Did you know that a science project can be useful even when its observations go in the opposite direction from the student’s prediction? Many learners think an experiment succeeds only if the hypothesis is ‘right’. That expectation can turn a normal surprise into pressure to change the claim, hide a trial or write a conclusion stronger than the evidence allows.
The correct response is to report what was predicted, what was actually measured and what the evidence supports—separately and truthfully. If the observations do not support the original hypothesis, say so. Do not rewrite the earlier prediction as if it had always matched the data. Explain possible reasons, measurement limitations and a sensible next test. This is learning from evidence, not a failed attempt to produce the ‘expected’ answer.
Begin by locating the real misunderstanding
Picture a student finishing a paper-airplane experiment. They expected a small paper clip to improve flight distance. Their recorded measurements are slightly shorter with the clip. They ask: ‘May I change my hypothesis so the conclusion looks right?’ Before answering, ask what the student believes a hypothesis is. A hypothesis is a testable proposed explanation or prediction, not a promise that reality must agree.
A good diagnostic is to show two possible opening sentences for the report: ‘I predicted the clipped plane would travel farther’ and ‘The clipped plane travelled farther.’ One is a statement about the prediction; the other is a statement about observed results. The student should be able to identify which is supported by which record. If not, more formula practice will not solve the problem: the student needs to distinguish an expectation from evidence.
How the scientific reasoning actually works
Before collecting data, researchers or students define a question, decide what variable will change, what will be measured, and what comparison is fair. They record a prediction based on a reason that can be examined. Once the experiment begins, the duty changes: measure consistently, preserve the original record, and interpret the results. A mismatch between hypothesis and observations can reveal a mistaken prediction, an unanticipated condition, measurement noise, limitations in design or several of these together.
Not supporting a hypothesis is not the same as proving its opposite. A small sample might be inconclusive. If the observed difference is smaller than measurement uncertainty or if the procedure was inconsistent, a careful conclusion may be ‘the evidence in these trials was insufficient to establish the predicted improvement’, rather than ‘the opposite is always true’.
The Science Buddies guidance on science-fair conclusions explicitly advises students to explain results that do not support their hypotheses and avoid manipulating outcomes to fit the prediction. In a school report, this is also a useful test of research integrity: the marks should reward a defensible connection between evidence and conclusion rather than a conveniently positive story.
Worked example: a paper clip that did not give the predicted result
All numbers here are invented for teaching, not results collected from pupils: A learner predicts: ‘Adding one 0.5 g paper clip at the marked point of my paper airplane will increase its mean horizontal flight distance compared with the same airplane without the clip.’ The supposed test uses the same folded model, an indoor launch marker, a consistent target direction and a floor tape measure. A teacher supervises safe launches. The outcome measured is horizontal distance from the launch line to first ground contact, in metres.
The illustrative no-clip distances over five trials are 4.2, 4.0, 4.1, 4.3 and 4.4 metres. Sum them: 4.2 + 4.0 + 4.1 + 4.3 + 4.4 = 21.0 m. Divide by five trials: mean = 4.20 m. The observed range is 4.0–4.4 m.
The invented clip distances are 3.8, 3.7, 3.9, 4.0 and 3.6 metres. Sum: 3.8 + 3.7 + 3.9 + 4.0 + 3.6 = 19.0 m. Divide by five: mean = 3.80 m. Observed range = 3.6–4.0 m. These are small sets of hypothetical classroom measurements; no statistical significance or general flight-law claim is implied.
The observed difference in mean distance is 3.80 − 4.20 = −0.40 metres for clip minus no clip. In this example, the clipped configuration travelled 0.40 m less on average across the five invented trials. If the learner wants a descriptive relative difference, 0.40 ÷ 4.20 ≈ 0.0952 or about 9.5% shorter relative to the no-clip mean; that percentage is not evidence that every clipped plane will travel 9.5% less.
The predicted direction was farther with a clip. The observed direction in this example was shorter with a clip. The correct immediate conclusion is therefore: ‘These five trials did not support the prediction that the clip would increase mean flight distance.’ It would be inaccurate to claim the experiment ‘proved’ all clips reduce distance or that the original prediction was always the opposite.
