One-way sensitivity analysis: method, table, graph and example

Learn how one-way sensitivity analysis works: vary one input at a time, read the table, draw the tornado plot and apply an example to your own shortlist.

3 min read3 viewsOne-way sensitivity analysis: method, table, graph and example
— · 0 votes
Contents
  1. How the method works step by step
  2. Reading the sensitivity analysis table
  3. Building the tornado plot
  4. Questions about one-way sensitivity analysis
  5. What is one-way sensitivity analysis in simple terms?
  6. Can you give a one-way sensitivity analysis example?
  7. What does a one-way sensitivity analysis graph show?
  8. How is OWSA different from varying several inputs at once?

One-way sensitivity analysis tests a model by changing a single input parameter while every other input stays at its base case value. The method shows how much the model output moves when one assumption shifts, which helps you see where your comparison is fragile. This page explains the method, the table and graph that present it, and a worked example you can copy.

How the method works step by step

One-way sensitivity analysis, often shortened to OWSA, starts from a base case that fixes every input at one chosen value. You then vary one input parameter across a range you can defend, and record the model output at each step. The other inputs stay fixed, so any change in output is tied to that single input. Repeating this for each input gives a set of output ranges you can compare.

Reading the sensitivity analysis table

The sensitivity analysis table lists each input, its low and high values, and the output produced at each end. You compare the size of each output range to judge which inputs move the result and which barely matter. A narrow range suggests the conclusion holds even if that input is uncertain. A wide range marks an input that deserves closer scrutiny before you commit.

Building the tornado plot

A tornado plot turns the table into a graph by drawing one horizontal bar per input, sorted so the widest sits at the top. Each bar spans the model output range produced when that input moves between its low and high values. The stack of bars narrows downward, which gives the chart its name. The tornado plot shows at a glance which assumptions carry the comparison and which are background noise.

Questions about one-way sensitivity analysis

What is one-way sensitivity analysis in simple terms?

One-way sensitivity analysis is a test that changes one input at a time to see how much the result moves. It is also called deterministic sensitivity analysis, because each run gives a single fixed output rather than a spread. The base case holds every other input steady, so the effect you see belongs to that one parameter. This makes it easy to explain which assumption shapes your decision.

Can you give a one-way sensitivity analysis example?

A one-way sensitivity analysis example: suppose you compare two options using a running cost that depends on an energy price. You hold every other input at the base case and rerun the model with the energy price low, then high, and note the output each time. The gap between the two outputs is the output range for that input. Doing this for each input, one by one, gives the full table behind your shortlist.

What does a one-way sensitivity analysis graph show?

A one-way sensitivity analysis graph shows the output range each single input produces when it moves between its low and high values. The tornado plot is the form used for this, with bars sorted by width so the strongest drivers sit at the top. The graph does not show interactions between inputs, because only one changes per run. Read it as a ranking of which assumptions you should pin down first.

How is OWSA different from varying several inputs at once?

OWSA varies only one input at a time, so it cannot capture cases where two inputs move together or interact. Varying several inputs at once, such as a probabilistic approach, produces a distribution of outputs instead of single values. Use OWSA to identify and rank the sensitive inputs before you invest in that wider model. The two methods answer different questions and work well in sequence.

Read next

Latest articles