Climate Change Mitigation and Adaptation in the Supply Chain: Agroforestry with Coffee Farmers in Rwanda

In partnership with a multinational coffee buyer, we design, implement, and evaluate the impact of financial incentive schemes on smallholders’ planting and survival of shade trees within the Rwanda coffee supply chain.

First, we examine the take-up and impact of different financial incentive schemes on planting and survival of shade trees. In both schemes, farmers earn incentives based on the number of trees planted and/or surviving. In Treatment 1, incentives are paid based on number of shade trees planted. In Treatment 2, farmers are eligible a smaller planting incentive and for a second incentives conditional on shade tree survival, and delivery to the Partner. Additionally, within treatment 2, we cross-randomize whether farmers receive a booklet to track their progress within the agroforestry program.

Second, we randomize information provided to farmers about the process of receiving payment and/or verification based on planting. We measure the impact on likelihood of call center engagement and planting.

Third, we cross-randomize low-cost verification approaches—leveraging call-center follow-ups to detect misreporting—to reduce plot verification costs, a key bottleneck to scaling agroforestry.

We roll out the intervention across five coffee washing stations (CWS) in Rwanda. The study builds on earlier collaborations in this supply chain and will generate practical guidance on implementing agroforestry programs that are credible, cost-effective, and scalable.

2025-10-23

2026-09-30

Take-up of scheme; Take-up of shade trees (n); Planting of shade trees; Survival of shade trees

1. The first primary outcome is the take-up of the scheme. This is a binary variable that indicates whether, after attending the training for agroforestry, a farmer is willing to take-up the scheme offered and commit to planting shade trees, or whether the farmer does not accept the scheme.

2. The second primary outcome refers to the number of shade trees the farmer is willing to take-up or collect at the point of signing up for the scheme. This number will be recorded by a surveyor who conducts distribution directly after the farmer signs-up for the scheme.

3. The third primary outcome, planting of shade trees, refers to the number of non-Grevillea shade trees that the coffee farmer has planted on their coffee plots. This measure will be collected in January 2026. We will have both a binary and continuous measure to reflect whether the farmer planted and how many trees the farmer planted.

4. The fourth primary outcome, survival of shade trees, refers to the number of non-Grevillea shade trees surviving on their coffee plots. This measure will be collected after dry season (2026, 2027, and 2028).

The last two variables will be measured through a) self-reported data through phone calls and in-person surveys, and b) by sending an agronomist to farmer's plots to verify the actual number of shade trees planted/surviving. We will create two variables, a binary indicating whether the farmers planted any trees/any trees surviving and a count variable indicating the number of trees planted/surviving.

Coffee cherry delivery (KG); Farmer communication through call center (farmer inbound call and topic of calls); Acceptance rate of counteroffer; Farmer understanding of scheme (clarity); Farmer trust in CWS and partner (credibility)

1. Coffee Cherry Delivery: This is a continuous variable indicating the amount of coffee delivered by farmers to the Partner’s CWSs during the 2025 harvest season between February and August. It is based on the administrative data on coffee delivery. To avoid results’ sensitivity to extreme values or data entry errors, the coffee delivery amount will be winsorized at the 99th percentile. The source of this data is the transactions records maintained by the Partner. The data is recorded automatically on the CWS data system. For each unique delivery made by a farmer, the data records the farmer’s unique ID, whether the coffee was delivered at the CWS or at a Site, the amount of coffee delivered in KGs, the price per KG paid to the farmer, and whether the farmer received a transport premium.

2. Farmer communication through call center. During the different stages of the project, we will construct both binary and a continuous variables indicating whether and the frequency at which an individual farmer has called within a designated timeframe (e.g., between signing the contract and planting verification, between planting and survival verification). Additionally, we will create variable capturing the topics for which the farmer called the call center.

3. Acceptance rate of counteroffer. A binary variable indicating whether the farmer accepted the lower counteroffer made before planting verification. This variable will be missing for those farmers who did not receive such an offer.

4. Farmer understanding of scheme (clarity). A variable capturing farmer’s understanding of the program, benefits, milestone, and requirement captured through unaided recall survey questions.

