A model produces an answer.
Change one assumption slightly.
The answer barely moves.
Change another.
Everything changes.
Those two assumptions are not equally important.
A student’s weekly study plan depends on:
available time,
travel,
sleep,
practice speed,
homework load.
Which one actually determines whether the plan works?
A financial projection depends on:
growth,
cost,
price,
interest rate.
Which assumption controls the conclusion?
A Science model depends on:
temperature,
concentration,
measurement error,
initial condition.
Which parameter matters most?
A policy argument depends on:
cost,
uptake,
effect size,
population.
Which assumption would reverse the decision?
That is sensitivity analysis.
A useful Wintour House definition is:
Sensitivity analysis is the disciplined study of how changes in inputs, parameters, assumptions or modelling choices influence an output, conclusion or decision—so that attention can be directed toward the factors that actually matter rather than distributed equally across everything that could vary.
The phrase actually matter is crucial.
People often spend time refining inputs that have almost no effect on the answer.
Meanwhile, one uncertain assumption quietly controls the whole result.
Sensitivity analysis exposes that asymmetry.
The newest practical teaching literature makes this explicit. A 2026 practical guide to teaching sensitivity analysis defines sensitivity analysis as attributing variability in model outputs to uncertainty in input parameters and assumptions and argues that it should be treated as part of model-based decision making rather than as an optional technical afterthought. The 2026 review of fifty years of decision analysis likewise identifies sensitivity analysis as a major tool for understanding uncertainty, model structure and information value.
The Wintour House question is therefore:
If a learner became excellent at ten sensitivity-analysis operations, which ten would still matter when models, spreadsheets, simulations and AI tools changed?
Before the Top 10: Uncertain Is Not the Same as Influential
Suppose a project model contains two uncertain inputs.
Input A could vary by 50%.
Input B could vary by only 5%.
Which matters more?
Not enough information.
Perhaps the output barely responds to A.
Perhaps a tiny movement in B causes the decision to flip.
Sensitivity is not the same as uncertainty.
Uncertainty asks:
How little do we know about this input?
Sensitivity asks:
How strongly does the output respond if the input changes?
A variable can be highly uncertain but low influence.
Another can be precisely known but extremely influential.
This distinction is one of the most important in model-based reasoning.
It prevents learners from assuming that the widest error bar deserves the most attention.
The right question is:
Which uncertainty matters to the outcome?
That is where sensitivity analysis begins.
1. Learn to Define the Output Before Testing Inputs
Sensitive with respect to what?
A study plan may be sensitive in:
completion rate,
sleep,
stress,
retention.
A financial model may be sensitive in:
profit,
cash flow,
break-even date.
A policy may be sensitive in:
cost,
benefit,
equity,
uptake.
One input can strongly affect one output and barely affect another.
So the first skill is:
name the output.
Use:
I am testing how changes in ______ affect ______.
Without that sentence, sensitivity analysis becomes vague.
Students often ask:
“Does this variable matter?”
Matter for what?
A variable may matter for speed but not accuracy.
For average outcome but not worst-case outcome.
For one stakeholder but not another.
Sensitivity is relational.
Input to output.
Not importance in the abstract.
Worth learning because: sensitivity only has meaning relative to a specified outcome, so the learner must define which result or decision is being tested before changing inputs.
2. Learn to List the Inputs, Parameters and Assumptions Separately
Models contain different kinds of things.
Observed inputs.
Estimated parameters.
Fixed constants.
Chosen assumptions.
Scenario conditions.
These should not be merged.
Suppose:
travel time is observed.
practice speed is estimated.
sleep minimum is a chosen constraint.
future workload is assumed.
Each has a different status.
A sensitivity analysis should make the list visible.
Then ask:
Which are uncertain?
Which are controllable?
Which are fixed by definition?
Which are modelling choices rather than facts?
This connects to Constraint Reasoning and Model-Based Reasoning.
Top 10 Constraint Reasoning Skills Worth Learning owns admissibility conditions.
Top 10 Model-Based Reasoning Skills Worth Learning owns assumptions and model structure.
Sensitivity Analysis asks:
What happens to the output when each of these model ingredients moves?
Worth learning because: separating inputs from assumptions and constraints makes it possible to identify which source of variation is genuinely driving the result.
3. Learn to Establish a Baseline Case
Before changing an input, record the starting result.
Baseline:
study plan completion = 85%.
Cost = $1,000.
Predicted score = 72.
Model error = 4%.
Now change one factor.
Without a baseline, movement becomes difficult to interpret.
