Science Education Systems · Article 56. Maya, Jia Jun, Hana and Ethan remain fictional Punggol learners. This article follows the sensitivity layer: how Science tests which assumptions and parameters actually control a conclusion.
The 50-second parent route
A scientific result can look precise while depending heavily on one uncertain assumption.
Sensitivity analysis asks what happens when we change that assumption deliberately.
The route is:
model → parameter or assumption → plausible range → controlled variation → output change → ranking → threshold check → robustness judgement → experiment priority → revision
The key question is:
If this input changes a little, how much does the conclusion change?
This article completes Articles 53–56 after How Scientific Variables Work, How Scientific Rates Work and How Scientific Thresholds Work.
1. Sensitivity means responsiveness
Change an input.
Observe the output.
If the output changes dramatically, the model is sensitive to that input.
If little changes, the conclusion may be relatively insensitive within that range.
2. Sensitivity analysis is controlled perturbation
We deliberately vary one parameter, assumption or boundary condition and observe the consequences.
It is an experiment performed on the model.
3. Maya’s sensitivity error is one-number dependence
She uses one estimated parameter and never asks how uncertain it is.
Her repair:
vary the parameter across a plausible range.
4. Jia Jun’s sensitivity error is changing everything at once
He changes three parameters and the output moves.
He cannot tell which input mattered most.
His repair:
begin with one-at-a-time tests, then study interactions deliberately.
5. Hana’s sensitivity error is assuming small uncertainty means small consequence
A parameter is uncertain by only 2%.
But near a threshold, that 2% can flip the model’s decision.
Her repair:
measure impact, not uncertainty size alone.
6. Ethan’s sensitivity error is exploring impossible ranges
He varies a probability from -2 to 4.
His repair:
respect scientific constraints and plausible parameter domains.
7. Primary Science can learn sensitivity intuitively
Add a little more water.
Does the plant response change much?
Move the light slightly farther away.
Does the measured brightness change substantially?
The idea begins with controlled variation.
8. Primary 3 sensitivity can stay qualitative
Which factor makes the biggest difference?
Which change hardly matters?
Children learn to rank influence.
9. Primary 4 can compare equal-sized input changes
Change each candidate factor by a comparable amount.
Observe which response is largest.
Fair comparison supports sensitivity reasoning.
10. Primary 5 can connect sensitivity to limiting factors
When water is abundant, adding more water may change little.
When water is limiting, a small increase may matter greatly.
Sensitivity depends on the system state.
11. Primary 6 can connect sensitivity to graph shape
Steep region:
small input change produces large output change.
Flat region:
input changes produce little response.
Graph gradient becomes sensitivity.
12. Secondary Science makes sensitivity quantitative
How much does current change per unit voltage?
How much does reaction rate change per degree?
How much does model output change per parameter change?
Sensitivity becomes measurable.
13. Local sensitivity examines nearby changes
Start at the current parameter value.
Change it slightly up and down.
Observe local response.
This resembles a derivative in Mathematics.
14. Global sensitivity explores a wider range
Some models behave differently far from the starting point.
Global analysis examines influence across the plausible parameter space.
15. One-at-a-time analysis is easy to interpret
Vary Parameter A while others stay fixed.
Then Parameter B.
This reveals individual effects clearly.
16. One-at-a-time analysis can miss interactions
Parameter A matters only when Parameter B is high.
Testing each independently may miss the combined effect.
Interaction-aware sensitivity is richer.
17. Sensitivity depends on scale
A parameter may dominate at microscopic scale and matter little macroscopically.
Another may become important only at long time horizons.
The analysis must match the intended scale.
18. Sensitivity depends on operating point
A system can be insensitive in one region and highly sensitive near a threshold.
There is no single universal sensitivity for many nonlinear systems.
19. Thresholds create high-sensitivity regions
Near a critical boundary, tiny parameter changes can switch the system state.
See How Scientific Thresholds Work.
20. Saturation creates low-sensitivity regions
Once a process is near maximum capacity, adding more input may change little.
This reveals a constraint.
21. Sensitivity analysis identifies leverage points
If one controllable input strongly affects an important output, it may be an effective intervention point.
Systems thinking uses this to find high-impact changes.
22. High sensitivity can be useful
A sensor should respond clearly to the quantity it is designed to detect.
High sensitivity improves detection when noise remains manageable.
23. High sensitivity can also create fragility
If a model conclusion changes wildly with tiny uncertain inputs, predictions become unstable.
The same property can help measurement and hurt decision robustness.
24. Sensitivity and robustness are opposites in one useful sense
A conclusion insensitive to plausible assumption changes is robust.
A conclusion highly sensitive to them is fragile.
See How Scientific Robustness Works.
25. Sensitivity and parameter estimation are inseparable
If the output is very sensitive to Parameter A, estimating A accurately becomes important.
If the output barely responds to Parameter B, spending enormous effort refining B may add little value.
26. Sensitivity analysis prioritises measurement
Which uncertain quantity matters most to the conclusion?
Measure that one better first.
This turns uncertainty analysis into an experimental strategy.
27. Sensitivity analysis prioritises research
Ten parameters are uncertain.
Only two materially affect the decision.
Future experiments can focus on those two.
Science becomes more efficient.
28. Sensitivity can reveal irrelevant complexity
A model includes twenty parameters.
Five have almost no influence across the intended range.
The model may be simplified without losing meaningful predictive power.
29. Sensitivity can reveal hidden dependency
One supposedly minor parameter dominates output.
The scientific explanation may have underestimated that mechanism.
