ICRAF (Probabilistic Causal Models)
World Agroforestry Centre (ICRAF)
Probabilistic Causal Models for Nutrition Outcomes of Agricultural Actions
PI: Eike Luedeling
Partners: Keith Shepherd, Ramni Jamnadass, Christine Jost (all ICRAF), Norman Fenton (Queen Mary, University of London), Michael Krawinkel (University of Gießen), Christian Borgemeister (Centre for Development Research), Martin Neil (Agena Ltd)
Start date: 1 September 2015
Duration: 24 months
Countries of research: Kenya, UK, Germany
Summary of project:
Decision-makers in international development often struggle to allocate resources to where they are most effective due to the difficulty of accurately projecting intervention impacts. They also often have difficulties in developing effective implementation plans, monitoring progress and evaluating project impacts. These difficulties derive from a number of factors, including lack of data, complex impact pathways, and risks and uncertainties that are difficult to factor into intervention planning. Scientific approaches to produce reliable impact projections are rarely applied in agricultural development, but decision analysis techniques commonly used in other fields have potential to improve development decisions. We propose to adapt such an approach to the needs of agriculture for nutrition interventions and apply it to two case studies.
The procedures we plan to develop will feature the construction of causal models – models that describe the mechanisms through which impact will be delivered – that are co-developed by experts, stakeholders and analysts through facilitated participatory processes. Models will be formalised as Bayesian Networks (BNs), a modelling approach that has been widely applied in a range of disciplines, including medical sciences, genetics, environmental sciences, and legal reasoning. BNs allow formal representation of causal models, such as intervention impact pathways. They can work effectively with incomplete information, combine expert knowledge with other sources of information, and they allow adequate consideration of risk.
We will apply our approach in two case studies: 1) introduction of fruit trees into agricultural landscapes of Eastern Kenya; and 2) conversion of traditional home garden systems in Uganda to large-scale staple crop production.
For both case studies, participatory workshops will convene experts on the systems, stakeholders involved in ongoing or prospective projects, and analysts skilled in facilitation and model building techniques. This team will jointly develop impact pathways for the interventions, which will be formalised into quantitative BN models. After several rounds of feedback elicitation, and inclusion of data from experts and other sources, stochastic simulations will be run to determine the likely impacts of the interventions on nutrition, disaggregated by gender and other social criteria. Results will be presented back to stakeholders for feedback. We will determine critical uncertainties in the models, which should be addressed through measurements or modified intervention designs. These are high-value variables that determine uncertainty about project outcomes, and their measurement can greatly support decision-making processes. All model results will be written up as reports and policy briefs to present the business case for the two interventions to decision-makers and stakeholders. Since the model building team will include members who are directly involved in agricultural development, we are confident that our results will be applied. We will also compose a manuscript for publication in a peer-reviewed open-access journal. The impact modelling procedures will be described in detail in a process manual to allow their replication elsewhere. By demonstrating improved intervention decisions with minimal additional investment and improved tools for intervention decision modelling, we have confidence that our approach will be widely adopted and used to enhance the efficacy of development activities for nutrition.