Sensitivity Analysis quantifies the impact of variable changes on a specific outcome within a model. Employed across various disciplines, it aids in risk assessment, model validation, and decision-making, offering metrics to represent sensitivity.
Definition
Quantitative technique to assess how different values of an independent variable impact a particular dependent variable under a given set of assumptions.
Types of Variables
- Input Variables: Factors being manipulated in the analysis.
- Output Variables: Resultant variables that change based on input variations.
Types of Sensitivity Analysis
- Local Sensitivity Analysis: Examines change over a small range of parameter values.
- Global Sensitivity Analysis: Covers a wide range of parameter values.
- Deterministic: Uses specific set values for inputs.
- Probabilistic: Incorporates randomness in inputs and/or outputs.
Methods
- One-at-a-Time (OAT): Changes one variable while keeping others constant.
- Monte Carlo Simulation: Uses random sampling to obtain numerical results.
- Factorial Analysis: Investigates the effects of multiple variables at once.
Applications
- Risk Assessment: Used to estimate uncertainties in outcomes.
- Model Validation: Helps in refining models by comparing results to real-world data.
- Decision Support: Assists in choosing between different strategies or scenarios.
Metrics and Indicators
- Elasticity: Measure of sensitivity, often expressed as a percentage change.
- Tornado Diagrams: Graphical representation ranking variables by their impact.
- Sobol Indices: Quantify the contribution of each input to the output variance.
Limitations
- Computational Complexity: Especially relevant for high-dimensional models.
- Assumptions: Results as good as the assumptions they are based on.
Interdisciplinary Usage
- Finance: Option pricing, portfolio optimization.
- Engineering: System reliability, material selection.
- Medicine: Epidemiological models, treatment effectiveness.
Ethical Considerations
- Transparency: Clear methodology essential for validity.
- Misuse: Risk of cherry-picking data or manipulating variables for desired outcomes.
