Doctor's Orders vs Divine Wisdom

How to deliver information most persuasively to individuals in developing countries is an important question with relevance not only for research, but also policy. Given the high rates of religiosity in many developing contexts, a recent literature considers the role of religious leaders and finds large impacts on behavior from information delivered by religious leaders. However, this work leaves unaddressed the policy counterfactual: namely, how would other, secular local leaders have fared in delivering the same information, especially when these leaders have more domain expertise? Whether utilizing religious leaders for information campaigns comes with unanticipated risks also remains understudied.

Working with the Pakistan Ministry of Health (MOH), I compare the effectiveness of religious versus secular leaders in conducting health information campaigns in the context of Punjab, Pakistan. By experimentally varying who leads community meetings on the importance of handwashing as well as the framing of the message, I hope to answer the following research questions: 1) are religious leaders more effective than secular leaders at communicating information on health, 2) to what extent is the influence of religious and secular leaders domain-independent as opposed to domain-dependent, and 3) does receiving information from religious leaders have the unintended consequence of making them individuals' preferred choice for information in the future?

2026-10-01

2027-04-01

- Hand hygiene as measured by an ATP test at endline
- Choice of imam or health worker on informational videos about family planning or religious giving

- Knowledge
- Self-reported health, beliefs, and behavior
- Beliefs about social norms
- Trust in state vs religious institutions
- Beliefs about the efficacy of handwashing relative to religiosity for health as opposed to fatalism

The study will be an experiment implemented in collaboration with the Pakistan Ministry of National Health Services that works directly with health workers in Pakistan. I will sample 1,800 couples at random from 30 communities, 18 rural and 12 urban, across three districts of Punjab with high under-5 diarrhea prevalence. Within each community, couples will be randomized to either a control condition, treatment 1 (in which the local imam delivers community meetings on the importance of handwashing), or treatment 2 (in which the local male health worker/dispenser delivers community meetings on the importance of handwashing). Treatments will also be cross-randomized such that in half the content is framed in fully religious language while in the other half it is framed in fully secular language. Randomization will be stratified by community. The unit of analysis is the individual, but randomization will be at the couple-level. Both husband and wife will be asked to attend meetings together. The randomization is as follows:

1. Control: 600 couples

2. Meetings with health worker: 600 couples
2a. Religious Framing: 300 couples
2b. Secular Framing: 300 couples

3. Meetings with imam: 600 couples
3a. Religious Framing: 300 couples
3b. Secular Framing: 300 couples

Not available

Randomization done in office by a computer

Couple-level

Yes

1800 couples

3600

- 600 couples assigned to meetings with male health worker (300 framed in religious and 300 framed in secular language)
- 600 couples assigned to meetings with imam (300 framed in religious and 300 framed in secular language)
- 600 couples assigned to control (no meeting)

I estimate that the MDE with 80% power to detect a statistical difference between the imam and health worker treatments (pooling my religious/secular treatments) for 1800 couple clusters is 0.14 SD for my primary ATP outcome. The MDE for comparing the sub-treatments to each other (e.g. religious vs secular) is 0.23 SD for my ATP outcome so the religious vs secular comparisons will be largely descriptive.