Interacting with familiar and predictable individuals facilitates communication and enables behavioral change in various spheres, including nation-building processes (Bazzi et al., 2019, Mousa, 2020, Lowe, 2021), financial decision-making (Fisman et al., 2017, Fisman et al., 2020), and experimental games (Habyarimana et al., 2007, Bicchieri et al., 2022). The propensity to adapt behavior based on shared characteristics and identities is notable in health-related interactions such as those between patients and doctors (Greenwood et al., 2018, Alsan et al., 2019, Greenwood et al., 2020, Hill et al., 2020). Leveraging race, gender or class concordance has been found crucial when promoting preventive healthcare (Alsan and Wanamaker, 2018, Torres et al., 2021, Alsan et al., 2021, Alsan and Eichmeyer, 2021). Religion remains understudied in this context, despite its significant historical influence, its heightened importance in times of unpredictable events (Bentzen, 2021), and its centrality to public health in low-income settings (Iyer, 2016, Benjamin et al., 2016, Banerjee et al., 2022, Taragin-Zeller et al., 2023). In particular, little is known about the role of shared religious identity in the diffusion of health information and the spread of misinformation about preventive health.1

This paper examines the effectiveness of a physician-delivered information campaign that promotes health-related preventive practices. We investigate how introducing religion concordance between the sender and the recipient enhances the campaign’s effectiveness. We do so among residents of densely populated informal settlements, often referred to as ‘slum dwellers’, a largely understudied population (Lilford et al., 2017). We document that promoting preventive behavior can increase compliance with recommended practices and beliefs about their efficacy. Our findings indicate that the campaign’s impact is primarily driven by a shared religion between the sender and the receiver. In this case, recipients listen to a greater portion of the message and are more compliant with recommended practices. Furthermore, we find that religion concordance helps to protect against misinformation.

We implement a field experiment in the Indian state of Uttar Pradesh (UP) in the context of a global outbreak of an infectious disease: the COVID-19 pandemic. At the onset of the pandemic, we designed a mobile-phone-based information campaign to raise citizens’ awareness about evidence-based practices to mitigate the spread of the virus, and to counteract the sudden rise in misinformation surrounding the pandemic (World Health Organization, 2020).2 To this purpose, between October 2020 and January 2021, we sent two pre-recorded voice messages to a representative sample of slum residents, in the two major cities of the state. The campaign held particular importance in this context, not only due to the overcrowded living conditions that made physical distancing challenging, but also due to the low-income and marginalized nature of the setting, which limited access to healthcare and adequate hygienic conditions.

Each voice message consists of two components: an introduction by a local citizen, the sender, followed by the content of the message. Using cross-randomization, we vary both components. First, to obtain exogenous variation in religion concordance between the sender and receiver, we randomly vary the greeting used by the sender at the beginning of the message to signal either a Muslim or Hindu identity. Religion is highly salient in our setting, particularly at the time of the experiment. In India, Hindu–Muslim tensions have been present since the pre-partition era, and are particularly relevant for UP, home to the largest Muslim population in India (Jha, 2013, Mitra and Ray, 2014). In line with religion being salient in the presence of unpredictable events (Sinding Bentzen, 2019, Atkin et al., 2021), the onset of the pandemic saw a sudden increase in these inter-religious tensions: misleading claims about the role of Muslim citizens in the spread of the virus were the primary driver of fake news on social media and spurred further violence (e.g. Yasir, 2020).

Second, to obtain exogenous variation in the content of the message, we randomize whether the receiver is sent messages about preventive practices or uninformative content. In the former, which we label as doctor messages, the content is provided by doctors of locally renowned hospitals, provides reminders about evidence-based policy recommendations, and debunks common misconceptions about the virus. The religious identity of doctors is not revealed. In the latter, which we label as control messages, the content consists of Bollywood gossip unrelated to the pandemic. Thanks to cross-randomization, both the doctor and control messages are either religion-concordant or religion-discordant.3

We gathered information about participants’ behavior related to preventive practices, particularly the extent to which respondents wear a face mask when going out, the frequency of hand-washing, and the extents to which they stay in the slum, do not receive visitors from outside the slum, and do not meet anybody from outside the slum. We aggregate these individual reports into an index of compliance with recommended practices. Additionally, we collected data on beliefs over the efficacy of both recommended and non-evidence-based practices, and about participants’ response to misinformation about the pandemic, during a baseline and two follow-up surveys. We base our main analysis on intention to treat (ITT) effects, which capture the effect of sending the messages. Using administrative data on the take-up of the interventions, we complement ITT estimates with local average treatment estimates (LATE) of the effect among compliers.

The design of the experiment allows us first to study the overall effect of promoting preventive practices and then to estimate the effect of introducing shared religion between the sender and receiver, a novel set up in the literature. Providing informative content via mobile phones is effective at promoting welfare-improving behavior. Compared with control messages, doctor messages significantly increase compliance with recommended practices and update recipients’ beliefs about the efficacy of these practices positively. However, despite being debunked in the message, doctor messages have no significant effect on the degree to which respondents believe that non-evidence-based practices such as relying on vegetarianism or on a stronger immune system can protect from infection, indicating the persistence of these beliefs to new information.

