Three Primary 5 pupils in Clementi are measuring the temperature of water as it cools. They repeat the measurement five times and write five very similar numbers. Their table looks reassuring. The values agree, their lines are neat and the group’s average is easy to calculate. Then the teacher notices something odd: every reading was taken with the same thermometer, which had not been checked against a reference. The pupils have shown that they can repeat a procedure consistently. They have not yet shown that their measurements are accurate.
Why can an experiment give the same result repeatedly and still be wrong? Repetition can reduce the influence of random fluctuations and can reveal whether an observation is stable under repeated conditions. But if every trial shares the same flawed instrument, misunderstood instruction, unrecognised confounder or measurement habit, the same error can appear each time. Consistency is evidence of repeatability; it does not by itself establish accuracy, a fair test or a causal explanation.
This is a guide for students learning practical Science, parents reading experiment write-ups and teachers designing inquiry tasks. The examples are hypothetical classroom situations, not reports about any particular Singapore school. The purpose is not to make children distrust experiments. It is to teach what repeated observations genuinely establish, what they leave uncertain and which next test can move a claim closer to reliability and validity.
The answer in 50 seconds
Repeated trials are useful because measurements contain random variation. Averaging suitable repeated observations can reduce some random noise. But repeated trials do not automatically remove systematic error—a bias that affects many readings in the same direction—or a shared design flaw. If a thermometer reads 2 °C too high on every attempt, five readings can agree perfectly and still be displaced. If two plant groups differ in both light and water, repeating the comparison does not isolate which factor caused the difference.
Ask four questions. Did the procedure stay consistent? Was the measurement tool checked? Did the comparison hold other relevant factors reasonably constant? What independent check could expose a shared mistake? Record actual data rather than forcing agreement. Use repeated trials alongside sensible calibration, comparison groups, clear operational definitions and appropriate scientific reasoning. A neat average is the start of interpretation, not a stamp saying “proved.”
A surprising result: five matching temperatures
Imagine a thermometer with a consistent positive offset of 2 °C over the range used in a simple school demonstration. The water’s actual temperature at a specific moment is 30 °C, but the device displays 32 °C. A group carefully reads 32, 32, 32, 32 and 32 °C in five repeated checks made under essentially unchanged conditions. Their mean is 32 °C. Their spread is zero. Within that setup, the measurements are precise and repeatable; relative to the true temperature, they are biased.
The example uses a deliberately simple offset. Real instruments can show more complicated errors that vary across a range, drift with time or depend on how they are used. A school lesson does not need to teach every metrology term before making the distinction. It can ask pupils to compare what their numbers say with what an independent check says. Agreement among readings is not identical to agreement with reality.
Suppose another checked instrument reads approximately 30 °C under matched conditions. The disagreement provides a clue, not yet a conviction. Perhaps the instruments differ in response time or placement, or the water is not equally mixed. The next task is to test those alternatives. A good Science lesson makes space for a puzzling disagreement and asks pupils to improve the procedure rather than automatically discarding inconvenient data.
Three ideas that should not be compressed into one
Repeatability concerns obtaining comparable results when a specified procedure is carried out repeatedly under specified conditions. Accuracy concerns closeness to the value or condition one intends to measure. Validity concerns whether the method and evidence actually support the particular question or inference. The terms connect, but a high score on one does not confer a high score on the others.
A tightly controlled classroom method could produce remarkably similar numbers while measuring the wrong thing. A parent may notice that the child’s graph is beautifully straight but does not represent the variable named in the conclusion. A teacher may see that the task measured how quickly paper towels soaked up spilled water, even though the question was framed as a broad claim about “the best absorbent material.” Operational definitions and claims must stay aligned.
Scientific work also distinguishes reproducibility across teams or conditions, though terminology is not completely uniform across fields and publications. The useful primary-school question is simpler: can somebody else understand what you did, check the measurement and judge whether the conclusion follows? If not, repeating the same private routine is a thin form of evidence. Clarity makes results open to challenge and improvement.