Sensitivity is about difference:
new output minus baseline.
Or:
percentage change relative to baseline.
The baseline should be plausible, not necessarily “most likely.”
It is simply the reference case.
A good baseline also records:
input values,
assumptions,
units,
output.
This prevents a common spreadsheet error:
changing cells without knowing what the original case actually was.
Worth learning because: a stable baseline makes the effect of each perturbation measurable and prevents sensitivity analysis from becoming a sequence of disconnected alternative cases.
4. Learn to Vary One Input at a Time for Local Understanding
One-way sensitivity analysis is simple and powerful.
Hold other inputs fixed.
Change one.
Observe output.
Example:
Study plan baseline:
90 minutes available.
Now:
75.
60.
45.
What happens to completion?
Or a model:
interest rate 3%.
Try:
2%.
4%.
5%.
This does not tell the whole story.
Inputs may interact.
But one-at-a-time variation is excellent for first diagnosis.
The 2026 decision-analysis review describes one-way sensitivity analysis as one of the longstanding tools in the field, while newer global methods extend analysis beyond one-factor changes.
Students should understand the limitation.
“One at a time” answers:
What is the local effect when everything else is held at baseline?
It does not automatically answer:
what happens when several inputs move together?
Worth learning because: one-way sensitivity gives a clear first view of direction and magnitude before interactions and combined uncertainty make interpretation harder.
5. Learn to Use Plausible Ranges Rather Than Arbitrary Changes
Change the input by 10%.
Why 10?
Sometimes sensible.
Sometimes meaningless.
Sensitivity tests should use plausible ranges.
Historical data.
Measurement uncertainty.
Expert estimate.
Policy bounds.
Physical limits.
Observed variation.
Suppose daily travel varies from 35 to 60 minutes.
Test that.
Not 0 to 500 minutes just because software allows it.
The quality of a sensitivity analysis depends partly on the input range.
Too narrow:
important vulnerability hidden.
Too wide:
unrealistic instability exaggerated.
The practical guide to teaching sensitivity analysis emphasises that sensitivity methods depend on explicit treatment of input uncertainty and model assumptions.
For learners, the rule is:
The perturbation range needs a reason.
Worth learning because: sensitivity conclusions are only as meaningful as the ranges tested, and arbitrary variation can make a stable model look fragile or a fragile model look safe.
6. Learn to Read Direction, Shape and Thresholds—not Only “More” or “Less”
An input increases.
Output increases.
Linear?
Perhaps.
Maybe:
initial increase matters a lot,
then effect saturates.
Maybe:
nothing happens until a threshold.
Then rapid change.
Maybe:
output rises and later falls.
Sensitivity is not only strength.
It also has shape.
Ask:
Is the relationship:
positive?
negative?
linear?
curved?
thresholded?
saturating?
reversing?
This matters because one local test can mislead.
At baseline:
small effect.
Near threshold:
huge effect.
The learner should therefore plot when possible.
Input on x-axis.
Output on y-axis.
Shape becomes visible.
This connects to Scale Reasoning.
Nonlinear relationships can hide behind ordinary percentage changes.
Worth learning because: sensitivity depends on where the system is operating, and the same input can have weak, strong or reversed effects across different ranges.
7. Learn to Test Interactions Between Inputs
Sometimes A matters only when B is high.
Study time matters strongly when practice quality is high.
Otherwise extra time produces little.
Temperature matters differently at different concentrations.
Price sensitivity changes with demand.
Inputs interact.
A one-way analysis can miss this.
So test pairs.
Low A + low B.
Low A + high B.
High A + low B.
High A + high B.
Now interaction appears.
Global sensitivity methods go further by varying multiple inputs across their distributions and estimating how much each factor and combination contributes to output variability. The 2026 practical guide explicitly distinguishes local and global sensitivity ideas and emphasises their different purposes.
For school learners, the mathematics can remain simple.
The habit is:
Could the effect of this input depend on another input?
That question is enormously powerful.
Worth learning because: real systems often contain interactions, so an input that looks unimportant alone may become decisive in combination with another condition.
8. Learn to Rank Inputs by Decision Relevance, Not Mere Numerical Movement
Suppose:
Input A changes output by 20%.
Input B changes output by 2%.
A seems more important.
But the decision threshold may be close to the B-driven change.
Example:
Pass threshold:
50.
Baseline:
51.
Input B moves result from 51 to 49.
Input A moves a secondary output that does not affect the decision.
Which matters?
B.
Sensitivity for decision-making is not just output variance.
