The field experiment was implemented at 77 stores of the partner payday lending company. To set up the random application of treatments, treatment was assigned at the store-day level; i.e. all treatment participants who entered a certain store on a given day received the same information treatment envelope. The following day, a new treatment envelope would be randomly assigned to all treatment individuals. The assignment algorithm forced dispersion of treatments by not allowing duplicate treatments for a given store until all possible combinations of information treatment and savings planner treatment had been assigned at least once.
After the conclusion of the intervention, the researchers received administrative data on all transactions by the participating borrowers from 2002 to October 1, 2008. A dummy variable indicating whether a participant borrowed during a given pay cycle was generated for all participants, e.g. the observation was recorded as "0" if there was no record of a transaction during a pay cycle. The main empirical specification regressed this dummy, "Indicator for Loan" on a set of treatment dummy variables while controlling for Store*Year fixed effects, day-of-the-week fixed effects, and individual fixed effects. A second specification featured the loan amount in a given pay cycle as the dependent variable in a Tobin model.