How random error behaves differently from a shared bias
Suppose a measurement varies a little because of small timing differences. Values such as 29.8, 30.1, 30.0, 29.9 and 30.2 °C might appear under controlled conditions. If those variations are approximately independent, are centred appropriately and do not conceal a systematic problem, repeated measurements and a mean may give a more stable estimate than a single reading. Those assumptions matter. Five numbers do not guarantee that the assumptions hold.
Now imagine each of those readings is also shifted upward by 2 °C by the same instrument. The observed values become 31.8, 32.1, 32.0, 31.9 and 32.2 °C. Their mean settles near 32 °C as the number of repeats grows. More trials can make that biased estimate look increasingly stable. In statistical language, repeated sampling can reduce the standard error of an estimate while leaving a systematic offset. A child need not memorise that sentence; the table makes the lesson visible.
There is also a mixed case. A pupil may consistently read the thermometer from above instead of at eye level, adding a systematic viewing error, while taking each reading at slightly different times introduces random variation. Repeating the same approach can narrow some variation while preserving the viewing bias. Asking the pupil to change position and compare readings may be more informative than asking for ten additional identical trials.
What does calibration actually do?
Calibration means comparing an instrument’s indications with an appropriate reference under defined conditions, so its measurement behaviour can be understood. It is not simply “press reset” or “make the number look right.” At school, a teacher might show two thermometers in the same well-mixed water bath and ask why their displayed values differ. In a more formal setting, reference quality, calibration range, uncertainty and traceability matter. Not every classroom has access to formal calibration equipment, and it should not pretend to.
The learning point is to make a reference check proportional to the task. A comparison instrument may reveal a likely offset but may itself have limitations. A ruler printed from a document can have incorrect scaling if the printer resizes the page. A weighing scale not zeroed before an experiment can add an offset to every mass. A timer that a student starts late can shorten every recorded interval. Repeating these measurements using the same flawed starting condition does not remove the problem.
In the classroom, a useful calibration conversation starts with the quantity and unit. “What exactly are we measuring? How will the instrument display it? What does zero mean? What reference could reveal a constant error? How large could the error be relative to the effect we are trying to observe?” These questions teach pupils to see measurement as a relationship between a physical event, an instrument and an interpretation.
The confounded plant experiment
Consider two groups of bean plants. Group A sits beside a bright window and receives 40 mL of water each day. Group B stays further from the window and receives 20 mL. After two weeks, Group A is taller. The children conclude that stronger light made the plants grow taller. They repeat the setup with new seeds three times, using the same light placements and water amounts, and observe a similar pattern.
The repeats make the pattern more credible under those bundled conditions. They do not isolate light as the cause because water also differed. The two factors moved together. Repetition has not removed the confounding variable. A fairer comparison would keep watering conditions suitably similar while changing the light condition, or design a more careful multi-factor experiment where both influences can be examined. Real growth also depends on soil, temperature, seed variation and other factors; school-level controls are approximations rather than total elimination of complexity.
Another mistake is to say the experiment “proved plants grow taller with more light” without distinguishing plant height from overall health. Under some conditions, plants in lower light may stretch taller while remaining weaker; the chosen outcome matters. An experiment about height is not automatically an experiment about biomass, growth rate, photosynthesis or vitality. The dependent variable must match the claim. This is why scientific validity is about the design and interpretation, not simply how many times a table was filled.
One child, two clocks and a rolling ball
A Secondary 1 group releases a small ball down a ramp and records how long it takes to reach a mark. They consistently start the stopwatch when they notice the ball moving, about a fraction of a second after release. If they use the same reaction pattern every time, their times may cluster well. They have repeated the timing method, but it may not measure the intended interval from the instant of release.
