The endline survey sample frame comprised 65 associations – 33 that had received the PROIRRI infrastructure and 32 that had not. These 32 “control” association were selected from the long-list of 90 associations that were considered by the project for rehabilitation during the inception phase. Among the non-selected sites on the long-list, the associations chosen for the survey were those that most closely mirrored the PROIRRI irrigation schemes, matching on district, value-chain and association legalization status. The 65 associations were of varying size and structure. Those in Manica generally have fewer members (45 on average), are smaller in area and produce horticulture crops. Those in Sofala and Zambézia generally cover a larger area, have a greater number of members (111 on average) and primarily cultivate rice. The map in Figure 2 shows the spatial distribution of surveyed farmer associations, denoting at the time of the survey whether the association had received finalized irrigation construction as part of PROIRRI. Any incomplete or never-functioning schemes intended for rehabilitation under PROIRRI are classified as non-PROIRRI. The majority of associations were located in the Manica uplands, around the Zambezi delta, or by the Buzi and Muda rivers in Sofala. As a first step of the endline survey, a member listing was conducted to obtain a sampling frame for the household survey. The association listing exercise yielded 4,748 total members in these 65 associations, from which 2,638 households were sampled – 1,051 from 34 associations in Manica, 1,153 from 20 associations in Sofala, and 434 from 11 associations in Zambézia. An average of around 40 households were sampled in each association. The sampling rules were as follows:
- All association members that participated in the PROIRRI matching grant scheme were sampled, so that this could be fully evaluated.
- All remaining association presidents and treasurers were sampled.
- If an association has fewer than (or equal to) 26 members, all remaining members were sampled.
- If an association has more than 26 members, all remaining (after steps 1 and 2) members are randomly sampled to reach 26, then 20% of the remainder are also added to the sample.
The size and structure (proportional oversampling of smaller associations) of the endline sample frame ensures that there is sufficient statistical power to identify differences in outcome variables over the groups of interest. We use sampling weights in the analysis to compensate for the difference in sampling probability across FAs. In total 1,159 replacements were made to the sample throughout the household data collection. Replacements occurred when a listed association member could not be interviewed, with the most common reasons being due to the member belonging to a household that has already been interviewed, the member being unknown, or difficulties locating the member. The majority of replacements were in Sofala (71%), with 70% of these coming from three large associations, where the association leaders and guides were unable to locate or did not know a substantial proportion of their members.