5. Farmer trust in CWS and partner (credibility). A variable capturing farmer’s trust in willingness and ability of partner to provide farmers with promised incentives conditional on compliance with the program.

In this study, we examine the impact of cash-based incentives for the planting and/or survival of shade trees on their take-up. Second, we examine the impact of information provided to farmers about the process of receiving payment and/or verification on actions taken within the program. Third, we explore the impact of low-cost verification approaches to reduce plot verification costs.

Our empirical strategy focuses on the following elements: i) the (differential) impact of agroforestry incentive schemes on planting and survival of shade trees, ii) the impact of information regarding steps to request payment/verification on the likelihood to call the call center and the planting of shade trees, and iii) low-cost verification approaches that detect misreporting to reduce plot verification costs, and to what extent both misreporting is detected.

Random assignment for financial incentives was conducted at the individual-level, at different intensities at the village level, stratified by village, gender, farm size, and participation in the stumping experiment. Importantly, a farmer’s random assignment was only revealed to them after they had arrived at the distribution site, completed agroforestry training, and expressed interest in collecting free shade trees.

Additionally, for the request intervention, Randomization is stratified at the level of Site/CWS x Day x Time of Day. For the counter-offer intervention, we stratify random assignment by CWS x Gender × Farm Size × Participation in Stumping in 2024 x Request Calling Status (1/0) x Type of Request (Payment/Verification).

Not available

Randomization done in office by a computer using software STATA.

1. Financial Incentive Intervention: Randomization is at the individual farmer-level, stratified by Village × Gender × Farm Size × Participation in Stumping in 2024.

2. RA Intervention: Randomization is at the individual-farmer level, cross-randomized with T2 of the Financial Incentives Intervention, stratified by Village × Gender × Farm Size × Participation in Stumping in 2024 x Distance from CWS.

3. Payment/Verification Request Intervention: Randomization is at the level of the distribution session. Distribution takes place at a total of 22 local sites + 5 CWS = 27 sites. At each local site, distribution is conducted over 3-9 consecutive days (18 consecutive days at CWS), in two batches (morning and afternoon). Randomization is stratified at the level of Site/CWS x Day x Time of Day.

4. Counter-offer Intervention: Randomization is at the individual-farmer level, stratified by CWS x Participation in Stumping in 2024 x Type of Request (Payment/Verification) x Number of Packages Collected.

No

1

3,600 (expected)

Financial Incentives Intervention (N = 3,600):
1. Control Group (30%) – 1,080
2. T1 (30%) – 1,080
3. T2 (40%) – 1,440

RA Intervention (N = 1,440):
1. Reverse Accountability (RA) – 720
2. Non-RA – 720

Request Calls Intervention (N = 3,600):
1. Payment Request – 1,800
2. Verification Request – 1,800

Counter-offer Intervention (N = 1,200):
1. Control Group (40%) - 480
2. Low xM (20%) - 240
3. High xM (20%) - 240
4. Honesty Appeal (20%) - 240

We provide the MDE for the ITT impact on the main outcomes: planting, survival and deliveries. We assume a take-up rate of 95 percent among farmers who attend distribution. The MDE for the outcome of PLANTING (= 0/1) is as following (assuming a control group mean of 0.31): 1. Comparing CG to T1: 0.054 2. Comparing CG to T2: 0.051 3. Comparing T1 to T2: 0.051 The MDE for the outcome of SURVIVAL (= 0/1) is as following (assuming a control group mean of 0.84): 1. Comparing CG to T1: 0.048 2. Comparing CG to T2: 0.045 3. Comparing T1 to T2: 0.045 The MDE for the outcome of DELIVERIES (in KGs/season) is as following (assuming a control group mean for the last periods to be 205 KGs (2025), 190 KGs (2024), 109 KGs (2023), 262 KGs (2022), 191 KGs (2021), and 108 KGs (2020); and assuming 3 pre-period and 1 post-period): 1. Comparing CG to T1: 86 KGs/season 2. Comparing CG to T2: 82 KGs/season 3. Comparing T1 to T2: 82 KGs/season