Experimental Design Details
Our empirical strategy focuses on the following three elements: i) the impact of the FDP, ii) the spillover effect of the FDP, and iii) the impact of relationship-building interventions.
* Impact of FDP program *
The selection of farmers to participate in the FDP was implemented in several stages. First, we assign scores to all farmers based on scoring criteria. The scoring variables are based on a baseline survey, plot visits data, and administrative data on coffee delivery. Farmers are scored relative to each other within their corresponding CWS.
Second, we randomize the percentages of invited farmers to be part of the FDP across zones. The aim was to select the top 60% of farmers to participate in the FDP. The Partner was keen on supporting only farmers who are more likely to participate in the program and adopt good agricultural practices. Therefore, our sample is split into two groups: 60% of the farmers above the corresponding threshold within their zones (FDP treatment), and the remaining 40% farmers scoring below the threshold (FDP control).
To overcome self-selection in measuring the impact of the FDP, we exploit the fact that all farmers were scored against pre-determined criteria and were not able to ex-ante manipulate scores to participate in the FDP. The notion is that those just above and below the threshold should not reveal any systematic differences of observable characteristics. This allows for the use of the regression discontinuity design (RDD) framework to causally identify the FDP impacts.
* Spill-over impact of FDP program *
Next, we randomize the scoring threshold required for farmers to be part of the FDP. We randomize 32 zones covering the four CWS into two groups. The first group includes 16 zones with 80% invited farmers (high-intensity zones), and the second group covers 16 zones with 40% invited farmers to the FDP (low-intensity zones). Using the block-level intensity random assignment, we plan to measure the spillover effects of the FDP by controlling for the intensity of invited farmers within each zone in our analysis. In other words, we will measure the average spillover effect of being invited from a low-intensity compared to a high-intensity zone. In order to examine the spill-over effect of the FDP, we focus on the bottom quintile of farmers based on the farmer score distribution.
* Relationship Building Interventions*
Third, we measure the impact of two different organization relationship-building interventions (i.e., relationship manager and communication intervention) on the main outcomes. Farmers were randomly assigned to either of the three following relationship manager interventions: i) fixed RM, ii) Control RM, and iii) Placebo RM. Second, we randomly assign farmers to one of the following communication conditions: i) SMS only, ii) SMS + fixed RM call, iii) SMS + placebo RM call, iv) SMS + Call center call. We plan to use to elicit three effects: 1) the causal effects of having a single designated RM (Fixed), 2) the impact of having an additional (introduction) session, and 3) the impact of receiving communication from an RM.