Sensitivity Analysis with Event Tree Analysis

One-at-a-time (OAT) sensitivity varies each input individually while holding others fixed, measuring local influence at the base-case values. Global sensitivity analysis (e.g., Sobol indices, Morris screening) samples all inputs simultaneously across their full uncertainty ranges and measures how much of the total output variance is attributable to each input and to interactions between inputs. OAT is simpler and faster; global methods are more informative but computationally demanding.

Yes, and in fact this is one of the strongest use cases. When probabilities are expert-elicited they carry wide uncertainty ranges, and sensitivity analysis reveals which judgments most influence the conclusion. If a critical input is entirely expert-driven and highly sensitive, it flags a priority area for additional data collection or physical testing.

No — they are complementary. Uncertainty analysis propagates input distributions through the model to characterize the distribution of the output (e.g., a 5th–95th percentile range for end-state frequency). Sensitivity analysis explains how much each input contributes to that output uncertainty. Both are typically reported together in quantitative risk assessments.

Sobol-based methods typically require N(k+2) evaluations, where N is a base sample size (often 500–10,000) and k is the number of uncertain inputs. For a tree with 20 uncertain branches this can mean tens of thousands of evaluations. Morris screening requires far fewer (r(k+1) runs, commonly 50–200 total) and is a practical first screen before committing to full Sobol analysis.

Not necessarily. A purely qualitative or screening-level ETA aims to identify scenario logic, not quantify probabilities precisely; sensitivity analysis adds little in that context. Sensitivity analysis is most valuable when the ETA is quantified, probability inputs are uncertain, and decisions (regulatory compliance, capital allocation for safety improvements) depend on the risk estimates.