A teacher can turn the problem into a productive comparison. Let the group write a precise operational definition: start time equals the moment the ball is released; finish time equals the moment it crosses the mark. Then try a method less dependent on individual reaction time, such as appropriate video-frame review where authorised and feasible, or a different person observing. Ask whether all trials were done from the same release point and whether the ramp angle changed. A second method introduces different possible errors but can reveal a shared bias in the first.
Safety and school equipment constraints matter. The improvement is not an invitation for unsupervised speed experiments or unsafe ramps. It is a model of scientific reasoning. If two methods disagree, do not automatically declare the newer one correct. Identify what each measures, what each may miss and which independent evidence best addresses the question. That habit scales from classroom investigations to advanced research.
Why a high correlation between repeated results is not enough
Students sometimes hear that repeated similar values make a conclusion more scientific. Similarity matters, but ask similarity of what. Ten devices built to the same incorrect specification may produce similar errors. Ten observers taught the same mistaken coding rule may agree perfectly. Ten worksheets copied from one incorrect answer key may show striking consistency. Shared origins can create agreement without independent confirmation.
The relevant distinction is not “repeat bad, variation good.” Random variation can make it difficult to see a meaningful pattern. Consistent methods are essential to learning how a system behaves. The point is that independence of error sources affects what agreement can establish. Replicating a procedure while changing one potential source of bias—for example, using a different checked instrument or independent observer—can test a deeper claim than repeating the same instrument unchanged.
An educational analogy is a child checking Mathematics with a calculator after typing the wrong expression. Repeating the same entry yields the same number. The machine’s consistency does not validate the student’s transcription. The science lab version is similar: compare the representation and procedure with the intended question, not only the consistency of the output.
The hidden assumption behind “take the average”
Averaging is useful when the observations represent repeated measures of the same intended quantity and the major uncertainties are appropriately understood. It can be misleading when pupils average measurements of different conditions, when the underlying process is changing rapidly, or when a few outliers reflect an important event rather than harmless noise. In a cooling-water experiment, the water temperature is meant to change over time. Taking five readings at successive minutes and averaging them does not estimate “the temperature at minute three” unless a specific model justifies it.
Averages can also hide drift. If a sensor warms up during use and reads 30.0, 30.5, 31.0, 31.5 and 32.0 under a stable reference, the rising sequence matters. The mean of 31.0 would erase that diagnostic pattern. Plot readings against time before deciding whether averaging is appropriate. A scientific summary should preserve the feature that answers the question rather than flatten it away.
In a simple school task, pupils can write both the readings and the average, then explain why averaging is reasonable or not. A teacher can reward a student for noticing a pattern that complicates the expected conclusion. This is not lowering standards. It teaches that calculation follows interpretation of the measurement process. The mathematics of a mean cannot rescue a poorly defined quantity.
Four checks before saying the experiment worked
Check one: the question. Can the group state what it is trying to find out in one sentence without vague phrases such as “best” or “more effective” unless those are defined? Does the stated outcome match the observations? A test about water temperature does not automatically prove anything about total energy transfer without a justified model.
Check two: the method. What changed intentionally, what was observed and what else might reasonably affect the result? Are there potential confounders that travelled together across every repetition? Could the instructions themselves lead everybody toward the same error? The question of control is not a ritual box to tick; it identifies what alternative explanations remain.
Check three: the instrument. Is the measuring scale, zero point, unit, resolution and placement appropriate? If the observed effect is one tenth of a degree but the available tool has coarse markings and large uncertainty, the tool may not distinguish the effect. More repeats cannot automatically create information the instrument did not resolve.
Check four: the independent challenge. What would a different method, observer, reference or comparison group show? What observation would make the group revise its conclusion? A claim that cannot survive a reasonable challenge is too fragile to present as settled. In primary school, “What else might explain this?” is a powerful scientific question even when the children cannot run every possible follow-up.