Model interpretation should change.
30. Sensitivity and uncertainty are different
Uncertainty asks how poorly an input is known.
Sensitivity asks how strongly the output depends on it.
Risk is greatest when an input is both uncertain and influential.
31. An uncertainty-impact matrix is useful
Low uncertainty, low sensitivity:
little concern.
High uncertainty, low sensitivity:
may still matter little.
Low uncertainty, high sensitivity:
monitor carefully.
High uncertainty, high sensitivity:
priority for better evidence.
32. Sensitivity and probability can be combined
Instead of changing parameters one at a time, assign probability distributions to uncertain inputs and propagate them through the model.
The output becomes a distribution of possible outcomes.
33. Monte Carlo methods support probabilistic sensitivity
Sample plausible inputs repeatedly.
run the model.
compare which inputs explain most output variation.
This connects probability, simulation and sensitivity.
34. Sensitivity and simulation are natural partners
Simulation makes parameter changes easy to run.
Sensitivity analysis gives those runs scientific purpose.
See How Scientific Simulation Works.
35. Sensitivity and experimental design are natural partners
Test the system where competing parameter values generate clearly different predictions.
This maximises information gain.
36. Sensitivity and anomaly detection are linked
An output suddenly changes dramatically after a tiny input adjustment.
That may indicate a threshold, numerical instability or hidden interaction.
The anomaly becomes diagnostic.
37. Sensitivity and validation are linked
A validated model should not merely fit known data.
Its conclusions should also behave plausibly when inputs are varied within real-world ranges.
38. Sensitivity and standards are linked
If a result is highly sensitive to measurement tolerance, the standard may need tighter control.
Operating specifications can be designed around sensitivity.
39. Sensitivity and decision-making are linked
A decision should be tested under plausible alternative assumptions.
If the preferred option remains the same, the decision is robust.
If it flips easily, uncertainty should be made explicit.
40. Tornado diagrams can visualise sensitivity
Inputs are ranked by how much their plausible variation changes the output.
The most influential variables appear largest.
This helps decision-makers see where uncertainty matters.
41. Scenario analysis is a coarse form of sensitivity testing
Best case.
central case.
worst plausible case.
Scenarios help explore how conclusions change under different combinations of assumptions.
42. Sensitivity analysis should use plausible ranges
If a parameter is realistically between 4.8 and 5.2, testing from -100 to 100 may produce dramatic but irrelevant behaviour.
Plausibility protects interpretation.
43. Sensitivity analysis should preserve scientific constraints
Probabilities stay between 0 and 1.
mass is not negative in ordinary contexts.
material limits apply.
Exploration should remain inside physically meaningful space unless the purpose is explicitly to test model failure.
44. AI systems need sensitivity analysis
Change the prompt slightly.
change the input image.
change demographic context.
change retrieval source.
Does the output change reasonably or unpredictably?
AI reliability includes input sensitivity.
45. AI prompt sensitivity can reveal fragile reasoning
If semantically equivalent wording produces contradictory answers, confidence should fall.
The system may be relying on surface form rather than stable reasoning.
46. AI can help perform sensitivity analysis
Useful prompts:
“Vary this parameter by ±10% and compare outputs.”
“Rank assumptions by effect on the conclusion.”
“Find the parameter value where the decision changes.”
“Identify interactions that one-at-a-time analysis would miss.”
47. Parents can teach sensitivity through everyday systems
“If we leave five minutes later, how much does arrival time change?”
“If we add a little more water, does the plant response change?”
“Which small change matters most?”
This builds leverage-point intuition.
48. Small-group tuition can run sensitivity maps
Give one scientific system with four possible input changes.
Each student predicts which matters most.
Then test or simulate.
Compare intuition with evidence.
49. A compact sensitivity-analysis checklist
- What output or conclusion matters?
- Which input or assumption is being tested?
- What plausible range should it take?
- What scientific constraints apply?
- How much does the output change?
- Does sensitivity vary across the operating range?
- Are there thresholds?
- Do inputs interact?
- Which uncertain variables matter most?
- Which measurements deserve improvement first?
- Does the decision change?
- What new experiment would reduce the most important uncertainty?
50. Frequently asked questions
What is scientific sensitivity analysis?
It is the systematic study of how changes in model inputs, parameters or assumptions affect outputs and conclusions.
Why is sensitivity analysis useful?
It identifies influential variables, reveals fragile conclusions, prioritises better measurements and helps test robustness.
What is the difference between sensitivity and uncertainty?
Uncertainty describes how poorly an input is known; sensitivity describes how strongly the output responds to that input.
Why do thresholds matter?
Near a threshold, very small input changes can create large output changes, producing high local sensitivity.
How does sensitivity analysis help PSLE Science?
It helps learners identify which factor matters most, interpret steep and flat graph regions and understand limiting factors.
How does it change in Secondary Science?
It becomes more quantitative through gradients, model parameters, simulations, uncertainty propagation and interactions.
51. Continue the Science Education Systems series
- How Scientific Variables Work
- How Scientific Rates Work
- How Scientific Thresholds Work
- How Scientific Robustness Works
- How Scientific Parameter Estimation Works
Conclusion: Sensitivity analysis finds the assumptions carrying the weight
Maya changes one input.
Jia Jun measures the output shift.
Hana asks whether the input was uncertain.
Ethan searches for interactions and thresholds.
Science needs all four.
Perturb.
compare.
rank.
stress-test.
find the leverage point.
then spend the next experiment where better knowledge will matter most.
