Digital Traceability for Deforestation-Free Supply Chains: Evidence from Cocoa Production in Ghana

Last registered on May 18, 2026

Pre-Trial

Trial Information

General Information

Title
Digital Traceability for Deforestation-Free Supply Chains: Evidence from Cocoa Production in Ghana
RCT ID
AEARCTR-0018649
Initial registration date
May 14, 2026

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
May 18, 2026, 7:26 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

Affiliation
University of Connecticut

Other Primary Investigator(s)

PI Affiliation
University of Connecticut
PI Affiliation
Wageningen University
PI Affiliation
University of Ghana
PI Affiliation
University of Geneva
PI Affiliation
University of Geneva

Additional Trial Information

Status
In development
Start date
2026-05-20
End date
2027-01-31
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
Global commodity supply chains face mounting pressure to verify deforestation-free status through property-level traceability, yet feasibility and effectiveness in low-income settings remain uncertain. We conduct a randomized controlled trial across 90 villages in Ghana's cocoa belt, partnering with a large farmers' cooperative to evaluate satellite-based deforestation monitoring. We develop a mobile application that digitalizes the cooperative's supply chain, automates data collection, and verifies farmers' deforestation status at the point of sale using high-resolution satellite alerts. Villages are randomized to (i) control, (ii) monitoring with visible certification, or (iii) monitoring with price premiums, isolating transparency effects from financial incentives. To quantify leakage, we measure deforestation at three spatial scales: inside mapped properties, buffer zones around properties, and village boundaries. Surveys capture cocoa output and quality, farmer awareness and understanding of monitoring, and willingness to participate under alternative contract designs. The experiment identifies compliance effects, spatial leakage, and farmer demand for certification in a major cocoa-producing region, informing the design of scalable deforestation-free supply chain policies.
External Link(s)

Registration Citation

Citation
Akandinge, George et al. 2026. "Digital Traceability for Deforestation-Free Supply Chains: Evidence from Cocoa Production in Ghana." AEA RCT Registry. May 18. https://doi.org/10.1257/rct.18649-1.0
Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
Experimental Details

Interventions

Intervention(s)
We partner with a large farmers' cooperative in Ghana representing over 100,000 cocoa farmers to test whether satellite-based deforestation monitoring can reduce smallholder deforestation. We develop a mobile app that digitalizes the cooperative's supply chain and integrates real-time deforestation verification at the point of sale. The cooperative operates through purchasing clerks who buy cocoa directly from farmers at designated purchasing sheds.

The app uses weekly high-resolution satellite alerts from Global Forest Watch to determine farmers' deforestation status. Purchasing clerks use the app to record transaction details and administer surveys. The app generates QR-coded labels for cocoa bags that enable full supply chain traceability from farmer to handoff to the Ghana Cocoa Board (COCOBOD) for export.

The intervention follows Ghana's cocoa season. In late May 2026, purchasing clerks deliver information interventions to farmers, communicating their zone's treatment status so farmers can potentially adjust land-use decisions during planting and growing seasons. Between October and December 2026 (harvest), when farmers bring cocoa to purchasing sheds for sale, clerks use the app to check farmers' deforestation status based on satellite alerts. In Treatment 1 and Treatment 2 zones, clerks print and attach certification tags to cocoa bags indicating deforestation-free status. In Treatment 2 zones only, farmers receive per bag premium payments for deforestation-free cocoa.
Intervention Start Date
2026-05-20
Intervention End Date
2026-12-31

Primary Outcomes

Primary Outcomes (end points)
On-property deforestation
Deforestation within property buffers (local leakage)
Zone-level deforestation (broader leakage)
Primary Outcomes (explanation)
Primary outcomes measure direct environmental impacts at three spatial scales. On-property deforestation measures direct compliance: whether monitoring and premiums reduce deforestation on monitored parcels. Deforestation within buffer zones tests for spatial leakage immediately around property boundaries. Zone-level deforestation tests whether monitoring reduces overall deforestation within village boundaries or merely displaces it spatially from monitored to unmonitored areas.

