Did you know that an experiment can look wonderfully precise on a computer screen and still tell us nothing about what a real sensor actually measured? A colourful graph is persuasive; its origin matters even more.
A science simulation calculates what a model predicts under specified assumptions. Treat its output as model-generated evidence about that model, not as an observation collected from the world. Students can learn a great deal by changing inputs, making predictions and testing explanations, but they must label simulated results honestly and use physical measurements or independent evidence when making claims about reality.
Why the distinction matters in ordinary science lessons
A learner may move a slider and watch particles move, a pendulum swing or water run downhill. The interaction can make an abstract relationship easier to explore. Yet the computer is following programmed rules. If it displays a smooth line, that line may be smooth because the model excludes wind, friction or measurement noise—not because nature supplies perfectly smooth readings.
The right question is not ‘Are simulations real science?’ Scientists use models and computational experiments for important research. The useful distinction is which question was actually investigated: how a model behaves, what a physical system does, or how closely one predicts the other. A simulation may be a rigorous experiment within a computational model while remaining insufficient on its own to establish a real-world claim.
Worked example: a simulated rainstorm and a measured water collection
Imagine an invented classroom drainage model. A simulation takes an input of 40 litres of water and assumes that 25% emerges as collected runoff during a specified period. The expected collected volume is 40 × 0.25 = 10 litres. This is a calculated model output, not a measured result. A separate physical tray activity uses 40 litres under its own conditions and records 12 litres in the collection container. These numbers are invented solely to teach reasoning; they are not evidence from a real school investigation.
The model predicted 10 litres; the physical activity recorded 12 litres, a difference of 2 litres. Relative to the model prediction, 2 ÷ 10 × 100% = 20%. Relative to the 40-litre input, the difference is 5 percentage points of the input volume. The two percentages answer different questions. Neither proves the software is useless, the tray is faulty or that all drainage systems behave alike.
- Prediction: simulated collected runoff, 10 L, conditional on the model’s 25% assumption.
- Observation: measured collection in the physical activity, 12 L, subject to the accuracy of the equipment and procedure.
- Comparison: observed minus predicted = +2 L, with the same collection period and units stated.
- Next check: are the surface, slope, water input, timing, catchment boundaries and collection losses genuinely comparable?
If the physical setup retained water after the measurement window, or the simulation assumed a different slope, comparing the numbers without correcting the conditions would be unfair. If the conditions are matched, repeat carefully and investigate uncertainty. If they are not matched, write ‘the activities used different conditions’ rather than inventing a hidden correction factor.
An accessible four-stage learning sequence
- Predict before clicking. Write the direction of change and a reason, not just a guessed number.
- Manipulate one model input while keeping other displayed settings constant; identify which outputs the model changes.
- Record the result in a table headed ‘simulation output’ or ‘model prediction’, never simply ‘experiment reading’.
- Explain which results follow from model assumptions, and identify what would need real-world checking.
This sequence is deliberately different from using a simulation as a digital answer key. A child who says ‘The bar went up when I increased the input’ has described a display. One who says ‘The bar went up because the model treats this input as increasing the rate; I would want to check that assumption outside the screen’ is starting to explain.
Which kind of simulation are we using?
Not every simulation does the same job. Some display a simplified conceptual analogy, designed to make an invisible process imaginable. Others compute from mathematical relationships and allow quantitative exploration. Some include randomness, so repeated runs may vary. A classroom needs to know whether the purpose is concept-building, prediction, parameter sensitivity or comparison with measured evidence.
The University of Colorado Boulder’s PhET project provides research-based interactive science and mathematics simulations. Its stated teaching approach is exploration and discovery. The simulations can support learning, but their popularity or visual clarity is not a reason to claim that a particular classroom activity established a measured natural result. Always read the selected simulation’s actual description, controls and limitations.
A diagnostic conversation with a student
Ask: ‘If the screen shows a temperature of 80 °C, who or what determined that number?’ If the child answers ‘The thermometer’, ask whether a physical thermometer was connected. If not, the model calculated or displayed a value, even if a virtual thermometer icon appears. Now ask, ‘What would a real thermometer add?’ The student should mention a physical observation under documented conditions and possible measurement uncertainty.
