Experimental Design
The experimental design for the two financial institutions will be quite similar.
First, the sample will be stratified by region to ensure representativity across Peru.
For both financial institutions, the unit of randomization is the financial advisor. Advisors will be randomly assigned to either the treatment or control groups.
For Institution A, financial advisors organize their clients into groups ranging in size from 10 to 25 members; advisors communicate with these groups via Whatsapp. The universe of advisors eligible for the experiment will have: i) at least 20 groups of clients, and ii) a minimum of two years working for Institution A.
Given the relatively small size of groups, our strategy will be to include all members of a group in the sample; as noted above, since advisors are randomly assigned to either the treatment or control group, there is no possibility of one group of a given advisor being in the treatment group and another group from this same advisor being in the control group. (This avoids the potential of contamination of the control groups via information accidently leaked by the advisor.)
From this universe, a sample of approximately 588 financial advisors will be randomly selected to participate in the experiment. Of these, half will be randomly assigned to the treatment group and half to the control group.
For Financial Institution B, the unit of randomization is also the financial advisor. Group size is larger than in the case of Institution A. Groups may have up to 250 members, and the average group size is about 150. From the universe of financial advisors, approximately 108 will be selected to participate in the experiment, with half being randomly assigned to the treatment group and half to the control group. Given the large group size, not all group members from a given advisor will be included in the treatment or control group; this participation will be capped at 30 individuals in order to ensure the participation of a sufficiently large group of advisors.