Multilevel simulation engine for the impact of biofortified maize
A tool that simulates, at national and department level, the effect of introducing biofortified maize in Guatemala through two lenses at once: the economic one for farmers and the epidemiological one for chronic child undernutrition.
New Seed's biofortification programme needed to bring two dimensions together at once: how much the economy of producer households would improve, and how chronic child undernutrition in Guatemalan families would be affected
Introducing biofortified maize is both an agricultural decision and a public health intervention. The simulation had to support both readings on the same scenario.
Maize is the staple food of the Guatemalan diet, which is why introducing a biofortified variety has effects on two levels that are not usually analysed together. On the economic level, adopting the new seed changes the economy of producer households: what they grow, what they sell, what their own household consumes and what they reinvest. On the epidemiological level, the maize that reaches households changes children's micronutrient intake and, with it, chronic child undernutrition. The challenge was to build a single simulation able to answer both questions for the same scenario, so that the economic and health readings were consistent with each other.
On top of this, the decision is rarely made on the national average: a donor or a ministry needs to know what happens in each department, under different assumptions about coverage, seed variety and subsidy. The simulation could not deliver a single number; it had to deliver a complete space of scenarios that could be explored at the resolution at which decisions are made.
2 lenses
economic for farmers and epidemiological for children, on the same scenario
22 departments
geographical resolution on top of the national aggregate
198 scenarios
precomputed: 11 penetration levels by 3 seeds, 3 subsidies and 2 impact pathways
A reliable population base, a redistribution that respects geography and an engine that precomputes every scenario
A calibrated synthetic population to build department-level scenarios
No survey held the child's diet, their height and a large enough sample in every department at once. The simulation population is built on the children of the consumption survey (ENCOVI), which provides the diet and the country's structure, and each child receives the anthropometric block of a donor from the nutrition surveillance survey (SIVESNU) through statistical matching: a bridge model places both on a common metric, and the donor is chosen from the closest ones in the same region and age band. The key is calibration: donor selection and weights are adjusted, department by department, to the undernutrition prevalence of the maternal and child health survey (ENSMI), rather than simply multiplying records.
A redistribution of produce that respects real geography
Because maize is not necessarily consumed where it is grown, the surplus from producing departments is redistributed to deficit departments with a gravity model: the flow increases with the deficit at the destination and decreases with the travel time between the two. That friction is calculated on an asymmetric matrix of real driving times between the 22 departmental capitals, not on straight-line distances. The formulation follows Anderson and van Wincoop (2003), a standard for interregional trade flows, and conserves total production throughout the allocation.
A parameterised engine that precomputes the scenario space
On those two foundations, adoption is modelled as a continuous process in which each farmer starts adopting at a point in the ranking and converts a growing share of their production as coverage increases. The engine solves the entire decision grid in advance — eleven penetration levels from 0% to 100%, three seed generations, three subsidy regimes and the two pathways for estimating health impact — in both its national and department-level versions. Precomputing, rather than calculating on the fly, is what makes exploration instant.
A dynamic Shiny dashboard for multilevel decision-making, not a static report
Both readings, economic and health, on every scenario
For any combination of coverage, seed and subsidy, the engine delivers both the economic effect on farmers and the effect on chronic child undernutrition. Both readings rest on the same scenario, so New Seed weighs each decision with the agricultural and health arguments together and consistent with each other, rather than in two separate analyses.
Modelled estimates at department level
Departments with lower representation in the survey are brought into the simulation through a synthetic population calibrated against the departmental reference prevalence. This makes it possible to compare scenarios across areas under explicit assumptions, without presenting the modelled estimates as direct observations.
Self-service exploration for non-technical decision-makers
The entire scenario space is delivered in an interactive app that translates the model's projections into clear readings for donors and management teams. They can explore combinations and compare areas without handling the underlying model: the analysis becomes a tool the client uses on its own, not a deliverable that only its builder understands.
What this case shows about the way we work
Two lenses in a single model
Integrating the economic reading for farmers and the epidemiological reading for children into a single simulation, rather than two separate studies, is what makes it possible to decide with the full picture.
Synthetic data with judgement, not statistical padding
Generating a synthetic population is not multiplying rows: it is matching with judgement and calibrating against an independent reference. It makes it possible to build scenarios in areas with low sample representation, preserving the observed structure without mistaking synthetic records for real observations.
Geography, inside the model
Modelling the redistribution of produce with real travel times over the road network, rather than theoretical distances, is what makes the geographical impact credible. Anchoring the formulation in an established trade-flow model avoids simulating an impact that real-world logistics would never allow.
The culmination of an analytical programme
The simulation is not a stand-alone piece: it brings together the farmer characterisation, the intake estimation and the undernutrition impact model in a single coherent engine, where each component feeds the next. It is the point at which all the previous work becomes a decision-making tool.
I've been working with Rubén and Julia on a complex modeling project for over a year now, and they have been exceptional partners throughout. As our project has evolved, they have approached every shift collaboratively, receptively, and with flexibility. They have consistently gone above and beyond to ensure we develop both a robust model and a user interface that translates model predictions into clear insights. What we have now is a fully documented, auditable methodology we can put in front of funders and external reviewers with confidence.
Katie Adams
Director of Impact and Cost Effectiveness, New Seed
Capabilities deployed in this project
Shall we talk?
Tell us what you are working on and we will give you a straight answer on how we can help.