We used a Randomized Controlled Trial (RCT) to evaluate the impacts of Anchor Farm Model's (AFM) access to extension and marketing assistance over a period of 5 years. This method entailed the random assignment of 250 villages into:
- Control villages;
- Treatment villages: Villages invited to establish demonstration plots and participate in farmer field-days throughout the project period;
In addition, a sub-set of both control villages and treatment villages were eligible to receive output marketing assistance in the two final years of the project.
We worked with CDI to identify 2500 farming households in 250 villages within two Extension Planning Areas (EPAs) covered by the ongoing project scale-up: Mtunthama in Kasungu district and Chibvala in Dowa district.
In 2014, we divided these 250 villages - randomly - into 2 groups: a control group of 125 villages with no access to CDI's extension activities and a treatment group of 125 villages who were invited to form farmer clubs and received access to CDI's extension activities through demonstration plots and field days. In 2017, we extended the randomization further by including marketing assistance sub-groups in each one of these two groups.
In each village, we selected 10 households, randomly in the control villages, and stratified by CDI club member status in the treatment villages.
We conduct this evaluation through a Randomized Controlled Design study that ran from 2014 until 2019, in which some villages, randomly selected, receive the certain components of the Anchor Farm Project, while others do not. Using base, mid and end-line surveys, we established the short and medium terms impacts of these various AFP components.
Data Collection & Instruments
The data collection consists of:
- A qualitative component
- An agronomic component
- Household questionnaires
- Lab-in-the-field games
We collected three rounds of data. A baseline in Fall 2014 (prior to all AFP components), a midline in Fall 2015 and an endline in Fall 2018. At base and endline, we also collect village level data. In addition, throughout the study, we collected program participation data from our partner, CDI.
A unique feature of our panel dataset is that we not only follow households over time, but we also follow fields over time, i.e., each field received a unique ID at baseline, and we followed up the field in terms of perceived and actual soil fertility, inputs and outputs over time. In addition, we follow network structures over time, allowing for an analysis of changes in networks and spillovers.
We conducted semi-structured interviews in five to ten treatment and control villages on a regular basis. During these interviews, we asked respondents to reflect on socio-economic aspects of their lives, and covered changes in household composition, livelihood strategies, expectations and perceptions, investment strategies and beliefs and hopes for the future. The reports are available on request from the team.
We also conducted focus group interviews among farmer clubs, once or twice a year, in about five treatment villages. Here, we focused on club dynamics, the perceived constraints and opportunities of club-based extension and marketing, the various club activities and plans and hopes for the future of the group. The reports are available on request from the team.
We collected data on all active farmer-led demonstration plots, including rainfall data using pluviometers, plant count data and yield data protocol described here. The daily rainfall data was collected by lead farmers whom we had trained for this purpose using a Rainfall Data Recording Sheet. The plant count data and yield data were collected by students and researchers from Bunda College.
In addition, we sampled and analysed soil samples from 600 farmer and demonstration plots including locational information collected by GPS (including selected track data) and plot history information using SOILDOC. These soil data are collected at base and endline and includes pH, organic carbon, phosphorus, nitrogen, calcium, magnesium, potassium, cation exchange capacity, and texture and bulk density.
The baseline questionnaires include: household composition, landholding and assets, beliefs regarding yields, group membership and networks, time preferences, input-output modules per plot, marketing and past agronomic practices (with a focus on ISFM). The questionnaires for the baseline are available via the links below.
The mid-line household questionnaire includes: updated household composition, networks, knowledge about ISFM, participation in program activities, adoption plans, beliefs regarding yields, input-output modules per plot, and marketing. The questionnaires for the mid-line are available via the links below.
The endline household questionnaire includes: household composition, networks, knowledge about ISFM, participation in the program activities, adoption plans, beliefs regarding yields, input-out modules per plot, marketing, dietary diversity, and landholding and assets. The questionnaires for the endline are available via the links below.
At baseline and midline, we played a public goods game among all members of the farmer's clubs aiming to measure cooperation within the clubs. The midline protocol involved a few randomized variations in terms of decision-making. At endline we developed a village-level game to measure inequality averse social norms. The game protocols are available via the links below.