Teacher route: make the disagreement teachable
A teacher can set up a short comparison with two temperature instruments whose readings differ modestly, explaining that they are using hypothetical or safe demonstration conditions. Ask students to record the values before interpreting the disagreement. Some pupils may immediately choose the higher or more familiar instrument as right. Instead, ask what independent check could help. The learning target is to separate repeatability from accuracy and to articulate what additional evidence is needed.
Avoid manufacturing fake data in a way that suggests pupils should tidy their own results to match the teacher’s expected pattern. If the teacher uses simulated data, label it clearly. When real observations differ from the anticipated result, discuss the limits of the setup and record the discrepancy. Classroom inquiry becomes more honest when unexpected evidence is treated as information rather than embarrassment.
A useful marking rubric can reward four features: accurate record of observations, clear procedure, recognition of at least one limitation and a conclusion proportional to evidence. This does not require sophisticated statistics from younger learners. “Our readings were similar, but we used only one ruler and did not check its scale” is better reasoning than “We repeated three times, so our answer is definitely correct.” The first student knows what their work can and cannot support.
Student route: how to write a conclusion without pretending certainty
Start with the observation, not the wish. “The five readings ranged from 31.8 °C to 32.2 °C” describes what was recorded. Then identify the measurement method and the comparison: “All readings came from the same thermometer in the same water sample.” Next make a limited claim: “The measurements were consistent within that setup.” Finally name an appropriate limitation: “We did not check the thermometer against another reference, so a common offset is possible.”
A school answer may need simpler language depending on level. A Primary 4 student might write, “Our results were similar, but we used the same instrument each time; we could compare with another thermometer.” The principle is the same. Honesty about uncertainty is compatible with confidence in what was actually observed.
Do not add vague boilerplate such as “human error may occur” to every conclusion. A useful limitation says which error, how it could change the result and what check might help. “We may have started the stopwatch late, which would make the measured travel time shorter; repeating with a different timing method could test this” is concrete enough to improve the next attempt. Science advances when limitations create actionable questions.
Parent route: ask a question the child can investigate
When a child proudly shows five matching results, acknowledge the careful work. Then ask, “What did repeating it help you find out?” The child might say it showed the measurement was similar each time. Follow with, “If the ruler was printed too small, would five measurements fix that?” This opens the distinction without attacking the child’s conclusion. A simple everyday analogy—a kitchen scale that starts at a nonzero number—can make the mechanism visible.
Parents should not turn every home Science assignment into a high-stakes technical audit. The educational aim is a transferable habit: compare an output with the way it was produced. If the child can name a possible shared source of error and suggest one fair check, the discussion has succeeded. Avoid inventing a “correct” experimental answer merely to make the worksheet neat. If the equipment is uncertain, say so and ask what could reasonably be checked in a school setting.
A parent’s question can also model respect for evidence: “Would your conclusion change if a second checked instrument disagreed?” The child learns that revising a claim is not failure. A teacher can then build on this willingness in future lessons about graphs, inference and scientific explanation. A culture that rewards correction is more useful than one that rewards flawless-looking tables.
When more repeats are exactly the right response
This article is not a warning against replication. If a process has meaningful random variation, additional well-designed observations can help estimate a pattern more precisely. If a result appears only once under poorly controlled conditions, a repeat may expose that fragility. If two groups use the same described method independently and achieve similar results, confidence can increase—provided the sources of bias are considered. Repetition is central to reliable Science because claims must face more than one opportunity to fail.
The decision depends on the dominant uncertainty. When the issue is random fluctuation, more suitable trials may be valuable. When the issue is a systematic thermometer offset, a reference comparison may add more information than another ten identical readings. When the issue is a confounder, redesigning the comparison matters more than increasing repetitions. When the issue is an ambiguous definition of “growth,” clarifying the outcome matters first. This is how a learner begins to match the test to the question.
It is possible for several problems to coexist. Real experiments need both enough observations and a credible design, as well as appropriately measured quantities. A classroom has limited time and equipment, so teachers should prioritise the check that most reduces uncertainty relevant to the learning goal. Good inquiry is not the largest possible number of trials. It is an efficient sequence of observations and challenges that makes the claim more defensible.
