Singapore School Mathematics Operating Manual · Chapter 21
Mathematics can describe the world powerfully because it simplifies.
A journey becomes distance, time and speed. A population becomes a growth rule. A school budget becomes income, cost and constraint. A shape becomes lengths, angles and relationships. A probability model replaces messy uncertainty with a defined sample space. A graph turns a changing quantity into a mathematical object that can be analysed.
The simplification is the strength of the model—but also its boundary.
This chapter develops a discipline for assumptions, model validity, limitations and return to context. The central question is not simply “can I calculate this?” It is: what had to be assumed for this calculation to describe the original situation, and does the conclusion remain trustworthy under those assumptions?
Assumptions · Model building · Validity · Limitations · Return to context · Practice · Worked answers
1. Every model keeps some information and discards some information
Suppose a 120 km journey is represented using one average speed of 60 km/h.
The mathematical model may ignore traffic lights, acceleration, road gradients, rest stops and changing speed. It keeps only the quantities needed for the chosen question.
If the question is “what travel time corresponds to an average speed of 60 km/h over 120 km?”, the simplification is sufficient.
If the question is “what is the maximum engine temperature during the journey?”, the model is useless.
A model is not simply right or wrong. It is suitable or unsuitable for a particular purpose.
2. An assumption is a condition accepted so the model can operate
Common school-model assumptions include:
constant speed, uniform density, negligible thickness, no leakage, equal likelihood, perfect circles, straight-line travel, constant interest rate, independent trials, identical units, exact measurements and no external forces.
Some are printed explicitly. Others are implied by the level of the question.
The important habit is to notice them rather than letting them disappear into the formula.
3. Assumptions are not automatically weaknesses
A useful model must simplify.
The idealisation of a particle as a point object can be very effective when size does not matter. Treating a road segment as straight may be sufficient for a map-scale exercise. Treating a rate as constant may be appropriate over a short time interval.
The problem begins when the omitted detail is large enough to change the answer or conclusion.
4. The same assumption can be reasonable in one task and unreasonable in another
Suppose water flows into a tank at 5 L/min.
For a ten-minute classroom calculation, assuming constant inflow may be fine.
For predicting the exact water level over several days, real pump cycling, pressure changes and leakage may matter.
Model validity therefore depends on scale, purpose and tolerance.
5. A model begins by choosing variables
Let t represent time and V(t) the volume of water in a tank.
If inflow is constant at 5 L/min and there is no outflow, a simple model is V(t) = V₀ + 5t.
The equation is not the tank. It is a representation of one aspect of the tank under stated assumptions.
Defining variables clearly prevents the model from drifting away from the original meaning.
6. Relationships should match the mechanism
A linear model assumes constant additive change.
An exponential model assumes multiplicative change proportional to the current quantity.
An inverse model may describe a fixed-total or fixed-product relationship.
A quadratic model may arise from area, constant acceleration or other second-order structure.
Choosing a formula because it “looks familiar” is weaker than choosing it because the mechanism supports the relationship.
7. Units help validate model structure
If V is measured in litres and t in minutes, the coefficient 5 in V = V₀ + 5t must carry units L/min.
Then 5t has units litres and can be added to V₀.
A unit mismatch can expose a structurally invalid model before numerical substitution.
This connects directly to Mathematical Type Checking.
8. A model can be exact relative to its assumptions
Suppose a perfect rectangle has length 8 cm and width 5 cm.
Under Euclidean geometry, its area is exactly 40 cm².
If those dimensions came from physical measurement, the physical object’s actual area may still carry uncertainty.
The mathematics can be exact inside the model while the model inputs are approximate in the real world.
9. Validity asks whether the model remains fit for the question
A model is valid enough when its assumptions support the required conclusion to the required accuracy.
This does not require perfection.
If a rough estimate is requested, a simplified model may be excellent. If a safety-critical threshold were involved in a real setting, a classroom approximation could be inadequate.
School Mathematics should teach the distinction between mathematical calculation and professional operational decision-making.
10. Internal validity comes before external validity
First ask whether the mathematics has been done correctly under the model.
Then ask whether the model itself describes the intended situation well enough.
A flawless calculation from a poor model can still give a poor real-world conclusion.
Conversely, a good model executed incorrectly also fails.
The two layers should be checked separately.
11. Parameter ranges can define where a model is valid
A linear approximation may work well near one point but poorly far away.
