Experimental Design
We study whether telling loan applicants what other borrowers ask for changes how much they themselves request and whether the behavior varies by applicant gender. Applicants are randomized into four groups: information about male peers similar to the participant, about female peers similar to the participant, about peers similar to the participant (gender neutral), or no information. The information given to the participants in the treatment arms is an average from random draws from real past applications (a sample of 5 or 20 applications, randomized, and the participants are aware of how many peers they are getting the information from).
This extra variation helps us examine mechanisms. The three treatment arms vary two things at once: the amount shown and who the information was about. Using this extra variation in the amount we can hold the information the clients were provided with fixed and test whether it is the amount or who the information is about that matters for behavior. We will also test what kind of rule borrowers follow: for example, do they only react when the number is above their own planned request? Differences between male and female applicants are of primary interest throughout.
Details
During the call, applicants first state how much they want to borrow. They are then asked what they think similar borrowers typically ask for. Applicants in the treatment groups then hear the average requested amount of a sample of comparable past applicants (male, female, or general, depending on the group); the control group hears nothing. Treated applicants are asked again about their beliefs about what other borrowers typically ask for. They are also asked some additional questions such as how common is taking a loan in their area and whether they know any other people who have borrowed from the same company. At the end, all applicants are asked whether they want to keep or change their requested amount, and it is made clear that keeping the original amount is fine. Approval and default rates are taken from the institution's records.
Our main comparisons will be:
1) Does providing gendered information (about female or male peers) have differential effects compared to providing gender-neutral information. Do these effects vary by respondent gender? The no-information group will serve as a benchmark for secondary analyses.
2) Is it the amount shown or the group it describes? The three treatment arms vary both at once: the amount and who the information is about. Because the amounts are random draws from real past applications, we can hold the amount fixed and isolate the effect of who the information describes: other men, other women, or peers in general.
3) Test whether people mainly change the requested amount when the information provided exceeds their own planned amount and whether that trigger differs by treatment group.