Scenario simulation and policy evaluation

Scenario simulation means building an engine that estimates the impact of different configurations of a policy or intervention before it is implemented. The engine works through the decision space, defined by variants, intensities and coverage levels, and returns, for each combination, its estimated effect on the indicators that matter. It applies when you need to decide how to configure an intervention before committing resources to its roll-out, and testing it in the field to see whether it works would be the most expensive experiment there is. Rolling out an intervention and waiting for the result costs years and budget, and if the configuration chosen was not the most suitable, correcting it afterwards can be far more costly. Simulating the scenario space lets you compare configurations and rule out the least promising ones without paying to learn the lesson in the field. What the service provides is a quantified scenario space to decide on, not a single assumption defended in advance.

It is for any intervention whose impact can be parameterised and depends on how it is configured and how it reaches the population: the simulation engine is domain-agnostic. It applies to food, medicines, public information, surveillance or public health policies. The machinery is the same whatever the field: characterise a baseline, work through a grid of configurations, model diffusion and aggregate indicators. What changes from one domain to another is the nature of the intervention and of the indicators, not the logic of the engine. A campaign circulating through a social network and a good distributed by road are simulated with the same framework and a different diffusion module. That generality is deliberate: the service is defined by the question, which configuration of the intervention performs best, and not by a particular sector. What you receive is an engine adapted to your intervention, applicable across domains.

The simulation runs on a baseline that characterises the population, supply and demand, as far as they are known or can be estimated for the decision context the intervention acts on, not on an assumed homogeneous population. This applies to any simulation, because the result depends on structural heterogeneity, whether geographic, demographic or in access, which an averaged population hides. The baseline identifies that heterogeneity and anchors the projections to it, so that the simulation does not assume the intervention reaches everywhere equally. One point of scope is worth making: the simulation estimates impact under explicit assumptions on a realistic base; it does not reproduce the future with certainty. Its credibility depends on how reasonable those assumptions are, and they are stated explicitly. When the available data are not enough to build the baseline, a synthetic population can be used as an input, but when real data exist, the engine runs on them. What you receive is a simulation anchored to the best available characterisation of the population, with its assumptions in plain view.

By modelling diffusion with the structure that matches how the intervention spreads, instead of assuming a flat spread. This applies whenever the reach of the intervention is not uniform, which is the norm: it depends on how the intervention is distributed across the area or through the contact network. When the intervention involves a good that is physically distributed, its redistribution is modelled with realistic geographic friction, for example with gravity models over real travel times, so that distance and access can be built in as factors in the expected reach. When the intervention spreads through a social or contact network, reach is modelled on that structure, not on an even spread. The diffusion module is chosen according to the nature of the intervention and the information available to parameterise it; none is applied by default. What you receive is reach modelled in a way that is consistent with the intervention's diffusion mechanism, such as geographic friction or network topology, not an idealised assumption of uniform coverage.

Yes: the engine combines variants, intensities and coverage levels into a grid of scenarios and estimates, for each one, its effect on the indicators that matter to the decision. This applies when there is more than one way to configure the intervention and more than one dimension at stake, which is common: a policy is rarely judged on a single metric. Outputs are built to be comparable across every configuration in the grid, not as isolated results, and the scenarios are precomputed so they can be explored quickly rather than calculated on the fly. Indicators of different kinds, economic, health or diffusion, are estimated on the same simulation. They often diverge, and estimating them together keeps the decision from optimising one dimension while ignoring another. What you receive is every configuration on the table, each with its estimated impact through each lens, so you can judge on solid grounds which configuration best fits the objectives and constraints of the decision.

At the scale at which the decision is made: the result is disaggregated to the geographic or subgroup resolution at which you decide, not only the national aggregate. This applies when a policy is rolled out or adjusted at regional, provincial or population-group level, and the national average hides differences that change the decision. Indicators are estimated at national scale and at the finest resolution the design allows, with metrics weighted by the sampling design where applicable and with aggregations consistent with each level of decision, so that the estimate for each area or subgroup is not a naive aggregation of the average. An intervention can perform well in aggregate and poorly in specific areas, and that heterogeneity is exactly what a roll-out decision needs to see. What you receive is the estimated effect at the scale at which you decide, not a single national figure that averages out different realities.

Scenario simulation is a decision engine that estimates and compares the impact of different configurations of a policy; synthetic population generation is a foundational technique that builds a statistically realistic artificial population when the available data are not enough. They are not the same and do not answer the same question: a synthetic population addresses a shortage of data; scenario simulation works out which configuration of the intervention is worth rolling out. They connect naturally. When a population base is missing, a synthetic population can be one of the inputs the engine runs on, but the engine also works on real data when they exist, and a synthetic population has uses of its own outside simulation. If the need is a population base you don't have, it is synthetic population generation; if it is comparing configurations of a policy, it is this service. One provides the base; the other, the decision across scenarios.

Not exactly. Ex ante modelling estimates the potential effect of a specific intervention before it is introduced; scenario simulation is the engine that works through a space of configurations (intensities, coverage levels, sequences or combinations of measures) and compares their impact. The first asks, ‘What change might we expect from this intervention?’; the second, ‘Which configuration offers the best balance across indicators, and for whom?’ They can sit within the same project: ex ante modelling can draw on the simulation engine when the decision requires comparing many alternatives under explicit assumptions. If you only need the expected effect of one option, ex ante modelling is enough; if you need to work through and compare a broad space of configurations, scenario simulation is the better fit. We simulate, estimate and compare; the client decides which policy to implement.