A formula may assume 0 ≤ t ≤ 10.
A probability model may apply only while trials are independent.
A growth model may assume resources are unlimited over a short interval.
The model’s domain is part of the validity statement.
12. Extrapolation is a model-validity stress test
A line fitted to data between x = 0 and x = 10 may predict reasonably at x = 8.
Using the same line at x = 100 assumes the relationship continues far beyond observed evidence.
The arithmetic may be easy. The assumption is strong.
This is why extrapolation needs more caution than interpolation.
13. Model limitations should name what was omitted
“The model is not perfect” is too vague.
A useful limitation statement identifies the omitted mechanism:
speed may vary; measurement error is ignored; outcomes may not be equally likely; demand may change over time; the shape may not be a perfect cylinder; the sample may not represent the population; the interest rate may change.
Specific limitations support better judgement.
14. A limitation matters most when it can change the conclusion
Suppose a bus capacity model gives 39 passengers for a 40-seat bus.
If the count is exact, one bus is enough.
If the passenger estimate could actually be 42, uncertainty crosses the capacity threshold and changes the decision.
A limitation becomes decision-critical when it can move the result across a boundary.
15. Some omitted details only change precision, not conclusion
Suppose a calculated distance is approximately 4.9 km and the only decision is whether it is under 20 km.
Small measurement uncertainty is unlikely to change that conclusion.
The model may be imprecise yet still decision-stable.
This links to Parameter Sensitivity and Stability.
16. A model can fail structurally
If cost includes a fixed fee, C = kn is incomplete.
If a discount begins after 100 items, one straight-line rule may fail.
If speed changes with time, one constant-speed equation may fail.
If probability changes after each draw without replacement, identical-trial assumptions fail.
Structural failure means the relationship itself has changed, not merely the parameter values.
17. Piecewise models can repair structural change
Suppose parking costs $3 per hour for the first two hours and $5 per hour afterward.
A single rate does not describe the entire system.
A piecewise model does:
C(t) = 3t for 0 ≤ t ≤ 2, and C(t) = 6 + 5(t − 2) for t > 2.
The model becomes more faithful by acknowledging a regime change.
18. More complexity is not automatically better
A model with twenty parameters may fit old data beautifully but be difficult to interpret or unstable on new data.
For school Mathematics, the simplest model that captures the required relationship is often best.
Complexity should be justified by a need, not added for appearance.
19. Assumptions should be testable where possible
If a model assumes constant rate, compare measurements across intervals.
If it assumes linearity, inspect residual patterns or graph shape.
If it assumes equal likelihood, ask whether the physical mechanism supports symmetry.
If it assumes independence, ask whether one event changes the state for the next.
Testing assumptions turns modelling into evidence-based reasoning.
20. A model can be useful even when known to be false in detail
A map is not the territory.
A straight-line approximation to a curved road may still estimate distance effectively over a short segment.
A simple interest model may be educationally useful even though real financial products involve additional terms and conditions.
The correct question is whether the simplification preserves what matters for the current purpose.
21. The return to context is part of the solution
Suppose the algebra gives x = 3.7.
If x represents metres, 3.7 m may be final.
If x represents buses, 3.7 buses is not an admissible operational answer.
If x represents a probability, 3.7 is impossible.
If x represents years in a continuous growth model, interpretation may depend on whether fractional years are meaningful.
The symbol becomes an answer only after its context is restored.
22. Interpretation should match the requested object
If a model predicts 4.2 containers are needed, the practical answer may be 5 containers if every item must fit.
If the question asks for average utilisation, 4.2 may be meaningful as a ratio rather than a count.
The same number can require different final decisions because the output job differs.
23. A model result should not claim more certainty than the inputs support
If measurements are approximate, the final result may also be approximate.
If assumptions are uncertain, conclusions may be conditional.
Useful language includes “under the stated model”, “assuming the rate remains constant”, “approximately”, “within the observed range”, or “this does not guarantee”.
Precision of language is part of mathematical integrity.
24. Correlation models do not automatically establish causation
A strong relationship between two measured variables does not by itself prove one causes the other.
A third variable, selection effect or reverse relationship may explain the pattern.
Mathematics can quantify association without settling causal mechanism.
Returning to context means respecting what the model can and cannot claim.
25. Optimisation depends on the chosen objective
A “best” design is meaningless until best is defined.
Minimum cost, minimum time, maximum capacity and maximum reliability can favour different solutions.