A second useful prompt is ‘Would changing the programme’s assumptions change what we see?’ If the learner cannot identify any assumptions, invite them to notice adjustable parameters and deliberately simplified features. Not knowing the internal code is fine; pretending the output is assumption-free is not.
Three mistakes that make polished graphs misleading
- Labelling a simulated number ‘measured’ because the screen resembles lab equipment.
- Comparing two curves drawn under different input settings without recording what changed.
- Treating a successful model fit as proof of a unique explanation; different mechanisms can sometimes produce similar outputs.
Correct the mistake in the notebook itself. Label the type of evidence, include units, identify the model version or settings where practical, and state the physical question left open. If a screenshot is used, its caption should explain which parameters were changed. Clear provenance matters even when the arithmetic is correct.
What teachers can assess instead of screen time
Give pairs of learners two runs from the same model and ask them to isolate the input responsible for a change. Then let them predict a third run before viewing it. A learner can pass a genuine understanding check only by giving a reason and noticing the model’s boundaries, not by producing a neat screenshot.
For a class with mixed confidence, begin with a visible everyday analogy before adding symbolic rules. Discuss why the analogy works and where it breaks. An extension group might compare sensitivity: change one parameter by 10% and measure the change in the output, making sure the base of each percentage is explicit. A pupil who can transfer the question to a new simulation is demonstrating broader reasoning.
A parent’s simple follow-up after digital science homework
Ask three brief questions: ‘What did you control?’, ‘What did the program calculate?’, and ‘What would you need to observe in real life to check the claim?’ These work better than asking a child to memorise every button. They also show whether the learner can describe the relationship when the animation is not in front of them.
You do not need specialist software to check progress. On another day, present a short written scenario with a calculated prediction and a measured result, and ask the learner to label each and explain the difference. The transfer question prevents confidence from being based only on familiarity with the interface.
When simulation is the right tool—and when it is not
A simulation is useful when real apparatus is unsafe, expensive, too slow, inaccessible or unable to show hidden mechanisms. It can also support many controlled comparisons. But a computer model may omit interactions, use uncertain inputs or represent a range of situations imperfectly. Physical experiments have their own limitations too: instruments may be miscalibrated, samples unrepresentative and procedures inconsistent. Neither type of evidence deserves blind trust.
When the question concerns a particular real-world outcome, consider whether observations, validated data or independent experimental findings support the model. When the task is explicitly to explain how the model works, avoid judging the learner unfairly for not doing an unavailable physical test. Match the evidence standard to the claim.
An independent practice question with an answer
A model predicts a plant height of 18 cm after four weeks under its specified rules. A student then measures an actual plant at 16 cm. Which statement is justified? (A) The model was measured at 18 cm; (B) the actual plant grew 2 cm less than the model predicted; (C) all plants grow 16 cm in four weeks. Only B accurately describes these two numbers. A confuses prediction with measurement; C makes a population claim from one plant. Even B needs comparable starting heights, growing periods and units to serve as a useful evaluation of the model.
Now alter the case: suppose the student measured from the soil surface while the model measured from the pot base. The 2 cm difference is no longer interpretable as a valid growth comparison until measurement definitions are aligned. This is the habit we want children to transfer to unfamiliar tasks.
Frequently asked questions
Is a virtual experiment fake?
No. It can be a genuine investigation of a computational model. The problem arises when model outputs are labelled as independent observations of physical reality.
Should students always test a simulation with real apparatus?
Not always. Cost, safety, time and access may make that unrealistic. They should still explain what the simulation does and does not establish.
Does a small difference prove the model is accurate?
Not by itself. Consider measurement uncertainty, input assumptions, the number of comparisons and whether the model predicts new situations, not merely one convenient case.
What if the simulation gives a different answer after a repeat?
It may include randomness or a setting may have changed. Record settings, seeds if available, and repeated outputs before making a conclusion.
What helps a younger child learn this?
Use plain contrasts: ‘The computer predicts’ and ‘We actually observed.’ Then build from concrete events to a diagram or equation.
Continue the wider learning journey
See the eduKate Education learning library for transfer from schoolwork to better decisions, and our Clementi Mathematics benchmark for diagnosis, guided practice and independent checking. The important lesson here is not distrust of software. It is respect for the difference between a useful model, a clear prediction and a real measurement.