Experimental Design
Each week, the researchers received a list of borrowers with payments due the following week. The first time a borrower appeared on the list, they were randomly assigned to a treatment group (receiving a text message reminder) or the control group (no reminder). Only one loan cycle per borrower was included in the study, since the likelihood of receiving future loans could be affected by treatment received in prior loans. The treatment messages varied by content and timing. The content dimension varied whether the text message was framed in terms of loss or gain, and whether the text message used the borrower's name or the loan officer's name. The timing dimension varied whether the text message was sent the day of, the day before, or two days before payment was due. The researchers observed administrative data on loan amount, term, and the number of weeks in the experiment. The outcome variables were five measures of loan delinquency/default. The researchers used an OLS model to estimate treatment effects, controlling for loan officer and month-year fixed effects.