It asks:
Which input can change what we decide?
This is where sensitivity analysis becomes value-of-information reasoning.
If one uncertain input can flip the decision, learning more about it may be valuable.
If another input cannot change the decision across its plausible range, further precision may not matter.
The 2026 decision-analysis review links sensitivity analysis directly with information acquisition and value of information.
Worth learning because: the most decision-relevant input is the one capable of changing the action, not necessarily the one producing the largest numerical fluctuation somewhere in the model.
9. Learn to Find Breakpoints and Decision Reversal Points
At what value does the decision flip?
This is a particularly useful form of sensitivity analysis.
Suppose option A is preferred if adoption exceeds 40%.
Below 40%:
option B.
Then 40% is a breakpoint.
Now ask:
How plausible is 40%?
What evidence do we have?
A student chooses between two study strategies.
One is better only if available time exceeds 90 minutes.
What happens on ordinary school days?
Breakpoints turn vague statements into conditional ones.
Instead of:
“This is the best plan.”
Say:
“This is the best plan if X remains above Y.”
That is much more transparent.
Threshold sensitivity also helps arguments.
A policy benefit may depend on uptake.
A scientific conclusion may depend on a parameter.
A financial decision may depend on price.
Worth learning because: breakpoints reveal the exact conditions under which a conclusion changes and expose which assumptions deserve the most scrutiny.
10. Learn to Communicate Sensitivity Clearly and Use It to Revise the Model or Decision
Analysis is not finished when the spreadsheet moves.
Explain.
Which inputs matter most?
Which do not?
What ranges were tested?
Where are the thresholds?
Which interactions matter?
Did the preferred decision change?
Visual tools may include:
tornado plots,
spider plots,
scenario tables,
response curves.
But clarity matters more than sophistication.
A strong conclusion sounds like:
The result is insensitive to A across the tested range, moderately sensitive to B, and highly sensitive to C near the threshold. The decision changes only when C falls below 0.4.
That is useful.
Then act.
Collect better data on C.
Redesign the model.
Choose a more robust option.
Narrow the claim.
Sensitivity analysis should change attention.
Not merely decorate a report.
Worth learning because: the purpose of sensitivity analysis is to redirect modelling, evidence collection or decision effort toward the inputs that actually control the result.
The Top 10 Sensitivity Analysis Skills as One System
The Wintour House route is:
DEFINE OUTPUT → LIST INPUTS/ASSUMPTIONS → BASELINE → ONE-WAY PERTURB → PLAUSIBLE RANGE → SHAPE/THRESHOLD → INTERACTIONS → DECISION RELEVANCE → BREAKPOINT → COMMUNICATE & REVISE
The quieter version is:
Know which result you care about. List what can change. Freeze a baseline. Move one input deliberately. Use realistic ranges. Watch direction and shape. Test combinations. Rank inputs by whether they can change the decision. Find the flip point. Then use the result to improve the model or collect the right information.
That is sensitivity analysis.
Not guessing.
Not random parameter fiddling.
Not robustness.
Not uncertainty alone.
Sensitivity analysis is influence made visible.
Sensitivity Analysis Is Not the Same as Robustness
Robustness asks:
Does performance remain acceptable under variation?
Sensitivity Analysis asks:
Which variation changes the output most, and how?
Top 10 Robustness Skills Worth Learning owns survival.
Sensitivity Analysis owns influence.
Sensitivity Analysis Is Not the Same as Uncertainty
Top 10 Uncertainty Skills Worth Learning owns reasoning when knowledge is incomplete.
Sensitivity Analysis asks a narrower model question:
If the uncertain quantity changes, what happens to the result?
Uncertainty can be large and sensitivity small.
Or the reverse.
Sensitivity Analysis Is Not the Same as Statistical Reasoning
Statistical Reasoning estimates patterns and uncertainty from data.
Sensitivity Analysis perturbs model inputs and assumptions.
Statistical estimates often become inputs to sensitivity analysis.
Different jobs.
Sensitivity Analysis Is Not the Same as Systems Thinking
Systems Thinking models feedback and dynamic wholes.
Sensitivity Analysis can be applied to a system model.
But it can also be applied to a simple spreadsheet or equation.
Sensitivity Analysis Is Not the Same as Scale Reasoning
Scale Reasoning interprets magnitude.
Sensitivity asks how much output changes when an input changes.
Scale can affect sensitivity.
But scale itself is not the owner.
For Primary Students
Primary learners do not need formal sensitivity analysis terminology.
They can ask:
What happens if I change only this?
