Sensitivity Analysis

Sensitivity analysis answers a fundamental question in decision-making: "How much would things have to change for us to reach a different conclusion?" If a small change in one assumption flips the recommended alternative, the decision is fragile and that assumption deserves careful validation. If the same alternative wins across a wide range of assumptions, the decision is robust and can be made with confidence.

There are three primary approaches to sensitivity analysis. One-way (or univariate) sensitivity analysis varies a single input while holding all others constant, producing a clear picture of each factor's individual influence. Tornado diagrams — horizontal bar charts showing how the output changes as each input swings between its minimum and maximum values — are the most common visualization for one-way analysis. Multi-way (or scenario) sensitivity analysis varies multiple inputs simultaneously, testing specific combinations that represent realistic scenarios (best case, worst case, most likely). Monte Carlo simulation assigns probability distributions to uncertain inputs and runs thousands of random combinations, producing a probability distribution of outcomes rather than a single point estimate.

In the context of multi-criteria decision analysis, sensitivity analysis typically focuses on criterion weights. The key question is: "At what weight for criterion X does the top-ranked alternative change?" This critical weight analysis reveals how much stakeholders would need to disagree on priorities before the recommendation shifts. If the winning alternative stays on top even when weights shift by 20-30%, the recommendation is robust. If a 5% weight change causes rank reversal, stakeholders need to reach genuine consensus on that criterion's importance before proceeding.

Sensitivity analysis is also essential for cost-benefit analysis, where it tests key financial assumptions — discount rate, growth rate, implementation cost, benefit realization timeline. A project with a positive NPV under the base case but negative NPV under reasonable pessimistic assumptions is riskier than one with a positive NPV under all tested scenarios.

The output of sensitivity analysis is not a single answer but a map of decision confidence. It tells decision-makers where to invest in better data (the most influential uncertain inputs), where to build in flexibility (factors whose values may change over time), and how much margin of safety exists in the current recommendation.