All temperatures used as observations here are illustrative, not recorded measurements or a student submission. The 15, 17.5 and 20 minute values are additional invented observations, chosen to demonstrate a residual pattern. This is a short reasoning example, not a complete IA.
1. Narrow the question
Instead of “the mathematics of coffee”, investigate: How well does Newton’s law of cooling model the temperature of a hot drink over its first 20 minutes in a room at 22°C?
This question defines a measurable outcome, a model and an observation period. In a real investigation, record the ambient temperature and explain how the thermometer, sampling interval and container affect the measurements.
2. Define and use the model
If the room temperature remains constant and the cooling rate is proportional to the temperature difference, a candidate model is:
Here, T is temperature in °C, t is time in minutes, and k is a positive constant in min⁻¹. The model starts at 82°C. Using an illustrative measurement of 62°C after 5 minutes:
The resulting prediction at 10 minutes is approximately 48.7°C. If an illustrative observation is 50.0°C, the residual (observed minus predicted) is about +1.3°C.
Keep one complete table
The starting value sets the temperature difference 82 − 22 = 60°C. The 5 minute value estimates k. The remaining four values are reserved for checking this fixed model; none is used to recalibrate it.
| Time (min) | Simulated (°C) | Model (°C) | Residual (°C) |
|---|---|---|---|
| 0Initial | 82.0 | 82.0 | 0.0 |
| 5Calibrate | 62.0 | 62.0 | 0.0 |
| 10Check | 50.0 | 48.7 | +1.3 |
| 15Check | 42.0 | 39.8 | +2.2 |
| 17.5Check | 39.5 | 36.5 | +3.0 |
| 20Check | 37.0 | 33.9 | +3.1 |
Residual = simulated temperature − model prediction. Table values are rounded to one decimal place after calculation with k = 0.0810930216… min⁻¹. “Initial” and “Calibrate” label the two values already built into the model; their zero residuals are not independent evidence of a good fit.
Reproduce the 10 minute check
Because e−5k = 40/60 = 2/3, the same equation can be written as:
At 10 minutes: 22 + 60(2/3)² = 48.666…°C.
Residual: 50.0 − 48.666… = +1.333…°C.
The model is about 1.3°C below this illustrative observation. Retain the unrounded parameter in subsequent calculations; repeatedly rounding k can introduce avoidable differences.
3. Evaluate the model
A plot of residuals against time can reveal a systematic mismatch. Mostly positive residuals late in the experiment, for example, would suggest the model predicts temperatures that are too low in that period. Random-looking residuals alone do not prove the assumptions are correct.
Compare the size of the errors with measurement uncertainty, and discuss possible changes in ambient temperature, heat loss through the container and evaporation. Do not assume a model fitted over 20 minutes remains equally useful outside that interval.
Read these residuals before proposing a cause
All four checking residuals are positive, rising from about +1.3°C to +3.1°C. For this deliberately constructed dataset, the fixed model increasingly underpredicts temperature across the later checks. In other words, it predicts more cooling than these illustrative values show.
Supported by this comparison
The discrepancy has a direction and changes with time. The early calibration does not describe these later invented values equally well. At 20 minutes the prediction is about 33.9°C, compared with a simulated value of 37.0°C.
Not established by the pattern
The residuals do not identify a physical cause. They cannot tell us whether room temperature changed, the probe lagged, a fixed-k assumption was unsuitable, or something else happened. Nor do four simulated checks establish statistical significance or real-world predictive accuracy.
For real measurements, compare discrepancies with justified measurement uncertainty and check the recording method before changing the model. A residual is not itself a thermometer uncertainty. Collect relevant evidence—such as an ambient-temperature record or repeated measurements—to distinguish explanations.
If you revise the model after seeing these checks, those observations have informed that revision. Explain the change and seek further observations for a fresh prediction check where feasible. Do not describe the same reused values as an untouched test set, or treat successive readings from one cooling run as automatically independent.
4. Before applying this to your IA
- Choose a question you can investigate with your own evidence.
- Explain each variable, assumption and fitting choice.
- Include graphs and enough observations to evaluate the model.
- Connect the mathematics and depth of analysis to your course and level.
This short excerpt illustrates reasoning and revision. It is not a complete IA and does not demonstrate every assessment criterion.
Keep this example with its own data. This page starts at 82°C in a 22°C room and uses 62°C at 5 minutes. The separate Mathematics AA sample report uses another dataset and model; do not combine its values or residuals with this one. For broader model comparison, see the Mathematics AI regression teaching example, which is labelled for that course.
Mathematical references
The data, added observations, calculations and residual figure on this page are original teaching material. These references explain the mathematical ideas; they do not validate the invented observations or assign an IB mark.


