How to teach fair tests and variables to children

"Is it a fair test?" is one of the most useful questions a child can learn to ask, in science and out of it. It applies to a toy race, a taste test and a news story that says one thing causes another. The idea is simple, but it takes practice, because children (and adults) naturally change several things at once. This guide explains how to teach fair testing to children of about 8 to 12, using a safe experiment with a ball and a ramp, with the arithmetic done out in full.

The core idea: change one thing

A fair test compares what happens when one thing changes and everything else that could matter stays the same. Scientists call the things that can change variables, and give them three roles:

• the independent variable, the one thing you change on purpose;
• the dependent variable, the thing you measure to see what happened;
• the controlled variables, everything else you keep the same.

Children usually find the words easier after the idea. Start with questions in plain language: "What are we changing? What are we measuring? What must we keep the same?" Introduce the names once they can answer.

What the research says about teaching it

The skill is called the control of variables strategy. In a well-known study, Chen and Klahr (1999) worked with children in the early primary years and looked at how they learned to design unconfounded experiments, where only one thing differs, and whether they could carry the skill over to new problems. They compared several kinds of teaching. The main finding was that explicit instruction in the strategy, with probing questions, helped children learn it and apply it to new situations, more than simply letting them explore alone.

Take that with care. It is one study, with particular tasks and ages, and it does not say that every child learns the same way or that every teaching method works. What it supports is a modest, practical point: telling children directly what makes a test fair, and asking them questions about their own designs, is worth doing. Do not rely on discovery alone.

Start with a question that can be tested

A testable question names what you will change and what you will measure. "Which surface makes a ball roll furthest?" is testable. "Which surface is best?" is not, because "best" is not measured. Rewrite vague questions together, and ask: "How would we know?"

Then predict. A hypothesis is a prediction with a reason: "I think the ball will roll furthest on the smooth floor, because there is less to slow it down." Accept wrong predictions as normal. The prediction is a starting point, not a grade.

A safe worked example: how far does a ball roll?

You need a ramp (a board propped on a book), a ball, a tape measure and four floor surfaces: a smooth floor, cardboard, carpet and a folded towel. The question is: how far does a ball roll after it leaves the ramp on each surface?

Plan the test with the children:

• Changed (independent): the surface.
• Measured (dependent): the distance rolled from the bottom of the ramp, in centimetres.
• Kept the same (controlled): the ramp and its height, the ball, the starting point on the ramp, how the ball is released (let go, do not push), and where you measure from.

Run three trials on each surface and write down every result. The numbers below are invented, to show the method. They are not data from any real run.

The arithmetic for the first row: 150 + 145 + 155 = 450, and 450 / 3 = 150 cm. The range is the largest value minus the smallest: 155 - 145 = 10 cm. Work out the others the same way. For the towel: 40 + 44 + 36 = 120, and 120 / 3 = 40 cm; the range is 44 - 36 = 8 cm.

Reading the results honestly

Now compare. The smooth floor mean is 150 cm and the cardboard mean is 130 cm, a gap of 20 cm. The ranges are 10 cm and 8 cm. The gap is bigger than the spread within either group, so the data support the claim that the ball rolled further on the smooth floor than on cardboard, in this test.

Now imagine a different pair of results: surface A with a mean of 150 cm and a range of 10 cm, and surface B with a mean of 146 cm and a range of 12 cm. The means differ by 4 cm, but the trials of each overlap and the spread is larger than the gap. You cannot say from this which is further. You would need more trials or a more careful method. Teaching this contrast, a clear gap against an unclear one, is one of the most valuable things a fair-testing lesson can do.

Finally, say what the data do not show. They apply to this ball, this ramp and these surfaces. They do not show that every smooth surface is faster than every rough one. Three trials are a small sample, and a repeated test on another day could differ.

Graphs and conclusions

A bar chart of the means is the natural display for this data: one bar for each surface, the bar height showing the mean distance. Ask children to label the axes with a name and a unit, and to give the graph a title that states what it shows. Then write a conclusion in three parts: what the data show ("The ball rolled furthest on the smooth floor, a mean of 150 cm"), whether that matches the prediction, and what could be improved ("We could use more trials or release the ball with a holder to make the start more exact").

Repeating and reproducing

Repeating a measurement, as above, checks that a result is not a fluke. A larger idea is that other people should be able to get the same result by following your method. The National Academies' 2019 report on reproducibility and replicability in science sets out these ideas in detail for researchers. For children, the version is: write the method so clearly that a friend could do it without asking you any questions, and see whether they get a similar result. If they do not, the next job is to find out why.

Common mix-ups

• Changing two things at once. A child tries a bigger ball on a rougher surface. Ask which change caused the difference.
• One trial treated as proof. "It rolled 160 once" is not a result.
• Measuring differently each time. Always measure from the same spot, to the same point on the ball.
• Hiding the odd result. Record it, then investigate. It may show a mistake, or something interesting.
• Claiming too much. The data are about this setup.

Tool that fits

Think Like a Scientist (EUR 12, paid once) is a set of 10 posters that take a child through an investigation: observation, questions, hypothesis, fair tests, variables, measurement, data tables, graph reading, conclusions and reproducibility. One simple, safe worked example, how far a ball rolls on four surfaces, runs through the whole set, with sample results and every mean and range calculated. You get print-ready PDFs in A4 and US Letter (10 + 10), 10 PNG files at 3508 x 2480 pixels, 10 editable SVG files and a 4-page educator guide with a run sheet, discussion prompts, common misconceptions and a checklist. Digital download only.

Questions people ask

Sources

1. Chen and Klahr (1999), All other things being equal: acquisition and transfer of the control of variables strategy, Child Development 70(5), 1098-1120
2. National Academies of Sciences, Engineering, and Medicine (2019), Reproducibility and Replicability in Science
3. HiddenWay worked example: invented measurements, our own arithmetic

We checked each source on 17 October 2026. If a fact here is out of date, tell us at hello@hiddenway.info.

Keep reading

• Retrieval practice and spaced review for children's science
• How to explain phase changes and latent heat to children
• How to teach Earth systems with simple diagrams (ages 8-12)
• More in Science for children