How to write a conclusion that remains useful
- Question: Does adding the specified paper clip change the mean flight distance for this paper-airplane setup?
- Original prediction: The clipped version will travel farther on average; preserve this wording from before the trial.
- Method and comparison: Name both conditions, what was measured, units, number of trials, and important controls or deviations.
- Observed result: No-clip mean 4.20 m from five trials; clip mean 3.80 m from five trials; difference −0.40 m.
- Claim with scope: ‘In the illustrative trials, the clip did not improve the mean distance and the data did not support my prediction.’
- Limits and next step: Different throw force, clip position, paper wear, launch angle and the small trial count may matter. Repeat under a more consistent procedure, without deleting the inconvenient observations.
Notice that the conclusion does not need emotional language such as ‘I failed’. It also does not use a strong word such as ‘proved’. It describes the question, the observed evidence and the boundary of that evidence. A curious learner might propose a new prediction for a new trial, such as ‘Moving the clip toward the centre will change the glide pattern.’ That new idea belongs under future investigation, not as a retrospective rewrite of the first hypothesis.
Three ways school experiments accidentally become misleading
Mistake 1: quietly rewriting the prediction
Changing ‘The clip will increase distance’ to ‘The clip will reduce distance’ after the results makes the story look perfect, but removes evidence of the learner’s original thinking and how that thinking changed. Keep the original with a date or version. Then explicitly write a new explanation or follow-up hypothesis where appropriate. Revision of understanding is welcome; rewriting the past is not.
Mistake 2: removing a trial because it ‘looks wrong’
Suppose the fourth clipped flight landed at 4.0 m, farther than the other clipped trials. That reading may be genuine variation rather than error. It should not be deleted merely because it weakens a neat conclusion. First check whether a documented procedural issue occurred: a gust from a fan, a different thrower, a measurement written in centimetres instead of metres, or a changed launch line. Record the decision and its reason. See our Core Aim article on investigating outliers before excluding data.
Mistake 3: treating a difference as a cause without checking the design
If every un-clipped flight was thrown in the morning and every clipped flight in the afternoon, fatigue or skill improvement could confound the comparison. If the same plane bent after repeated flights, the condition may not be the only changing factor. Alternating or randomising condition order where safe and feasible, defining launch procedures and recording the same outcome helps. Even good controls do not automatically establish universal causation from ten launches.
How to improve the next test without inventing a result
A sound follow-up begins with a new written plan, not corrected old numbers. Mark clip placement, use a clear flight-distance measure, photograph the design without including personal data, and record each raw trial promptly. If possible, standardise who launches, where the throw begins, and how the first ground contact is measured. Alternate the order of conditions, note any damaged paper and collect enough repeats for the instructional question.
For this made-up airplane investigation, real classroom outcomes would depend on many untested factors. The paper clip may change mass distribution, stability and drag, while human launch variability may dominate a small sample. Therefore, do not promise a particular result in a real demonstration. The goal is to practise setting up a comparison that could genuinely surprise the student.
Sometimes a null-looking result is itself meaningful only in a limited way: if measured values are close, it may mean the tested difference was too small to detect, the method lacked precision, or the proposed effect is absent in that setup. These are distinct possibilities. Before saying ‘there is no effect’, check the uncertainty, sample size, design and what size of change the method could reasonably notice.
Progressive practice: let the learner move from a sentence to a defensible report
Guided exercise: prediction or observation?
Read these two statements aloud: ‘I thought a shaded paper boat would dry more slowly’ and ‘The shaded paper boat took 18 minutes to appear dry.’ Ask which is prediction and which is observation. Point out that ‘appear dry’ is an observation rule that needs defining; it is not necessarily a direct measurement of remaining moisture. That small clarification prepares the pupil to design better work.
Supported exercise: check a complete dataset
Using the five-trial airplane numbers, have the learner recalculate both totals (21.0 m and 19.0 m), both means (4.20 m and 3.80 m) and the signed difference (−0.40 m). Have them write two sentences: one describing the data and another judging whether they support the prediction. They should not borrow the words ‘proved’ or ‘guaranteed’ from a template.
