Relational Sourcing and Tree Rejuvenation Schemes in the Rwanda Coffee Sector

In partnership with a leading coffee exporter, we design, implement and evaluate the impact of incentive schemes on tree stumping behaviour by farmers in the Rwanda coffee chain. In both schemes, farmers receive financial incentives for stumping based on the number of trees they stump. Conditions and timing of payment vary across the different schemes.

Across three related experiments, we first examine the take-up of schemes and the choice of scheme by farmers. Second, we evaluate the effectiveness of different schemes on the adoption of tree-stumping practices. Third, we evaluate the role of a relationship-building program on the choice of scheme and the effectiveness of schemes. Additionally, we cross-randomize a digital intervention and evaluate its effectiveness in improving take-up of schemes and adoption of tree rejuvenation.

In this (pilot) project, we roll out a tree stumping intervention in seven coffee washing stations (CWS) in Rwanda. The intervention builds on an earlier pilot (AEARCTR-0008725) in which we set up a farmer development program (FDP) aimed to build trusting relationships between coffee farmers and the company.

2023-07-24

2023-09-30

1. Take-up of scheme
2. Choice of scheme
3. Adoption of tree stumping & number of trees stumped

Take-up of scheme, is a binary variable that indicates whether the farmer is willing to take-up the scheme offered and commit to stumping part of their eligible trees, or whether the farmer does not accept the scheme.

Choice of contract, refers to the type of contract that farmers will choice when offered a choice; upfront menu or the relational menu. We will measure this categorical outcome variable by recording the farmer's final choice in Survey CTO.

The third primary outcome refers to the number of trees stumped in the given period (i.e., June - September 2023). We will measure this variable through a) self-reported survey data and b) by sending an agronomist to farmer's plots to verify the actual number of trees stumped. We will create two variables, a binary indicating whether the farmers adopted stumping and a count variable indicating the number of trees stumped.

1. Coffee cherry delivery (KG)
2. Trust in CWS and Partner
3. Take-up of Lime & Fertilizer

Coffee cherry delivery - This is a continuous variable indicating the amount of coffee
delivered by farmers to CWSs during the 2024 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 administrative. The data is recorded automatically on the CWS data system. The data includes the amount of coffee delivered to CWS by individual farmers, the price paid to farmers, the number of transactions, whether a farmer received a transport premium. We do not expect to find a short-term impact of the tree rejuvenation schemes on the delivery volumes coffee. Coffee delivery volume is treated as a long-term outcome, given that implementation of stumping techniques will only result in increase produce volume over an average of 2-3 years.

Trust in CWS - These are indices that take a value between 0 and 10, where 0 reflects no
trust at all and 10 indicates full trust. These indices involve trust in CWS RM, CWS promises,
CWS management, and CWS farmers.

Take-up of Lime and Fertilizer – a separate binary variable for lime and fertilizer indicating whether the farmer came to the CWS to pick-up freely distributed inputs in September 2023 essential to tree stumping.

This study partly builds on an earlier pilot in which we designed and evaluated a farmer development program (FDP) aimed at building relationships between farmers and the company.

Across three related experiments, our empirical strategy focuses on the following elements: i) take-up of tree rejuvenation schemes, ii) the (differential) impact of tree rejuvenation incentive schemes on tree stumping, iii) the impact of FDP on the effectiveness of choice and take-up of scheme, and iv) the effect of a digital support intervention.

We stratify random assignment within each individual CWS and over phone ownership, gender, FDP status, traceability indicators, and distance to CWS and relative distance to partner and other competitors.

Experiments A & B are conducted in four CWS, Musasa, Karenge, Kibirizi and Kayumbu, in which two years of FDP have been implemented. In these CWS, we launched a farmer loyalty program and informed farmers that they may become eligible for benefits, such as tree stumping, based on their performance. See AEARCTR-000872 for details on this intervention.

After harvest season 2023 (June 2023), farmers in catchment areas of these four CWSs were ranked based on their performance. Key performance indicators include volume of coffee delivery, traceability of coffee delivery (i.e., usage of farmer card when delivering), loyalty in delivering across last two years, and engagement with partner (i.e., phone number verified). The top 15% of farmers were assigned 'gold' status, the following 35% of farmers received 'silver' status, and the bottom 50% received 'bronze' status. Experiment A focuses on the silver and gold farmers, and Experiment B focuses on the bronze farmers in these four CWS.