Detailed explanation:
On-Property Deforestation: Our primary outcome is a binary indicator for whether any high-confidence deforestation alert occurred on a farmer's mapped property during the intervention period (October-December 2026). We use 10×10m resolution weekly deforestation alerts from Global Forest Watch, which processes Landsat and Sentinel-1 and Sentinel-2 imagery to detect tree cover loss.

Buffer Zone Deforestation: To test for spatial leakage immediately around monitored properties, we construct buffers around each mapped property boundary and create a binary indicator for whether any deforestation alert occurred within these buffer zones, excluding the property itself.

Zone-Level Deforestation: We measure deforestation at the zone level (where zones correspond to a village or cluster of villages that sell cocoa to a single purchasing shed) to capture broader spatial leakage. We construct spatial boundaries for zones using Voronoi tessellations around zone centroids. We then divide each zone into 1-hectare grid cells (each containing 100 10×10m pixels) and count the number of deforestation alerts occurring within each grid cell over the study period. We exclude mapped properties and their buffer zones from this count, isolating broader deforestation spillovers.

Secondary Outcomes

Secondary Outcomes (end points)
Cocoa production (number of bags sold per farmer)
Cocoa quality (re-bagging due to moisture content)
Farmer compliance and sustainability practices
Cooperative participation and sales behavior
Willingness-to-accept deforestation-free premium payments
Risk tolerance for variable premiums
Tolerance for enforcement
Secondary Outcomes (explanation)
Secondary outcomes capture behavioral and economic responses to the intervention. Valuation outcomes inform the design of scalable certification programs by identifying which contract features farmers value most and what premium levels are necessary to maintain participation.

Detailed explanation:
Cocoa Quality: Cocoa quality is measured at the district warehouse, where managers measure moisture content using handheld sensors to determine whether bags require re-bagging. The re-bagging procedure destroys property-level traceability by mixing cocoa from the affected bag with other farmers' cocoa during drying and re-sealing. Re-bagged bags become ineligible for deforestation-free premiums, since premiums are paid conditional on maintaining traceability from shed to handoff to the Ghana Cocoa Board (COCOBOD). This outcome tests whether monitoring or premiums induce complementary improvements in cocoa quality.

Cooperative Recruitment, Retention, and Sales Volume: We measure cooperative membership status and total bags sold to the cooperative over the season for each farmer. Treatment 2's deforestation-free premiums may attract new members to the cooperative or increase sales from existing members who might otherwise sell to competing buyers. Conversely, Treatment 1's monitoring without premiums could deter farmers who wish to avoid reputational costs. These outcomes test whether deforestation-free certification serves as a recruitment tool and whether farmers value the premium enough to consolidate sales with a single buyer.

Adaptive Responses: We measure farmer adaptation strategies that may emerge in response to deforestation monitoring and premiums, including likelihood of household members working off-farm, planting new cocoa trees, practicing agroforestry by maintaining shade trees on properties, and stated intentions to reduce or exit cocoa farming. These outcomes capture whether certification leads households to shift toward alternative livelihoods or adopt less deforestation-intensive production practices.

Willingness-to-Accept: The survey elicits farmer willingness to accept deforestation monitoring under alternative contract structures. For all farmers, we elicit willingness to participate in (or continue) deforestation monitoring by randomly assigning hypothetical premium offers drawn from a uniform distribution of integers between $0 and $15 per bag. Unmapped farmers are asked whether they would agree to have their farms mapped and monitored at the assigned premium. Mapped farmers are asked about their willingness to participate in (control) or continue (T1 and T2) monitoring at the assigned premium.

Risk Tolerance: We follow up with a risk tolerance question: Would you participate in monitoring and certification if the premium averaged the same amount but varied between $0 and twice that amount? These questions test whether farmers discount uncertain payments relative to fixed premiums of equal expected value, revealing risk preferences that determine the feasibility of market-based premiums.

Tolerance for Enforcement: For Treatment 2 farmers, we measure tolerance for enforcement by asking the minimum premium they would accept if the penalty for detected deforestation changed from loss of premium only to exclusion from selling to the cooperative for the remainder of the season. Response options range from $0 to $20+ or unwillingness to participate at any premium, informing whether cooperatives can implement credible sanctions without damaging supplier relationships.

