Simulation studies
Experiment inside an explicitly validated model
Simulation studies execute a model to examine system behaviour under chosen conditions. Discrete-event simulation advances through events; continuous simulation evolves state over time. Model construction, verification, calibration, validation and experimental analysis are separate tasks. Repeated stochastic runs estimate simulation variability, while uncertainty about real-world structure or parameters requires additional sensitivity analysis.
Choose simulation when direct experiments are impractical or costly and credible mechanisms, inputs or calibration data exist, especially to investigate queueing, computing systems or plausible intervention scenarios.
Strengths
- Allows controlled exploration of difficult or unobserved scenarios
- Separates mechanisms and parameters for sensitivity analysis
Limitations
- A precise result can be based on an implausible model
- Calibration and validation on the same data can overstate credibility
Know the boundary
Simulated interventions identify effects inside the model, not automatically causal effects in the world.