In addition to the standard variables mentioned above, randomization for Experiment A & B was also stratified on zone intensity (i.e., zones in which top 80% farmers in FDP or top 40% farmers where assigned to FDP), FDP tier 2023, and phone ownership.

Experiment C is conducted in three different CWSs (i.e., Mpanga, Karambi, Nyamyumba). FDP was introduced to 50% of the farmers in the catchment areas of these CWS in January 2023. In these CWS, all farmers, regardless of position relative to a certain threshold based on size and delivery volume, were not excluded. In Experiment C, all farmers are offered a choice between the relational scheme, upfront scheme, or no scheme. Additionally, we cross-randomize digital support intervention (Tx).

In addition to the standard variables mentioned above, randomization for Experiment C was also included Village Treat Intensity (75% farmers in FDP or 25% farmers in FDP), FDP assignment, FDP tier 2023 and phone ownership.

We plan to use to elicit two effects: i) the (differential) effect of incentive schemes on tree rejuvenation adoption and ii) the impact of FDP on choice and take-up of incentive schemes. Additionally, we aim to evaluate the additional effect of a digital support intervention.

By comparing choice of scheme and take-up between FDP and non-FDP farmers in Experiment C, we are able to estimate the impact of FDP on the take-up of relational scheme. By comparing farmers offered the relational scheme to farmers offered the upfront scheme in Experiment A, we are able to estimate the differential effect of relational compared to the upfront scheme. By comparing farmers offered the relational scheme to no scheme in Experiment B, we can estimate the effect of the relational scheme compared to the status quo on tree stumping.

Last, we will present both the intention-to-treat (ITT) effects (i.e., impacts of being assigned to a scheme) and the treatment effects on the treated (TOT) (i.e., impacts of taking-up a scheme).

Randomization done in office by a computer using software STATA.

Randomization was individual Farmer-level for all treatments.

Experiment A - Farmers were stratified by the following:

“CWS X Intensity of FDP in Zone X Gender X FDP Tier X Certification Survey Done X Clarity-Credibility Calls Receipt X Relative Distance to Partner’s Coffee Collection Site X Relative Distance to Competitor’s CWS”

Experiment B - Farmers were stratified by the following:

“CWS X Intensity of FDP in Zone X Gender X Certification Survey Done X Clarity-Credibility Calls Receipt X Phone Ownership X Relative Distance to Partner’s Coffee Collection Site X Relative Distance to Competitor’s CWS”

Experiment C - Farmers were stratified by the following:

“CWS X Intensity of FDP in Village X Gender X FDP Eligibility (=0,1,2) X FDP Tier X Certification Survey Done X Clarity-Credibility Calls Receipt X Phone Ownership X Relative Distance to Partner’s Coffee Collection Site X Relative Distance to Competitor’s CWS”

Additionally, in experiment C, FDP Status was randomized at individual level in January 2023.

No

1

Experiment A: 1477 farmers Experiment B: 1482 farmers Experiment C: 3045 farmers

Experiment A:

T1: 377
T2: 374
T3: 367
T4: 359

Experiment B:
C: 767
T1: 344
T3: 371

Experiment C: 3045 farmers
Choice only: 1522
Choice + Tx: 1523

Power is estimated for a two-way comparison for each experiment separately. We describe below the MDE for ITT impact on following main outcomes. Experiment A: 1. For Extensive margin take-up of tree stumping: MDE is 4.9%age points (baseline rate of 13.2%) 2. For average percent trees stumped: MDE is 2.9%age points (on a baseline rate of 6.5%) 3. For average number of trees stumped: MDE is 33.1 trees, (on a baseline rate of 53 trees) Experiment B: 1. For Extensive margin take-up of tree stumping: MDE is 3.9%age points (baseline rate of 4%) 2. For average percent trees stumped: MDE is 2.25%age points (on a baseline rate of 3.8%) 3. For average number of trees stumped: MDE is 30.2 trees, (on a baseline rate of 23.1 trees) Experiment C: 1. For Extensive margin take-up of tree stumping: MDE is 2.4%age points (baseline rate of 6.2%) 2. For average percent trees stumped: MDE is 1.1%age points (on a baseline rate of 2.1%) 3. For average number of trees stumped: MDE is 15.8 trees, (on a baseline rate of 16 trees)