We will also collect additional descriptive data on farm characteristics, farmer understanding of deforestation monitoring, and purchasing clerk experiences with the app:
Farm Characteristics: We measure total farmed land and land dedicated to cocoa cultivation, number of plots managed, and labor sources
Deforestation and Sustainability Awareness: We assess farmers' awareness of nearby forest boundaries, understanding of "deforestation-free cocoa," and familiarity with monitoring and traceability. We also measure farmers' ranking of deforestation relative to other challenges they face
Treatment Awareness and Comprehension: We measure whether purchasing clerks explained the experiment and treatment assignment to farmers. For mapped farmers, we assess their understanding of monitoring technology.
Purchasing Clerk Feedback: In January 2027, after data collection concludes, we will administer a survey to all purchasing clerks who participated in the experiment. The survey will assess practical implementation challenges, including tablet and printer reliability and time burden per transaction. For T2 clerks, we will assess difficulties explaining the premiums and processing payments.

Experimental Design

Experimental Design
We conduct a three-arm randomized controlled trial across 90 cocoa purchasing zones (roughly corresponding with villages) in Ghana's cocoa-producing regions. Zones are randomized to: (i) Control (data collection only), (ii) Treatment 1 (satellite monitoring with visible certification stickers on cocoa bags), or (iii) Treatment 2 (monitoring with per bag premiums for deforestation-free cocoa). This design unbundles monitoring effects from financial incentives.

We measure deforestation at three spatial scales to quantify both compliance and leakage: (i) on mapped properties (direct compliance), (ii) within buffer zones around properties (local leakage), and (iii) at the zone level (broader spatial leakage). We also measure secondary outcomes including cocoa production, quality, cooperative participation, and farmer behavioral responses. Survey modules elicit farmer valuation of certification through willingness-to-accept questions with randomly assigned hypothetical premiums, risk tolerance for variable versus fixed premiums, and tolerance for stronger enforcement.

Prior to randomization, we selected 9 districts spanning the cooperative's operating area. Within each district, we randomly selected an average of 10 zones with sufficient georeferenced farm properties, resulting in 90 experimental zones. We then randomized the 90 zones into Control (30 zones), Treatment 1 (30 zones), and Treatment 2 (30 zones), stratified by district to ensure balance across the cooperative's operating area.
Experimental Design Details
Not available
Randomization Method
Randomization was conducted in Stata using random draws from a uniform distribution, using "set seed" for reproducibility. Zones were stratified by district and randomly assigned to Control, Treatment 1, or Treatment 2 with equal probability (30 zones per arm).
Randomization Unit
The unit of randomization is the purchasing zone (village), with all farmers in each zone assigned to the same treatment arm.
Was the treatment clustered?
Yes

Experiment Characteristics

Sample size: planned number of clusters
90 purchasing zones (villages)
Sample size: planned number of observations
3,600 (approximately 40 mapped farmers per zone x 90 zones)
Sample size (or number of clusters) by treatment arms
30 zones control, 30 zones Treatment 1 (satellite monitoring only), 30 zones Treatment 2 (satellite monitoring with premium payments)
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Minimum detectable effects for primary outcomes are: (i) on-property deforestation: 0.147 SD (0.021 alerts per property, ICC=0.00664), (ii) buffer-zone deforestation: 0.17 SD (0.459 alerts per 200m buffer, ICC=0.01926), and (iii) zone-level deforestation: 0.028 SD (0.017 alerts per hectare, ICC=0.00117, average 3,338 grid cells per cluster). These calculations use intra-cluster correlations from five years of historical satellite alert data (2020-2025). The design enables detection of small treatment effects on deforestation (0.028–0.17 SD; 0.017–0.459 alerts), with MDEs well below a single additional alert for primary outcomes. For a secondary outcome of bags sold per farmer, the MDE is 0.39 SD (2.66 bags, ICC=0.26715), requiring slightly larger effects to detect changes in sales behavior.
Supporting Documents and Materials

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
IRB

Institutional Review Boards (IRBs)

IRB Name
Commission Universitaire pour une Recherche Ethique à Genève (CUREG)
IRB Approval Date
2026-02-06
IRB Approval Number
CUREG-2025-08-133