The study sample consists of 53,000 households in 30 municipalities of Colombia. Colombia contains 1123 municipalities. First, municipalities that are department capitals were excluded as the government interest was to evaluate interventions to boost the take up of digital wallets in smaller localities with harder-to-reach beneficiaries. Second, municipalities were ranked by the number of unbanked Ingreso Solidario beneficiaries and the top 10% were retained. Third, from this list, the 30 municipalities with the greatest number of locations of the financial institution were selected.
The first randomization assigned municipalities to treatment and control. Stratified randomization was used to assign 15 municipalities to treatment and 15 municipalities to control. Two strata based on the number of locations of the financial institution were used. The second randomization assigned beneficiaries to treatment and control groups in equal proportion within each municipality. Therefore, only beneficiaries assigned to the treatment group in the treatment municipalities receive the intervention.
The treatment effect can be estimated by comparing outcomes for the treatment beneficiaries in the treatment municipalities to outcomes for beneficiaries in control municipalities. Spillovers can be estimated by comparing outcomes for the control beneficiaries in treatment municipalities to outcomes for beneficiaries in control municipalities. The randomization of beneficiaries in control municipalities into treatment and control groups can be used for placebo tests.
We will analyze outcomes in administrative records (mode of payment, ownership and use of formal financial products, formal employment, and usage of health services) using this study sample of 53,000 households. We will analyze treatment effect heterogeneity on the main outcomes based on predetermined municipality-level characteristics such as number of bank branches, ATMs, point-of-service terminals and demographic characteristics of local agents. Also, we will analyze heterogeneous effects by gender and household size for the main outcomes and time use. We plan to conduct household surveys using a sample of 5,000 of the 53,000 households, approximate equally split between treatment and control groups, that will allow us to examine additional outcomes of interest.
Control: no intervention
Treatment: Encouragement intervention