Uganda and Malawi: First, researchers formed partnerships with a formal bank in each site, and determined study areas based on low density of comparable saving services and high degree of access to the partner bank. Second, to determine study eligibility, a census was taken in the catchment areas around the branches of the study. (In Uganda, 3 bank branches were utilized for the study; in Malawi, 2 branches.) The sample was randomly selected from census households that were deemed eligible: they did not currently use a savings account and at least one member of the household had some form of income. The baseline survey was then administered to the eligible sample.
After the baseline was analyzed, treatment status was assigned as described below. The treatment households were delivered the intervention by bank staff, monitored by IPA. We then administered follow-up surveys at 6-, 12- and 18-months will be used to estimate the impacts on a range of household activities, including agricultural and business practices, expenditures, household income, response to shocks, and savings and credit practices. In Uganda, qualitative data was collected to understand the mechanisms through which access to a bank account affected (or not) the study participants.
Chile: JPAL identified a partner financial institution and selected rural areas in which the partnering institution was operating in Region IX, one of the poorest regions of the country. A probabilistic sampling strategy was then used to enroll into the study a representative sample of unbanked households in those areas. Upon enrollment and completion of a baseline survey, study households were randomly assigned to either a treatment or comparison group as described below. Those assigned to the treatment group received procedural assistance with account opening.
In Chile, a low take-up rate of 17% prompted a new approach to the study design, as detecting impacts on downstream outcomes was no longer feasible. To explore the issue of low take-up, we decided to conduct a qualitative follow-up survey among the 938 treatment respondents. Control respondents were not included in this round. 639 treatment respondents were successfully surveyed.