To assess the added benefit of shared religion, we focus on the sample that was sent the doctor message and we exploit the cross randomization in the religion concordance between the sender and the receiver of the information. First, we find that religion concordance leads participants to listen to a larger portion of the doctor message, an increase of 13.3% compared with religion-discordant messages. Second, the effect of doctor messages on compliance with recommended practices is primarily driven by religion-concordant messages. Third, religion concordance in the doctor messages effectively reduced beliefs over the efficacy of non-evidence-based practices, particularly those with a religious connotation.

The last two results are specific to the combination of informative content provided by the doctor and religion concordance. Studying the differential effects of religion concordance in the control messages, which serves as a placebo test, indicates no effect in any of the outcomes studied. In addition, the effects are specific to misinformation. In fact, none of the interventions influences agreement with non-factual opinions about the spreading of COVID-19, by definition more persistent and harder for information campaigns to influence than pure misinformation (e.g. Walter and Salovich, 2021). Finally, we provide evidence that spillover effects were not present in the interventions, suggesting that mobile-phone campaigns are effective at targeting individuals rather than communities.

To understand the drivers behind these impacts, we first analyze respondents’ fact-checking behavior, an important determinant of factual knowledge (Barrera et al., 2020). The findings reveal that doctor messages significantly reduce the likelihood of verifying the truthfulness of information. This reduction is likely because individuals, having heard the messages from doctors, feel more confident in dismissing misinformation. We further use a novel survey instrument to measure whether respondents agree with misinformation shared by other citizens and show that doctor messages reduce agreement with misinformation shared by citizens outside the religious group of the respondent (out-group citizens), while keeping unchanged their level of agreement with citizens of the same religion (in-group citizens). Religion concordance in the doctor messages is effective at detaching in-group norm compliance in the response to misinformation. When the sender and the receiver have the same religion, doctor messages reduce agreement with misinformation shared by in-group citizens by 4.6% compared with religion-discordant messages. This finding aligns with existing research and for high-income countries, which emphasizes that the perceived credibility of information is influenced by the social distance between the communicator and the recipient (Tabellini, 2008, Alsan et al., 2019).

Our results suggest that the information campaign somewhat reduces the effort to verify information’s truthfulness while creating a protective layer against misinformation. However, this layer is crucially affected by salience within a group, suggesting a high level of in-group norm compliance in our setting (e.g. Akerlof and Kranton, 2000). However, this compliance can be reduced through a carefully designed information campaign that takes into account social proximity with the objective of leveraging social norms, challenging the assumption that in- and out-groups agree with prevailing norms.

To address concerns related to the treatment group exerting more social desirability bias in the self-reported outcomes, we collect measures of the (Crowne and Marlowe, 1960) social desirability scale at baseline. Although individuals with a strong tendency toward social desirability may show more endorsement for recommended practices or widespread beliefs, we demonstrate that this pattern is not more pronounced in the treatment group tan in the control group. In addition, we show that, at baseline, social desirability does not influence reporting differently depending on the religion, the gender, and the caste of the respondents.

Our findings offer novel insights into the design of information campaigns, an instrument that has been extensively used to communicate risk and best practices for health behavior (Dupas, 2011). We complement available evidence on the effectiveness of communication technology to raise health awareness in the US (Alsan et al., 2020, Breza et al., 2021, Torres et al., 2021), in the Indian state of West Bengal (Banerjee et al., 2020), and in rural India and Bangladesh (Siddique et al., 2022). We further the understanding of these interventions by providing novel evidence on how the effectiveness of information campaigns on preventive behavior is crucially influenced by shared identity. Our design is unique in the literature because it allows identification of the effect of the initial signal of shared religion (i.e. the first word of the message), while keeping the content of the message indistinguishable in terms of religious identity. Previous literature focuses instead on micro-targeting (i.e. the shaping of both the sender and the information content to the individual characteristics of the receiver). This approach has been used to influence interactions with patients (Yom-Tov et al., 2018, Alsan and Eichmeyer, 2021).

By linking compliant behavior with beliefs and response to misinformation, we provide novel evidence not only on the drivers of information, but also on the mechanisms of misinformation, whose persistence remains a puzzling result in the literature (Van der Linden et al., 2017, Zhuravskaya et al., 2020). In particular, despite the recognition that understanding how beliefs are affected by information is crucial, few studies explicitly elicit the effect of information on beliefs over practices and on (dis)agreement with misinformation (Kremer et al., 2019).

Finally, highlighting the role of religion also complements available evidence on the role of identity in decision-making. The literature shows how identity affects cooperation, political mobilization trust, and violence (Philpott, 2007, Bhalotra et al., 2014, Lowe, 2021, Alsan and Wanamaker, 2018), but there is limited evidence on information-sharing. We reinforce the role of religious identity among interacting citizens, a growing field of study in both economics and political science (Iyer, 2016). The specific focus on the use of religion for spreading information through mobile phones furthers our understanding of how these technologies stimulate social mobilization (e.g. Enikolopov et al., 2020, Manacorda and Tesei, 2020).