Did you know that an experiment can be worth learning from even when the two results look almost the same? A student may feel disappointed after building a fair test and finding no clear winner. But not seeing a difference is a finding to describe carefully, not a reason to invent one.
When a school science experiment finds no clear difference, report what was actually measured, how variable the results were, and what the test was capable of detecting. Do not rewrite the prediction to match the outcome, claim that two things are universally identical, or call the work a failure merely because the result was unexpected.
What exactly does ‘no clear difference’ mean?
It might mean that the average values happened to match, that measurements varied too much to distinguish the conditions, or that a difference was smaller than the method could resolve. These are not identical explanations. A good scientific conclusion distinguishes the observation from its possible causes.
Before interpreting the outcome, check the investigation question. Did the student compare a single result or several trials? Were the measured units the same? Was the intended change actually applied? Were equipment and conditions comparable? A conclusion can be honest without being overly confident.
Worked example: two paper-bridge designs support the same average load
Imagine an invented classroom investigation comparing Design A and Design B, two different ways of folding equal-size sheets of paper into short bridges. Each bridge spans the same distance. Students add identical paper clips at the same central position, one at a time, and record the greatest number supported before the bridge fails. For each trial, they make a new bridge; they do not reuse a damaged one.
In the illustration, Design A supports 7, 8 and 9 clips across three trials. Design B supports 9, 7 and 8 clips. The calculation for A is (7 + 8 + 9) ÷ 3 = 24 ÷ 3 = 8 clips. B is (9 + 7 + 8) ÷ 3 = 24 ÷ 3 = 8 clips. Each design has the same observed range, 7–9 clips, and the same average in this tiny sample.
A careful conclusion reads: ‘In these three trials per design, both paper-bridge types supported an average of eight identical paper clips before failure. Our measurements did not show a difference in average load under the tested conditions.’ That is stronger than ‘Design A and Design B are exactly equally strong everywhere’ because it tells the reader the size of the observation and its limits.
The fictional data are not actual student results. In a real class, new bridge construction, paper grain, where the clip sits, how quickly clips are added and a one-clip measurement increment could influence the outcome. A larger or more consistent investigation may reveal a difference that this small test did not distinguish. It might also continue to find none.
Check four parts of the investigation before explaining the result
- Design: Was only the intended bridge construction changed? Keep paper dimensions, support spacing, clip type and placement as similar as reasonably possible.
- Replicates: Did each trial use a new test unit? Adding a second clip to the same bridge is not a second independent bridge trial.
- Measurement: Was the number of clips counted consistently, with the same decision about exactly when the structure failed?
- Records: Are all original observations available, including awkward or outlying trials, rather than only a neat summary?
If the procedure was not consistent, do not quietly remove the inconvenient result. Explain what changed, whether the work remains comparable and what a fairer retest would need. The Science Buddies guidance on conducting experiments emphasises clear records, controlled variables and transparent adjustments when a procedure needs revising.
How to write the explanation without proving too much
A science conclusion can use four sentences in sequence: question, measured result, supported inference and limitation. For example: ‘We tested whether the two paper folds differed in supported load. In three independent builds per design, both averaged eight clips. We did not observe a difference in this small set of trials. The coarse one-clip step and variation between builds limit what we can conclude about small differences.’
Do not write ‘there is no effect’ just because the averages match. At the same time, do not write ‘there must be a hidden effect’ simply because your prediction favoured one fold. Both would go beyond the evidence. When the measured distinction is unclear, uncertainty is an honest part of the answer.
Notice how this differs from declaring statistical significance. Three classroom trials per design with the numbers above are not a basis for inventing a p-value or guaranteeing a particular effect size. Students can first learn to read the raw measurements and describe what their method did—and did not—test.
What should the next investigation change?
Change the measurement method only for a documented reason. For the bridge test, a teacher may suggest finer load increments or a more consistent way to identify failure. A different design could provide stronger evidence if the first comparison did not meaningfully separate the conditions. Record the new method as a new round; do not mix the new and old trial results as though they used the same protocol.
If time permits, increase the number of independently rebuilt bridges and summarise both the centre and spread of results. But more repeats cannot repair an unfair comparison, such as letting Design A span a shorter gap. Repair comparability first. Our Core Aim article on pilot-testing investigations shows why this check belongs before full data collection.
A diagnosis-first approach for students, parents and teachers
Students: highlight the measured values in the table, calculate each average accurately, and say what those figures do not establish. Explain the first likely source of uncertainty without blaming yourself for every variation.
Parents: ask your child to tell you the difference between ‘we did not detect a difference’ and ‘we proved there is no difference’. Avoid making a disappointing result sound like wasted effort. A respectful question about procedure is more useful than demanding a positive result.
Teachers: include at least one fictional no-difference dataset alongside dramatic results. Ask learners to rank conclusions by how closely they fit the evidence. Model why a modest conclusion can be more scientific than an exciting but unsupported one.
How to observe progress
Give learners a second invented dataset—say, Condition C gives 5, 6 and 7 units while Condition D gives 8, 9 and 10 units under a consistently applied method. Ask them to calculate averages of 6 and 9 units, compare the spreads and explain why a visible difference still does not automatically establish a general causal rule. The progress signal is accurate calculation combined with appropriately limited language.
If students can separate raw observation, average, alternative explanation and next test without prompting, they have moved beyond merely reciting ‘repeat the experiment’. Where there are safety or equipment concerns, experiments must remain under suitable school supervision; a home demonstration is not a substitute for supervised laboratory work.
Frequently asked questions
Is a science experiment wrong if the original prediction was not supported?
No. Predictions can fail. A clear comparison and an honest account of what happened can still teach the learner about the system, the method and what evidence is missing.
Should an unusual measurement be erased?
Not merely because it is unusual. Check for a documented error and preserve the raw record. If a reading is excluded under an agreed rule, explain why and show how the conclusion changes.
Can identical averages hide important differences?
Yes. Different distributions can have the same average. Compare the raw trials, range, method and uncertainty rather than treating the mean as the entire finding.
The purpose of science is to learn what the test can support
The most valuable scientific sentence may be ‘We cannot tell yet, for these reasons.’ The related Core Aim guide to honest limitations develops the conclusion, while the eduKate education library connects this habit with independent checking. The untouched Clementi learning reference likewise treats reasoning and verification as more important than guessing a pleasing answer.