Experimental Design
Three small and medium agribusiness enterprises (SMAEs) namely: ACILA enterprises, ALITO Joint, and OKEBA operating in the eastern, northern, and western and central regions, respectively are implementing an intervention involving three key incentives. First, farmers organized in groups are trained through training workshops to create awareness and their understanding of different CSA practices and technologies. Thereafter, trained farmers begin to access agricultural extension services through demonstration sites or farmer field schools (FFSs) established within their communities. At the demo sites, the extension agents also residing within the same communities practically teach farmers how to apply different CSA practices and technologies. Second, the SMAEs facilitate smallholders’ access to quality inputs either directly through direct sales on cash/credit, or indirectly through brokered linkages with agro-input dealers. Upon harvest, farmers sell their output to the SMAEs at a higher than average market price. This SMAE-farmer relationship is established through either a written or verbal contract. Through this relationship, input and output market risks are addressed. Finally, the SMAEs with support from SNV identify key insurance service providers to sell index-based insurance to farmers. Farmers pay half of the premium while the other half is subsidized by the government of Uganda. Index-based insurance insulates smallholder farmers against risks associated with weather changes including excessive rains or seasonal drought.
Famers do not pay for the supply-driven training and extension services. However, farmers may incur other costs associated with travels to and from the training venues and demo sites. We combine training and extension and label these push incentives since both incentives address informational constraints and any uncertainties associated with the technologies. On the other hand, demand-driven technologies/inputs such as improved seed, and insurance services are accessed through production/marketing contract, and index-based insurance incentive instruments, respectively. In other words, farmers have to decide whether or not to participate (buy) in contract farming arrangements (index-based insurance). Therefore, production/marketing contract is labelled pull-1 while index-based insurance is labelled pull-2 incentive.
We form different combinations (bundles) of push and pull incentives and assess their effectiveness on adoption intensity of CSA practices and technology before assessing impacts of CSA innovations on factor productivity, incomes and household welfare. The bundles include: (i) push; (ii) Push + Pull-2; (iii) Push + Pull-1; and (iii) Push + Pull-1 + Pull-2. Incentive bundles/combinations of push and pull incentives can potentially effectively address the multiplicity of risks and other constraints faced by smallholders thereby improving adoption of seemingly profitable innovations such as improved seed and rhizobia inoculants. In other words, raising adoption might require incentive bundles that address various risks and constraints, contemporaneously. This is particularly important because standalone incentives are normally offered to address specific risks or constraints but not a continuum of risks/constraints.
The SMAEs roll-out the intervention to the participants in phases by targeting new farmer groups every season for two years (Table 2). Also important to note is that, index-based insurance becomes available to farmers in the second year due to the delays in the identification and selection of insurance service providers. Therefore, we implement a concurrent stepped-wedge Cluster Randomized Controlled Trial (CRCT) with SMAEs to assess among their farmer groups the effectiveness of the four incentive bundles. Randomization is done in three steps. First, we randomly assign farmer groups across seasons due to phased roll-out of the intervention where new farmer groups are enrolled at the beginning of a new season. Using block-cluster randomization approach, we randomized farmer groups (clusters) into three instead of four cropping seasons because the SMAEs (blocks) had already selected beneficiaries for the first season of 2020 by the time of randomization. Random assignment of farmer groups to the seasons helps us to randomly determine the sequencing of enrolment of farmer groups into the CRAFT program. Second, farmer groups were randomly assigned to push or push + pull-1 incentive bundles. Finally, we randomize access to index-based insurance among farmer groups that begin to receive push or push + pull-1 in either year 2020 or year 2021.
Each farmer group starts with (1) the business as usual condition or the baseline phase in which usual farming without incentives is assessed, followed by (2) a second (mid-line) phase in the year 2020 in which 73 farmer groups receiving push (T1-1 and T2-1) and/or 76 farmer groups receiving push+pull-1 (T3-1 and T4-1) incentive bundles are compared with 92 and 88 control farmer groups that begin to access push (T1-2 and T2-2) and push+pull-1 (T3-2 and T4-2) incentives in the year 2021, respectively (Table 1). In the final (end line) phase, five more comparisons can be made. That is, 89 farmer groups receiving Push+Pull-1+Pull-2 (T4-1 and T4-2) are compared with: (i) 87 farmer group receiving push (T1-1 and T1-2) incentives; (ii) 77 farmer groups receiving push+pull-1 (T3-1 and T3-2); and (iii) 78 farmer groups receiving push+pull-2 (T2-1 and T2-2) incentives. Similarly, recipients of Push+pull-2 incentives are also compared with recipients of (i) push, and (ii) push+pull-1 incentives. Noteworthy, 180 farmer groups that wait to receive interventions in the year 2021 serve as the control groups for assessment at the mid-line period. Farmer groups that began to receive push and push+pull-1 incentives in the first year (2020) continue to receive the same incentives in the second year (2021) except that part of these are randomly assigned to receive index-based insurance as an additional incentive (the last two bars in seasons 1 and 2 of year 2021 in Figure 1). We collect data from 2,533 households on several variables of interest at the baseline period in the year 2019 before implementing the CRAFT program to facilitate measurement and estimation of outcomes of interest such as adoption intensity.
It is possible to have cases of non-compliance or partial compliance within treatments. Due to different number of incentives in each incentive bundle, each treatment can have different degrees or levels of compliance. We can define two types of non-compliers: never-takers that will always reject a new intervention if they are offered it, and always-takers that will always receive a new intervention even if they are not offered it (Ye et al., 2014). For instance, a member of a farmer group in the push treatment is regarded as a complier if he or she takes both the training and extension services, but would not have done so if it was not offered. A member is regarded as a partial complier if he or she takes only extension services or only training but not both. There is a possibility for farmers to access inputs and technologies and other services under a (verbal or written) contract and, for some reasons, still sell soybean output outside the contract. In this study, a farmer takes up pull-1 incentive if he or she sells the largest share or all of marketed soybean output to the SMAE, regardless of receipt of inputs and other services. A farmer takes up pull-2 incentive if he or she uses it, irrespective of whether or not it was bought. By design, receipt of pull-1 and pull-2 incentives depends on receipt of push incentives. Therefore, it is not possible that a farmer can access only pull-1 or pull-2 incentives. However, a farmer can choose not to take push incentives although such cases are expected to be very rare. Instead, there might be cases where some farmers take only extension services or training. Similarly, a farmer can take both or part of push incentives but chooses not to take pull-1 or pull-2 depending on assigned treatment.
The effects of the four incentive bundles will be assessed in two ways. First, we estimate the average intention-to-treat (ITT) effect among those assigned to treatment (incentive bundles). The ITT effect reflects how farmers respond to te randomized offer regardless of their actual take-up of the treatment. Second, we can account for non-random compliance by using the random offer as instrumental variable for non-random take up of the treatment, yielding a local average treatment effect (LATE).