Concept
A tornado chart is the standard display of a one-way sensitivity analysis — varying one input across its credible range with all others held fixed: one horizontal bar per input, each spanning the output’s low-to-high swing, sorted longest at the top down to shortest. The tapering, funnel-like shape gives it its name, and its vertical order makes the model’s drivers visible at a glance.
How it’s built. Run a one-way sensitivity on each input — low and high across its credible range, all others at base — and draw a horizontal bar from the low-case to the high-case output, against a vertical reference line at the base case. Stack the bars longest at the top; they shrink down the chart, producing the tornado silhouette.
What it encodes. Three things read directly off it: vertical order = the driver ranking (top bar is the dominant driver); bar length = leverage (how far the output swings when that input moves across its range); position relative to the base line = direction and asymmetry (which end is the high input value, and whether the bar reaches further on one side — a signature of a nonlinear response). The usual pattern — a few long bars over a long tail — is the visual form of uncertainty concentrating in a handful of parameters.
What it is not. A tornado is a presentation of one-way results, inheriting every limitation of the one-at-a-time analysis behind it. It’s not a probability distribution, and its bars are not additive: each is measured with all other inputs fixed, so their lengths can’t be summed or enveloped into a combined uncertainty. Moving several inputs together is a scenario (a coherent bundle of inputs moved at once), which a tornado doesn’t show.