Experimental Design
Sampling and data collection
Our sample of households is drawn from 61 villages located in 9 communes and departments of Dogon-Doutchi and Tibiri in the Dosso region in Niger. The villages have been randomly chosen, but we had to avoid insecure villages while constituting the sample.
The household baseline survey was conducted in February 2024 with 978 households. We recorded GIS locations of villages and households. We also implemented a phone survey in September 2024 interviewing the same households to assess the impact of a small randomized information treatment embedded in the baseline survey (see below for further details).
Our sample of agents was obtained through a listing of mobile money and airtime agents in 38 markets in the departments of Dogon-Doutchi and Tibiri. The markets were identified as the key markets associated with the villages from our household sample.
The agent baseline survey was carried out in August 2024 and it was conducted with 190 agents from 36 markets. Out of the 190 agents, 92 are mobile money as well as airtime agents and 98 are only working as airtime agents.
We will do two additional follow-up surveys with the households as well as with the agents.
Randomization procedure
We are implementing several treatment interventions throughout this project. First, as part of the baseline survey, we conducted an information experiment. We grouped communes into three regions: North, Middle, and South. Next, we created strata based on commune group and participation in a previous study on an adult literacy program. Villages were then randomly assigned to receive no information (1/3) or information (2/3). The villages that received information, the treatment intensity was either 50% or 100%. In the 50% treatment intensity villages, the random assignment of the information was stratified by respondents’ gender.
Second, for the random assignment of the financial incentive treatment, we stratified the 61 villages based on their baseline treatment status: whether households received information (grouping 50% and 100% villages together) or did not receive any information. Within each of these groups, we randomly assign villages into ”Treatment R”, ”Treatment R+S” or the control group. Villages are assigned into one of the three groups and in each village all households receive the same treatment. We will test the treatment design in a pilot with villages that are not part of our sample.
Spillovers
Treated households may share the information or the flyer that is handed out as part of the intervention with other households in the same village or other villages. Information sharing within the same village is not problematic for our identification, because we randomize the treatment at the village level. Since we will analyze intent-to-treat effects, information sharing across treated and control villages may only underestimate the effect of the information. However, we will monitor the extent of information sharing. Since from a policy perspective information sharing is a desired action which should increase the cost-effectiveness of the intervention, we will ask households whether they shared the information about mobile money with other households.
Given the personalized nature of our financial incentive and the in-person monitoring done by our implementing partner, we do not expect significant spillovers from our financial incentive treatment on control villages. The treatment and control villages are geographically dispersed, which minimizes the likelihood of interaction between treated and control units. These factors together suggest that any unintended influence of the treatment on control groups is likely to be minimal.