Constraints also matter.
An optimisation result is therefore conditional on objective and feasible set.
26. A probability model is conditional on its sample space and mechanism
A fair die gives each face probability 1/6 under the ideal fair-die model.
If the die is biased, the same numerical assignment is no longer valid.
The formula did not fail. The model assumptions changed.
27. Financial mathematics examples need especially careful return language
A school compound-interest calculation can teach exponential growth.
Real financial decisions may involve fees, taxes, changing rates, legal terms, risk and individual circumstances.
Educational Mathematics can explain the mechanism without pretending to provide investment or financial advice.
28. Measurement models carry instrument limits
A ruler marked to the nearest millimetre cannot justify arbitrary decimal precision in a final physical conclusion.
More calculator digits do not create more measurement information.
The model should carry the precision of its evidence.
29. Simulation is a model, not reality itself
A simulation can explore thousands of possible outcomes under defined rules.
It is powerful for uncertainty and complex systems.
But the results remain conditional on the rules, distributions and dependencies programmed into the simulation.
A simulation cannot rescue an invalid assumption by repetition.
30. Model comparison should focus on purpose, fit and interpretability
When two models are possible, compare:
Which assumptions are stronger? Which fits the observed range? Which behaves sensibly at boundaries? Which is easier to interpret? Which supports the decision being asked? Which fails more gracefully when inputs change?
The mathematically most complicated model is not automatically the most useful.
31. A model should survive a reasonableness check
If a model predicts negative population, probability above 1, infinite speed from a modest input, or cost decreasing when every unit price is positive, investigate.
Some surprising results are real; many expose domain or structural failure.
Reasonableness is not proof, but it is a powerful alarm.
32. A practical model-validity audit
Ask:
What is being represented? What variables matter? What has been ignored? Which assumptions are explicit or implicit? Over what domain does the relationship make sense? Are the units and types consistent? How sensitive is the conclusion? Does the model cross any threshold? What would make the model fail? What does the result mean back in the original context?
A mature solution should be able to answer these questions in proportion to the level of the task.
33. Independent practice
1. A 180 km journey is modelled at constant speed 90 km/h. Find the travel time and state the main assumption.
2. A tank starts with 20 L and fills at 3 L/min with no outflow. Write a model for volume after t minutes.
3. State one reason the model in Question 2 could fail in a real tank.
4. A line fitted to data for 0 ≤ x ≤ 10 is used at x = 8 and x = 100. Which use is interpolation and which is extrapolation?
5. A calculator gives 2.4 buses. What must be considered before reporting a final answer?
6. A rectangle model uses measured dimensions. Explain how the mathematical area can be exact relative to recorded inputs but the physical area remain uncertain.
7. A fair-coin model assigns P(heads) = 0.5. What assumption gives that value?
8. A cost model C = 5n ignores a $20 fixed fee. Write a repaired model.
9. Why can a highly accurate calculation still give a poor real-world answer?
10. Give one limitation that could matter in a constant-speed travel model.
11. Explain why more decimal places do not automatically make a measurement-based answer more truthful.
12. State the final “return to context” question a learner should ask.
34. Worked answers
1. Time = 180/90 = 2 h. The model assumes 90 km/h is an appropriate constant or average speed for the whole journey.
2. V(t) = 20 + 3t litres.
3. Examples: inflow may vary, leakage may occur, the tank may overflow, or the pump may stop.
4. x = 8 is interpolation. x = 100 is extrapolation.
5. Whether buses must be whole numbers and whether every passenger must fit. If so, a ceiling-type decision may be required.
6. The formula exactly multiplies the recorded values under the rectangle assumption. The recorded measurements themselves may differ slightly from the true dimensions.
7. That the coin is idealised as fair, so heads and tails are equally likely.
8. C = 20 + 5n.
9. The model assumptions may not describe the real situation well enough even if the arithmetic is flawless.
10. Traffic, stops, acceleration, changing speed or route conditions.
11. Calculator precision cannot exceed the information quality of the measurements and assumptions.
12. “What does this mathematical result mean in the original situation, under the assumptions actually used?”
35. Continue through the School Mathematics operating manual
Use Scale, Order of Magnitude and Estimation to test whether a model output is plausible, Equivalence and Canonical Forms when several representations describe the same object, and Monotonicity and Ordering when direction of change can settle a decision without full calculation.
Return to the BTT Mathematics Hub for Batch 06.