Sensitivity Analysis (Tornado Chart)

Sensitivity analysis in a QRA context identifies the inputs that most influence the variability of the output. For a cost risk model, it answers: which risk items and which cost estimate uncertainties are driving the spread of the total cost distribution? The results are displayed as a tornado chart — a horizontal bar chart where inputs are ranked from top to bottom by their influence on the output, with the longest bar at the top. The name comes from the characteristic shape of the chart, which widens at the top like an inverted tornado.

Sensitivity analysis is arguably the most actionable output of a QRA. The S-curve tells you the probability of different outcomes; the tornado chart tells you why those outcomes vary. The top five or ten items in a tornado chart define where the project team should focus its risk management effort. If ground conditions and procurement lead time together account for 60% of the variance in the project cost distribution, those are the two risk areas that need active management, detailed monitoring, and the most sophisticated mitigation strategies. Everything else on the risk register is secondary to getting those two right.

There are two main types of sensitivity analysis in Monte Carlo models. Regression-based sensitivity measures the statistical correlation between each input variable and the overall output across all iterations — the higher the correlation, the more influential the input. Contribution to variance decomposes the total variance of the output into the percentage contributed by each input. Both approaches give broadly similar rankings for most models, though they can diverge when inputs are correlated. Some tools also produce criticality index charts for schedule models, showing the proportion of iterations in which each activity appears on the critical path — which is the schedule equivalent of a cost sensitivity analysis. All of these outputs should be reviewed alongside the S-curve, not as a standalone analysis.