Global Agriculture and Food Security Program Impact Evaluation 2018, Endline Survey

Sample size: 570 households

The core strategy for the evaluation is a cluster-randomized phase in of subsidy offers with individual assignment of input deliveries within treatment communities. What this means is that, because SAPEC can only provide a fixed quantity of inputs in a particular season due to capacity constraints, the study aims to randomly assign which communities are first in line to receive the inputs and which farmers in these communities are first in line to receive the inputs. Random selection of farmers and communities is a fair way to decide who receives inputs first that allows us to compare farmers who have been offered inputs with those who have not been offered yet without ultimately affecting which farmers receive benefits at the project closing date. Including some communities and not others in the first round allows to rule out spillovers of farmers who get inputs to other farmers in the same community by comparing to communities where no one has received any inputs. Randomly selecting farmers within communities in contrast also allows us to make comparisons between farmers who are very similar to each other. Finally, strategically make offers of inputs to youth vs older farmers in order to also make careful comparisons of impacts on these groups of particular interest.

There are two dimensions on which the offer of improved seeds, cuttings, and tools will be made in order to assess the impact of this offer on agricultural practices and outcomes, both in general and differentially by the age of the farmer.

First, we propose to compare households in communities where SAPEC will offer subsidized inputs against communities where subsidized inputs will not be offered during the evaluation year. Assignment to SAPEC input provision (treatment communities) or no provision (non-treatment) occurs at two levels. First, 100 communities in SAPEC treatment and control districts will be randomly selected to take part in the study, using the list of SAPEC-eligible communities in the LATA database as a sampling frame. Each community will have a minimum of 25 farmers. Of the selected communities, 50 will be treatment communities and 50 will be control communities. Since SAPEC can deliver inputs and technical assistance to a maximum of 5,000 farmers in a given year and the potential number of beneficiaries in all SAPEC-eligible communities is much larger than this 5,000, the random assignment of potential treatment communities to the set of controls only changes the order of who receives the benefits next, rather than withholding benefits from anyone in particular.

The second dimension to create a counterfactual for beneficiaries will be the random selection of specific beneficiaries within SAPEC treatment communities. From the list of all farmers from the randomly-selected treatment and control communities that have been registered in the e-platform system, we will randomly select an average of 10 per community to be SAPEC beneficiaries (or survey respondents, for control communities) in the upcoming round. This will allow us to sample farmers within treatment communities as well as farmers who were randomly selected to not receive them this year, allowing us to see the causal impact of this input provision on the delivery of tools.

Within the communities to be treated this year, we will start from the list of all households listed in the mobile phone registries who are eligible to receive SAPEC benefits in this year’s wave of benefit delivery. In advance of distribution, we will stratify these eligible farmers by age and gender, and randomly assign 1,500 of them to be invited to report to a SAPEC office in order to be registered as beneficiaries, with equal proportions of male and female and under- and over-35-year-old farmers receiving invitations. All of the farmers who appear to be registered will receive the package of benefits from SAPEC. If less than 1,500 farmers reply, a corresponding number of additional invitations will be sent to randomly selected households who were not selected in the first round until 1,500 beneficiaries have been registered. When invitations are sent, the invitee will be randomly assigned to receive one of 5 different types of messages which emphasis different features such as the potential to earn higher income or to meet other farmers the invitee as described in section 5. This will allow us to test whether different types of farmers (young vs. older) respond to different aspects of program benefits or design.