Do prayers to private charities crowd out giving?

Seminal work by James Andreoni and others suggests that large contributions to charitable organizations partially crowd out individual donations. We extend this literature by assessing whether non-monetary actions - specifically prayers - similarly crowd out donations to charitable causes. To do so, we designed a survey experiment in which religious Christian respondents were given the opportunity to donate either to an individual or to a community in need following a natural disaster. A baseline condition simply solicits a donation. Treatment conditions inform subjects that either one or one hundred other people had previously offered a prayer for the individual’s or community’s full recovery. We hypothesize that others' prayers reduce people's perceived marginal benefit of donating and so decrease both the likelihood and the amount of giving.

2026-02-26

2026-03-31

We are primarily interested in estimating treatment effects on i) the rate of giving and ii) the amount that people give.

Within the survey experiment, subjects first provide some basic demographic information about themselves before learning about a town or individual (depending on the treatment) that was recently impacted by a natural disaster. Subjects are then solicited to donate part of a $5 bonus payment to the recovery of either the town or individual. Treatments inform subjects that either one or one hundred other people have previously prayed for the recovery of either the town or the individual. As such, our primary outcomes of interest surround the rate and amount of giving across the two treatments relative to the control in which other people's prayers are not mentioned.

In addition to estimating treatment effects on donation rates and amounts, our survey includes questions that will allow us to estimate a perceived, "prayer production function." To do this, we ask subjects about their perceptions of how likely an individual or community are to recover when i) no one prays, ii) 1 person prays, or iii) 100 people pray.

We will explore several sources of heterogeneity. Here, we are especially interested in how treatment effects vary by a person's level of religiosity. In particular, we will estimate heterogeneous effects among people who i) attend church at least once per week, ii) read religious scripture at least once per week, iii) pray at least once per week, iv) define themselves as "very religious", and v) interpret the Bible literally. We will additionally estimate heterogeneous effects among people who believe that a person (or community, depending on treatment assignment) is more likely to recover from a natural disaster after i) anyone prays for their recovery, ii) one person prays for their recovery, and iii) one hundred people pray for their recovery.

The survey experiment was designed and implemented using Qualtrics. Subjects first consent to participate in the study and then provide basic demographic information. After learning they have received a $5 bonus payment, they are informed of either a person or a community that is in need of support following exposure to a natural disaster (Winter Storm Fern). The study leverages a 2x3, between-subjects design in which subjects are randomly assigned to learn about a community or individual in need following a natural disaster and are then asked to donate up to $5. Subjects in the control group are solicited for a donation without any mention of prayer. Subjects in the two treatments are either told that i) one person has prayed for full recovery (for either the town or individual, depending on treatment assignment) or that ii) one hundred people have prayed for full recovery. Following the donation decision, subjects answer a set of questions designed to elicit their level of religiosity.

Not available

Subjects are randomly assigned to treatments using randomization logic within Qualtrics.

NA

No

1

Our sample includes 800 respondents; roughly 266 subjects in the control group and each of the two treatments.

Our sample includes 800 respondents; roughly 266 subjects in the control group and each of the two treatments.