Three decision cards for classroom investigations
Card A: identical measurements, uncertain instrument. A group records mass as 25 g five times, using a scale that may not have been zeroed. They should not claim that repetition proves the true mass is 25 g. The next step is to check zero, assess the scale and compare a suitable known reference. Their original results remain useful as evidence of consistency in that setup.
Card B: different measurements, controlled instrument. A ball’s travel time varies between attempts because release timing is difficult. The group should examine whether the measurement protocol can be made more consistent and whether the variation is random or connected to a factor such as release position. Additional trials may help after defining what counts as the same condition. Do not average across clearly different ramp positions without justification.
Card C: repeated pattern, two variables changing. Plants near the window grow differently from plants in the corner, but the watering schedule differs too. Repeatability supports an observation about the combined conditions. To test light specifically, the group needs a better comparison. A clear limitation and a safer, more controlled plan show scientific understanding even if the experiment cannot be rerun during the lesson.
The most tempting errors in student write-ups
“We did it three times, therefore it is true.” Three repetitions can be a sensible classroom minimum for some practicals, but there is no universal number that converts a flawed design into proof. What matters is the question, variation and potential bias. Use language such as “consistent with” when it better reflects the evidence.
“Our readings were close, so there was no error.” Closeness addresses some forms of variation. It does not rule out a common instrument offset or confounder. Accuracy and repeatability should be judged using relevant checks, not treated as synonyms.
“The answer must match the textbook diagram.” Diagrams are simplified models of particular conditions. Real materials and school equipment can behave differently. When observations and a model disagree, inspect both the procedure and the assumptions before rewriting the results to fit expectations.
“Anything unusual is human error.” This phrase often conceals the mechanism. A thermometer at an angle, a stopwatch started late and a wrongly labelled axis are specific, testable issues. Naming them improves learning; generic blame does not.
“One accurate-looking graph settles the cause.” A graph can show a relationship without isolating a mechanism. A shared third factor may explain the pattern. A scientifically justified cause requires appropriate design, evidence and reasoning beyond visual neatness.
“An independent check is a threat to my result.” The opposite is more productive. If a well-chosen check agrees, the claim is better supported. If it disagrees, the discrepancy identifies work worth doing. Teaching students to welcome a fair challenge helps them keep conclusions proportionate to evidence.
Evidence and caveats: what current guidance can and cannot establish
The Singapore Ministry of Education Primary Science syllabus emphasises inquiry practices and evidence-based reasoning appropriate to pupils’ development. It supplies curriculum context, not a guarantee that one simple experiment can establish a universal scientific rule. Science education resources on cause and effect distinguish seeing a pattern from developing and testing causal explanations. An independent check is part of disciplined inquiry because repeated observations can share limitations.
Research-methods guidance from the Australian Education Research Organisation asks educators to scrutinise design, measures and the strength of evidence rather than accepting a result at face value. Those ideas are transferable to evaluating classroom demonstrations, though school experiments are not clinical trials or large research programmes. We should not claim that one parent question, a second thermometer or a revised graph will by itself improve examination scores. The immediate goal is a sounder inference from an observable procedure.
Precision and accuracy are sometimes used casually in school settings. Here we use repeatability to describe consistency under defined conditions and accuracy to discuss agreement with a suitable reference or intended true quantity, while acknowledging uncertainty in any reference. Causal validity is a separate question: whether the design isolates the proposed influence. Experts may define measurement terms more formally for particular disciplines. The classroom definitions are teaching tools, not substitutes for full standards of metrology.
Frequently asked questions
If I repeat an experiment five times, does that make the result accurate?
Not automatically. Five readings can reduce concern about random variation when the method is suitable, but all five may share a biased instrument or design flaw. Check whether measurements represent the intended quantity and whether an independent reference or improved comparison could expose common error. Repeating is valuable evidence about consistency; accuracy requires more than the count of attempts.
