eVoucher and Farmer Field School Impact Evaluation 2023

Midline Round

The sampling procedure for the program impact evaluation includes four steps:

Step 1 – Identification of possible intervention communities.
At the beginning of the program, the research team asked FAO to provide a list of extension agents (EAs) who would be responsible for identifying communities for intervention, and established the catchment for each EA. From this list, the research team assigned each EA, by random lottery, a pipeline of communities in which to establish a FFS or register farmers for eVouchers. Working with a total of 102 EAs, the team identified 799 potential intervention communities.

Step 2 – Community level randomization.
From the long-list of communities in each EA’s catchment, four communities were randomly assigned to compose the experimental sample. Each of these communities was then assigned to one of the following four groups receiving either:

1. Farmer Field School
2. eVoucher
3. Farmer Field School + eVoucher
4. Control group
   At that time, only 56 of the 97 EAs had all 4 experimental sample communities within range of an agrodealer participating in the FAO eVoucher program. In areas where there is no eVoucher coverage, the EA is assigned two FFS and two control communities. In total, the evaluation sample includes 388 communities. 15 EAs were later dropped from the IE due to performance issues.

Step 3 – Within community identification of FFS interested participants.
Participation in the FFS is voluntary and based on farmers expressing interest. This means that farmers who choose to participate are likely not representative of the average farmer in the community. To allow for identification of likely FFS participants in a similar way in both the FFS treatment and control groups, each of the EAs visited their four experimental communities to list community members interested in participating in a FFS group prior to the roll-out of the interventions. During the listing, communities were asked to also identify two likely FFS facilitators of each group. On average 29.3 members per community were listed as FFS interested participants.

Step 4 – Within community farmer randomization of eVouchers.
To allow for measurement of spillovers of eVouchers within communities, a second randomization was done to select treatment and control farmers within communities. This is done among both FFS likely participants and among other members of the community. To obtain a list of all members in the experimental communities, an extensive household listing was performed in October and November 2020. The identification of treatment and control eVoucher households within communities is done for all evaluation communities, not just those assigned to receive eVouchers. This permits the identification of the equivalent households in both treatment and control groups.

The baseline survey sample is composed of all 388 evaluation communities. Within each community, 12 households are sampled from the following groups:

1. FFS interested + eVoucher treatment (6 households)
2. FFF interested + eVoucher control (3 households)
3. Not FFS interested + eVoucher treatment (2 households)
4. Not FFS interested + eVoucher control (1 household)

Both likely facilitators from the FFS listing were prioritized to be included in the survey sample. Sampling weights are applied when translating our sample averages to community wide average or other combinations of groups with different sampling probabilities.