Bayesian Scenario Analysis — Probabilistic weighting of future scenarios via Bayesian inference

Use structured expert elicitation protocols (e.g., the Sheffield method or SHELF) to obtain calibrated priors. Conduct sensitivity analysis over a range of plausible priors to test whether conclusions are robust to prior choice. When priors are truly uninformative, equal weighting may be defensible but should be stated explicitly as an assumption.

Monte Carlo simulation samples from continuous probability distributions to propagate uncertainty through a model, treating the model structure as fixed. BSA explicitly defines discrete alternative futures (scenarios) and uses Bayes' theorem to update the probability weight of each scenario based on evidence. Both can be combined: Monte Carlo can be run within each scenario to capture within-scenario uncertainty.

Typically 3–6. Too few scenarios risk missing important futures; too many make prior elicitation and likelihood assessment cognitively unmanageable. Scenarios must remain mutually exclusive — if you cannot clearly assign evidence to one vs. another, consider merging or restructuring.

Yes. BSA assigns probabilities to scenarios and optimizes expected utility; RDM avoids probability assignments and instead seeks strategies that perform acceptably across all scenarios. In practice, analysts sometimes use BSA for scenarios where evidence is available and RDM for deeply uncertain futures — a complementary rather than competing approach.

BSA can be implemented in R (using packages such as rjags, brms, or custom scripts), Python (PyMC, Stan via cmdstanpy), or spreadsheet tools for simple discrete cases. Specialized platforms for health economic modeling (TreeAge, Analytica) also support probabilistic scenario weighting natively.