The sample size of 917 households was projected to capture as much information as possible about AR participant farmers but also to have a basis for comparison with non-participant farmers to whom the program could yet be expanded. To achieve these twin objectives, households from all seven intervention villages already hosting AR activities were included alongside households from eighteen control villages that up to that point held no AR activities but that could serve as direct comparators.These latter villages were purposely selected on a ‘distant-but-comparable’ criterion, essentially a requirement that they exhibit similar agro-ecological conditions as existing program areas within the same district but be geographically apart, thereby ruling out the possibility of contamination. Lists of candidate comparator villages were established in concert with local-area extensionists and then verified with district-level agricultural officials.

The majority of households from the first group of communities, the seven intervention villages already hosting AR activities, were chosen on the basis of their active participation in AR activities, either from involvement in one or more of the work packages dating from program onset or from membership in an input-provision experiment conducted by IFPRI in mid-2013. In this experiment, farmers in Babati District attending field demonstrations were invited to participate in an experiment surrounding the use of modern seed varieties and non-traditional local fertilizer that resulted in their assignment (via randomization) into two sub-groups: coupon (and, a few months later, input) recipients and non-recipients.

To facilitate their identification, an initial listing of program participants in the seven intervention villages was sourced over several site visits and from ongoing dialogue with research team leaders. This was then matched with lists of participants from the experimental group in the three Babati District villages. These lists were later field-verified by EDI during the survey preparation phase. Ultimately, after accounting for duplicate households, households from non-intervention villages, and non-existent households, this portion of the sample was whittled down to 435 households (from 542 originally), comprising 328 households who took part in the above mentioned input-provision experiment and 107 who fell outside of it.

In addition, it was decided to include a randomly sampled subset of 15 non-participants from each of the seven intervention villages, to meet the question whether – within Africa RISING communities – anticipated program benefits could filter to non-participating farming households via indirectlearning, neighborhood interactions and networking (so-called ‘spillover effects’). These 105 households are referred as ‘indirect beneficiaries’.

Finally, households from the second group of communities, the comparator villages, were selected by two-stage sampling methods in which, first, one sub-village was randomly chosen from within each control village and, second, 20 households were randomly chosen from the list of all households within that sub-village. Fifteen households so chosen were entered in the final sample for enumeration and five served as reserve units in the event of non-response or other anomaly. Thus 270 households (and 90 reserves) were chosen in this way to serve as comparators to households actively participating in the program; they are referred to in the report as ‘control’ households.Thus, the final sample design for TARBES 2014 included purposeful selection of 435 AR and experimental households (from intervention villages only) and sampling of 25 sub-villages (from control and intervention villages) followed by 15 households in each sampled sub-village. In combination, this produced the ultimate sample size of 810 households.