Skip to main content
v2026.11,986 entries · CC-BY 4.0

Scenario Modelling: Conditional Projections in Infectious-Disease Research

Scenario modelling shows what would happen under stated conditions, not what will happen, and how it differs from forecasting.

Written and maintained by CASRAI Editorial Board

Published

On this page

Scenario modelling is the use of a model to show what would happen under a stated set of conditions, rather than to predict what will happen. In infectious-disease research the output is called a scenario projection: a conditional prediction that contrasts several possible futures, such as different vaccination levels or different values of a transmission parameter, without claiming to know which future will occur.

Definition and scope

The clearest published statement of the distinction comes from Runge and colleagues (Epidemics, 2024), who set out a design framework for infectious-disease projections. They describe a forecast as an unconditional prediction of what is expected to happen, ideally expressed as a probability distribution. A scenario projection is a conditional prediction: it describes what would happen if a defined set of conditions were met. It typically gives the probability of outcomes given those conditions, but it does not usually quantify how likely the conditions themselves are.

That last point shapes how scenario results should be read. A scenario is exploratory. It contrasts multiple futures so that a decision maker or researcher can see how outcomes move when assumptions move. Isolating particular conditions of interest as separate scenarios is what distinguishes it from a forecast, where uncertainty about assumptions is folded into a single embedded prediction.

Scenario modelling is not unique to epidemiology. The same paper notes that climate research (the CMIP6 shared socio-economic pathways), conservation, fisheries, energy and transport all use scenario-based projection, and that the field is fragmented, with no common lexicon across disciplines. In climate work, projections are likewise conditional on scenario assumptions, and the likelihood of those conditions is not estimated. For the climate side of this family of methods, see the CASRAI guide to downscaling. Business and finance pages that use the same words usually mean base-case, best-case and worst-case planning; this page covers the research sense of the term in infectious-disease epidemiology.

Scenario projection versus forecast

The two products answer different questions and are judged differently. The comparison below restates the wording of Runge et al. (2024) and the Scenario Modeling Hub.

Feature Forecast Scenario projection
Question answered What is expected to happen What would happen if a set of conditions were met
Type of prediction Unconditional, ideally probabilistic Conditional
Treatment of assumptions Uncertainty in assumptions is embedded in the forecast Conditions of specific interest are isolated as separate scenarios
Probability of the conditions Part of the prediction Not usually quantified
Character Predictive Exploratory, contrasting multiple futures

The Scenario Modeling Hub makes the practical consequence explicit on its site: forecasts are unconditional estimates of what will happen, and even strong models struggle beyond about three to four weeks. Longer-term questions are therefore handled with scenarios, which compare outbreak trajectories under different assumptions.

How scenario modelling is used in infectious-disease epidemiology

Runge et al. cite a long record of epidemic scenario work, including UK foot-and-mouth disease (Tildesley 2006), avian influenza (Longini 2005), Ebola in West Africa (Meltzer 2014), COVID-19 and the roll-out of new vaccines. The common pattern is that a team agrees a small set of scenarios with the people who will use the results, runs one or more models under each, and compares the outcomes. The model structure behind this work is covered in the guide to epidemiology models, and the trajectories that come out of them are the kind of curve described in epidemic curve. For the discipline as a whole, start with what is epidemiology.

The U.S. COVID-19 Scenario Modeling Hub as a worked case

Runge et al. use the first 17 rounds of the U.S. COVID-19 Scenario Modeling Hub (February 2021 to April 2023) as a case study of design practice. Their account gives several concrete design facts:

  • Most rounds used four scenarios, and every round except rounds 1 and 17 was a 2 x 2 design.
  • Designing a round took between 3 and 86 days, with a median of 32.5 days.
  • The motivating audience was mostly federal, state and local public health agencies.
  • Round 6 (29 May 2021) crossed the Delta variant with vaccination and hesitancy, a situational-awareness design. Round 8, on waning immunity, was a sensitivity analysis. Round 14 (4 June 2022) crossed age targets for boosters with variant, a value-of-information design.

The hub’s website now carries pathogen-specific projections for COVID-19, seasonal influenza and RSV, plus research rounds. Its COVID-19 scenarios examine behaviour, control measures, variants and vaccination; its influenza scenarios examine vaccination coverage and effectiveness, prior immunity and dominant subtypes.

