0.914 1144 0.909 1169 0.923 1109 0.915 0.009 0.013* -0.001 0.008 (0.161) (0.178) (0.181) (0.163) (0.171) 28 Avoided meeting non-household members 1169 0.814 1179 0.816 1144 0.803 1169 0.817 1108 0.808 -0.001 0.012 -0.003 0.006 (0.233) (0.233) (0.243) (0.231) (0.239) Avoided Friday prayers at mosque 1160 0.518 1168 0.514 1140 0.521 1162 0.500 1104 0.507 0.004 -0.003 0.019 0.011 (0.419) (0.413) (0.411) (0.417) (0.417) Voting behavior and media consumption Turnout in 2018 elections 1169 0.791 1179 0.809 1144 0.810 1169 0.802 1110 0.803 -0.018 -0.019 -0.010 -0.011 (0.407) (0.393) (0.392) (0.399) (0.398) Vote for ruling party (PTI) in 2018 elections 925 0.418 954 0.472 927 0.434 937 0.451 891 0.424 -0.053** -0.015 -0.033 -0.006 (0.494) (0.499) (0.496) (0.498) (0.495) Watch ARY 1169 0.290 1179 0.324 1144 0.300 1169 0.307 1110 0.313 -0.034* -0.010 -0.017 -0.023 (0.454) (0.468) (0.458) (0.461) (0.464) Notes : This table reports the mean and standard deviation for each variable in each treatment group. The value displayed for t-tests are the differences in means between the control group and each treatment group. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. The t-tests show covariates are balanced across treatment groups at baseline: out of the 52 comparisons made (13 variables * 4 columns), 5 are significant at the 10 percent level, 3 are significant at the 5 percent level, and 1 is significant at the 1 percent level. This is to be expected given natural sampling variation. Table A2: Additional Beliefs towards State Directives Perceived Effectiveness Perceived Sanctions Shutdown of public places Suspension of Friday prayers From Others From Govt (1) (2) (3) (4) Treatment effects Past state success 0.003 0.010 -0.003 -0.007 (0.010) (0.014) (0.009) (0.013) Citizen cooperation 0.001 -0.003 0.013 0.005 (0.010) (0.014) (0.009) (0.013) Religious authority -0.006 0.019 0.008 -0.016 (0.010) (0.014) (0.009) (0.013) N 4593 4537 4306 4432 Mean of control group 0.828 0.524 0.762 0.565 Experimenter demand effects Past state success 0.002 -0.002 0.000 -0.001 (0.010) (0.014) (0.009) (0.013) Citizen cooperation -0.000 -0.014 0.015* 0.010 (0.010) (0.014) (0.009) (0.013) Religious authority -0.007 0.007 0.011 -0.010 (0.010) (0.015) (0.009) (0.013) N 4525 4475 4242 4369 Mean of demand group 0.831 0.534 0.755 0.564 Notes : OLS regressions of perceived effectiveness of policies and perceived sanctions for not following policies on treatment. The unit of observation is the individual. The first set of regressions estimates the effect of each treatment arm. The second set of regressions estimates the effect of each treatment arm relative to the experimenter demand group, thereby netting out potential experimenter demand effects. Respondent perceptions of the effectiveness of shutdown public places and suspension of Friday prayers in limiting the spread of coronavirus are measured on a 5-point Likert scale with higher values indicating higher perceived effectiveness. Respondent perceptions of sanctions from others and from the government are measured on a 5-point Likert scale with higher values indicating higher perceived sanctions. All specifications include stratum fixed effects, enumerator fixed effects, and the post dummy. Robust standard errors are in parentheses. * p<0.10, ** p<0.05, *** p<0.01