Make the ramp steeper.
Does the car travel farther?
Add one block.
Does the tower still stand?
Change one ingredient.
What changes?
The habit is controlled comparison.
For Secondary Students
Secondary students can begin parameter thinking.
Formula.
Model.
Experiment.
Graph.
Change one quantity.
Predict first.
Then calculate or test.
Ask:
Which variable changes the result most?
Where does the relation stop being linear?
This is especially powerful in Mathematics and Science.
For JC Students
JC learners can use sensitivity reasoning explicitly.
Economics:
elasticity and assumptions.
Physics:
parameter dependence.
Chemistry:
rate sensitivity.
Mathematics:
functions and derivatives.
GP:
which factual assumption controls the conclusion?
A strong JC sentence is:
The conclusion is robust to A but highly sensitive to B.
That is compact intellectual control.
Sensitivity Analysis in Mathematics
Functions are sensitivity machines.
Change x.
Observe y.
Derivative:
local rate of change.
But sensitivity analysis is broader than calculus.
It asks which inputs matter and over what range.
A model may contain several variables with very different influence.
Sensitivity Analysis in Science
Science uses sensitivity in:
models,
experiments,
instrumentation,
parameter fitting,
uncertainty analysis.
Students should learn that not all measured variables deserve equal precision.
If an output barely responds to one parameter, extreme measurement effort there may be wasted.
Sensitivity Analysis in English and GP
Arguments have inputs too.
Suppose a conclusion depends on:
cost estimate,
participation rate,
effect size.
Change each.
Does the argument survive?
This turns “assumptions” into testable pressure points.
Sensitivity Analysis in Studying
A study plan is a model.
What controls success?
Hours?
Sleep?
Question difficulty?
Practice quality?
Travel?
Change one.
Observe.
Students often think “more study time” is the dominant variable.
Sensitivity analysis may reveal that practice quality matters more.
Sensitivity Analysis in the Age of AI
AI can run sensitivity analysis instantly.
That makes question quality more important.
Ask:
“Which assumptions should I vary?”
“Use plausible ranges and explain why.”
“Show one-way and interaction effects.”
“Find the breakpoint where the decision changes.”
“Which input is worth collecting better data for?”
AI can calculate.
The human must decide whether the tested inputs and ranges are meaningful.
The Sensitivity Paradox: The Most Uncertain Input May Not Matter
Huge uncertainty.
Tiny output effect.
Do not waste precision there.
The Sensitivity Paradox: A Tiny Input Change Can Control the Decision
If the baseline sits near a threshold, even a small variation can flip the conclusion.
The Sensitivity Paradox: A Stable Average Can Hide an Unstable Decision
Outputs may move only slightly while crossing a critical decision boundary.
That is why decision sensitivity matters.
The Wintour House Test: Does Sensitivity Analysis Survive When AI Can Model Everything?
Yes.
Because someone still has to decide:
which output matters,
which inputs are uncertain,
what range is plausible,
which interactions matter,
what breakpoint changes action,
and which uncertainty deserves better evidence.
That is why Sensitivity Analysis belongs permanently in the Skills Worth Learning series.
The mature learner can eventually say:
I know the output I care about. I have separated inputs from assumptions, established a baseline, varied plausible ranges, identified direction, thresholds and interactions, ranked inputs by decision relevance, found the breakpoints and used the analysis to improve the model or focus evidence collection.
That is sensitivity analysis becoming influence intelligence.
Research Anchors
The ten skills above are a Wintour House editorial synthesis, not a claim that education research has validated one universal ten-part sensitivity-analysis taxonomy.
The 2026 practical guide to teaching sensitivity analysis defines sensitivity analysis as attributing output variability to uncertainties in input parameters and assumptions and draws on the authors’ experience across 12 summer schools and specialist training programmes. It emphasises local and global methods, model-based decision making and risk analysis.
The 2026 review of fifty years of decision analysis identifies sensitivity analysis as a major component of decision analysis and reviews developments from one-way sensitivity to broader uncertainty and information-value methods.
Robust decision-making literature complements this by showing how sensitivity and scenario exploration help reveal conditions under which strategies fail or decisions change.
The strongest defensible Wintour House conclusion is therefore:
Sensitivity analysis is disciplined influence testing: define the output, identify inputs and assumptions, establish a baseline, perturb plausible ranges, inspect direction and nonlinear shape, test interactions, rank variables by decision relevance, identify breakpoints and use the result to redirect modelling, evidence collection or action toward what actually controls the outcome.