Independent exercise: a different expectation
Fresh fictional data: A pupil predicts that a design change will lower a marble’s travel time over a measured track. Without the change the three recorded times are 6.2 s, 6.0 s and 5.8 s; with the change they are 6.1 s, 6.3 s and 6.2 s. The first mean is (6.2 + 6.0 + 5.8) ÷ 3 = 6.0 s. The second mean is (6.1 + 6.3 + 6.2) ÷ 3 = 6.2 s. The difference is +0.2 s, not the predicted reduction. Ask the learner to state what the three-trial sample supports and what it cannot show conclusively.
Transfer test: refuse an impossible conclusion
Now remove the numerical examples. Hand the pupil an unfamiliar experimental graph and ask: ‘What had the investigator predicted? What did the measured outcome show? Which sentence would be unfair to write from this evidence?’ Progress is visible when the learner independently separates prediction, observation, explanation and uncertain inference—even when no arithmetic hint is supplied.
A helpful parent and teacher response
A parent hearing ‘My experiment went wrong’ can reply with an investigative question: ‘What did you expect, what did you measure, and what might explain the difference?’ That invites accuracy rather than embarrassment. Avoid telling the child to repeat a trial until the numbers match the desired answer. Repeat a test because the method needs improvement or a new question deserves investigation—not because a particular outcome must be produced.
A teacher can assess four independent qualities: Was the initial hypothesis specific and recorded before testing? Are the measurements and calculations intact? Does the conclusion fit the evidence with appropriate scope? Does the evaluation identify plausible limitations and a useful follow-up? This rewards scientific reasoning over convenient confirmation.
Useful evidence of progress includes a dated initial prediction, an unaltered observation table, transparent correction of any transcription error, a conclusion that names contradictory findings, and a next-step plan driven by limitations. A polished poster without those traces may be attractive but reveals less about whether the learner understands inquiry.
What this lesson does not claim
It does not say every unsupported prediction is therefore false, or that an inconclusive study produces a scientific breakthrough. Weak designs can produce unreliable findings; honest reporting should identify that weakness. It does not substitute for appropriate equipment, teacher safety instructions, consent or supervised statistical analysis where these are needed.
A school science report should use approved materials and age-appropriate experiments. The aircraft example here is only a numerical teaching device. Actual aircraft performance and experimental design are substantially more complex than this toy problem; we make no external claims about what paper clips generally do to flight distance.
FAQs
Do I have to change my hypothesis when my results disagree?
Keep the original prediction as the prediction you actually tested. You can form a revised hypothesis for a subsequent test, clearly labelled and dated. Do not silently replace the old one after seeing the results.
Should I throw away a measurement that contradicts the trend?
No, not on that basis alone. Investigate possible recording or procedure errors and explain any justified exclusion. A genuinely unusual result may tell you something important about variability or your method.
Does ‘no support for the hypothesis’ mean the opposite has been proven?
No. It means the available data did not support that specific prediction under the reported conditions. Sampling variation, measurement limits and confounding can prevent a decisive general conclusion.
What if the results are almost identical between conditions?
Report the closeness and the measurement resolution. With a small dataset, ‘we did not detect a clear difference’ is usually more defensible than ‘the treatments are exactly equal’. Decide whether a better-designed follow-up is worthwhile.
Will a result that disagrees with the hypothesis make the school project worthless?
No. A carefully designed test, honestly reported data, reasoned evaluation and thoughtful next experiment can be valuable learning. What matters is the quality of the evidence and the student’s explanation, not whether a prewritten answer appears on the last page.
Connect the surprise to better science
For more help keeping a prediction, an observation and an explanation distinct, read our companion Core Aim lesson on those three reasoning stages. To compare a written conclusion with its displayed evidence, see how to check a school project conclusion against its own graph. These lessons strengthen different points of the same evidence chain.
The eduKate Education learning hub and the Secondary 1 Mathematics Clementi small-group reference offer the broader teaching principle: locate the misconception, explain the mechanism, practise with support and build independence through checking.
Final thought: The point of an experiment is not to make reality agree with your prediction. It is to find out what your particular evidence can and cannot tell you. When a learner can write that with confidence—and show the unaltered numbers underneath—their scientific reasoning has taken a meaningful step forward.