This repository contains the code, data, and documentation for a Bayesian Network model developed to support decision-making around a nutritional intervention in Kenya. The project aims to assess the impact of introducing fruit trees on smallholder farms on household nutrition, specifically focusing on gaps in dietary energy, iron, zinc, and provitamin A.
This model is the product of a collaborative knowledge co-production process with domain experts, including government officials, nutritionists, agricultural practitioners, and farmers. It translates a detailed impact pathway into a probabilistic model to forecast outcomes under different decision scenarios and evaluate the value of acquiring additional information.
Key Features
Probabilistic Decision Support Model: A Bayesian Network implemented in AgenaRisk, representing the causal relationships between the fruit tree intervention and nutritional outcomes.
Explicit Uncertainty Quantification: All model parameters incorporate expert-elicited probabilities, explicitly accounting for uncertainty in the system.
Value of Information (VOI) Analysis: Includes scripts/results for calculating the Expected Value of Perfect Information (EVPI) to identify which variables, if better understood, could most improve decision confidence.
Structured Workflow: The repository is organized according to the five-step methodological framework described in the accompanying paper.
Model Details
The model is designed to compare two primary scenarios:
Baseline: Farms without the fruit tree intervention.
Intervention: Farms where fruit trees (e.g., papaya, mango) have been planted.
The key outcome (utility) nodes are the annual per-person gap for four nutrients:
Dietary Energy (kcal)
Iron (mg)
Zinc (mg)
Provitamin A (Retinol Activity Equivalents - RAE)
The goal of the decision analysis is to minimize these nutritional gaps.
How to Use This Repository
Understand the Framework: Begin by reading the /docs to familiarize yourself with the impact pathway and the workshop process.
Explore the Model: Open the .agn model file in AgenaRisk to visualize the network structure, inspect node probability tables, and run simulations.
Run Scenarios: Within AgenaRisk, you can:
Set the decision node (e.g., "Fruit Tree Intervention") to different states.
Observe the resulting probability distributions for the nutritional gap nodes.
Determine the decision option that minimizes the expected nutritional gap.
Conduct VOI Analysis: Use the built-in VOI tools in AgenaRisk to replicate the EVPI analysis. Set the optimization goal to "Minimize" the expected value of the utility nodes (nutritional gaps) to identify critical uncertainties.
Important Notes
Software Requirement: The core model requires AgenaRisk Professional software, which is a commercial product. The free AgenaRisk Viewer may allow for inspection but not for running new analyses.
Expert-Dependent Parameters: Many model parameters are based on expert knowledge from a specific context (Kenya, specific farming communities). Caution should be exercised when applying this model directly to other regions or populations without re-eliciting local expert knowledge.
Transparency: The use of the Hansson & Sjökvist method for CPT generation is documented to ensure transparency in how expert judgment was translated into model parameters.
Contributors
We thank the team of experts, analysts, and farmers in Kenya whose knowledge was essential to building this model. The work was done in collaboration with the Center for International Forestry Research and World Agroforestry (CIFOR-ICRAF). The work was financially supported by the Innovative Metrics and Methods for Agriculture and Nutrition Actions program (IMMANA; project number 1.36), funded by the UK Department for International Development (DFID). We also acknowledge support from the ‘Water, Land and Ecosystems’ research program of the Consultative Group on International Agricultural Research (CGIAR).
For questions or support, please open an Issue in this GitHub repository.