What is the difference between random error and systematic error?
Random error produces unpredictable variation between observations under otherwise comparable conditions. Systematic error shifts results in a recurring way because of the instrument, procedure or another persistent feature. A mean of well-designed repeats can reduce some random variation but generally does not eliminate a shared offset. Real classroom errors can include both kinds at once, so inspect the pattern and the method.
Can an experiment be reliable but not valid?
Yes. An experimental procedure may consistently produce the same measurements while failing to address the intended question. For example, repeated plant comparisons that change both light and water do not isolate light as the cause. The observations can be stable under the bundled conditions yet be inadequate evidence for a single-factor conclusion. Reliability and validity answer different questions.
Does a larger number of trials always make the conclusion stronger?
More appropriate trials can help estimate random variability, but trials cannot fix a poorly defined outcome or a confounded design by themselves. Ten repetitions of the same flawed setup may make an inaccurate estimate look impressively stable. Before adding trials, ask what uncertainty needs reducing and choose a test designed to address that uncertainty.
Why do two thermometers disagree when placed in the same water?
Possible reasons include calibration differences, different response times, placement, water mixing or the resolution of the displays. Their disagreement does not identify the correct instrument without further checking. Examine the setup and compare against an appropriate reference if available. Keep the original readings visible rather than silently altering them to match the expected answer.
Is averaging wrong in Science?
No. Averages are useful when repeated observations refer to a comparable quantity and the dominant sources of uncertainty are understood. Averaging can be misleading if the measured quantity changes systematically over time or if a shared bias remains. Inspect the raw readings and graph any meaningful sequence before deciding what a mean represents.
How can a Primary 5 pupil explain systematic bias without advanced statistics?
Use a familiar instrument example. “If the scale is already showing 2 g before we put the object on it, repeating the measurement will still include that extra 2 g.” Then let the pupil suggest checking zero or comparing with another appropriate instrument. Understanding the shared mistake matters more than mastering terminology at this stage.
Does a fair test remove every other variable?
A fair test aims to control relevant influences sufficiently for the question being asked, but real conditions can never be perfectly identical. Students should identify major alternative explanations and try to keep them from changing with the tested factor. A school practical can be informative without claiming impossible total control. A conclusion should say what conditions were actually compared.
What if the results disagree with the teacher’s expected answer?
Record them accurately and investigate. Check units, procedure, instrument, graph and assumptions. Ask whether the discrepancy comes from the setup or from a model that applies only under particular conditions. Fabricating tidy results hides the very evidence that helps pupils learn. Teachers can use real disagreement to discuss uncertainty, measurement and responsible inference.
How should a conclusion change after discovering a shared mistake?
Do not erase the history. State what was observed, describe the discovered limitation and explain how it affects the original claim. If suitable, propose or perform a revised test. A conclusion such as “Our readings were consistent, but the thermometer had not been checked, so we cannot claim the exact temperature was accurate” is more scientifically responsible than pretending the issue never occurred.
Sources and canonical neighbours
- Singapore Ministry of Education, Primary Science Syllabus — curriculum context for inquiry and science learning.
- Next Generation Science Standards, Cause and Effect in Earth and Human Activity — a science-education example of analysing evidence and testing effects.
- Australian Education Research Organisation, Assessing research evidence — guidance on the quality and interpretation of evidence, updated June 2026.
- What Works Clearinghouse, Using Student Achievement Data to Support Instructional Decision Making — an education decision-making perspective; not experimental proof of a particular thermometer exercise.
Within eduKateSG, read the related observation-window explanation of why ‘nothing happened’ can still be a Science result, how research variables work, and the broader what causality means. Those pages own observation timing, variable design and general causal thinking. This article owns one exact question: if every attempt agrees, could every attempt still be wrong for the same reason? A good answer is yes—and the next scientific move is to find a check that the shared mistake would not survive.