Six purposes a scenario set can serve

A central contribution of the Runge et al. framework is that a scenario set should be designed for a stated purpose, and that the purpose determines what varies across scenarios. They describe six purposes, each pairing the two axes of a scenario grid as either decision axes (things the decision maker controls, such as intervention choices) or uncertainty axes (things nobody controls, such as a transmission parameter). The data table below sets out the six purposes with the axis types and examples given in the paper.

The sensitivity-analysis example in the paper shows how a grid is built. Cross two values of the basic reproduction number (R0 of 2.2 and 3.0) with two values of the serial interval (3 and 4.5 days) and you obtain four scenarios, each an uncertainty-by-uncertainty cell. Comparing the cells shows how much the projected outcome depends on each parameter. A decision-making design instead crosses intervention options with each other, and a value-of-information design crosses a decision axis with an uncertainty axis. The paper’s example is a 3 x 2 grid of three spatial intervention levels by two R0 values, which tests whether the better intervention changes when the uncertain parameter changes.

Where it matters for research offices and labs

The sourced lessons that carry over to research planning are narrow but practical. First, state the purpose before choosing the scenarios: the same model can serve decision making or sensitivity analysis, and the grid that suits one wastes effort on the other. Second, keep decision axes and uncertainty axes distinct, because a reader who cannot tell which axis is a choice and which is a guess will misread the output. Third, expect design to take time. The hub’s median of 32.5 days for scenario design, with a range from 3 to 86 days, shows that agreeing scenarios with stakeholders is a substantial part of the work and not a preliminary formality.

Evaluation also differs from forecasting. Bay and colleagues (Epidemics, 2024) note that scenario projections rely on specific assumptions about plausible states of the world that may or may not come to pass, and that long-term projections therefore need different evaluation criteria from short-term forecasts. Their Ensemble2 approach for the U.S. COVID-19 Scenario Modeling Hub assesses the performance of a scenario ensemble without having to identify which scenario was the most plausible, and they report the Ensemble2 models as well calibrated. For a related quantitative method that projects a single population path, see the guide to population projections.

Common misunderstandings

A scenario is not a forecast

A scenario projection says what follows from its conditions. It does not say those conditions will occur, and a scenario that did not match later events is not automatically a failed prediction, because the conditions may simply not have been realised.

Scenarios do not carry a probability of being true

Scenario projections express the probability of outcomes given conditions; the probability of the conditions is typically left unquantified. Reading the highest-impact scenario as the most likely one, or the middle one as a best guess, adds a claim the design never made.

More scenarios are not better

The hub’s most common design was four scenarios in a 2 x 2 grid. Each added axis multiplies the number of cells, so a design with several axes quickly becomes hard to run and hard to read. Choosing which two conditions matter most, for the stated purpose, is the design task.

One field’s definition does not transfer unchanged

Because there is no shared lexicon across fields, “scenario” in finance, climate science and epidemiology overlaps without being identical. State which sense you are using when you cite or compare results.

Sources

Data: Six purposes of infectious-disease scenario projections and their design axes

Compiled from the scenario-design framework of Runge et al. (Epidemics 2024) and the Scenario Modeling Hub's published rounds: for each of six purposes, the axis types crossed in the scenario grid and the example design the paper gives.

Period covered: 2021/2023.

Six purposes of infectious-disease scenario projections and their design axes
PurposeAxis types crossedWhat the design doesExample in source
Decision makingDecision x decisionContrasts intervention optionsContrast of interventions (framework description)
Sensitivity analysisUncertainty x uncertaintyShows how outcomes depend on uncertain parametersR0 of 2.2 vs 3.0 crossed with serial interval of 3 vs 4.5 days; Hub round 8 (waning immunity)
Situational awarenessUncertainty axesShows what might be comingHub round 6 (29 May 2021): Delta variant crossed with vaccination and hesitancy
Horizon scanning (scenario planning)Plausible extremesExplores plausible extremes to motivate preparationFramework description
ForecastingUncertainty embedded, not isolatedGives an unconditional prediction with assumption uncertainty built inFramework description
Value of informationDecision x uncertaintyTests whether the better intervention depends on the uncertainty3 x 2 design: three spatial intervention levels by two R0 values; Hub round 14 (4 June 2022): booster age targets crossed with variant

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · full access

Ask about Scenario Modelling: Conditional Projections in Infectious-Disease Research

Full access: large monthly usage, up to 10 source passages behind each answer, and MCP and API access for Claude Code, Codex and other CLIs. Cancel any time.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →