Did you know that calculating an average can hide the very measurement problem a science experiment needs to reveal? Three readings might look reassuring once they become one neat number. But if one reading is wildly different, the average is not a substitute for asking why.
Before averaging repeated science measurements, keep the original readings, check that they refer to the same quantity under comparable conditions, inspect their spread, investigate unusual values and only then decide what summary is justified. Repeats can help reveal variability; they do not automatically make a measurement correct or prove a conclusion.
The diagnosis: what exactly was repeated?
A learner may write “Trial 1, Trial 2, Trial 3” without understanding what a trial means. Were three readings taken of the same stationary object using the same ruler? Was a whole experiment repeated with a fresh sample each time? Or did the system itself change as time passed? Those are different questions. An average across incomparable conditions can produce a precise-looking but meaningless result.
Ask the student to point to the instrument, object or sample, measurement unit, timing and conditions. Then ask: “If we did the experiment again, which things would we deliberately keep the same?” If they cannot explain the setup, re-teach the measurement procedure before practising more arithmetic.
Worked example: one rod, three lengths, two very different stories
Here is an entirely invented classroom example, not a real student experiment. A pupil repeatedly measures the same rigid rod with a ruler. The first set of recorded lengths is 12.1 cm, 12.0 cm and 12.2 cm. The mean is (12.1 + 12.0 + 12.2) ÷ 3 = 36.3 ÷ 3 = 12.1 cm. The range is 12.2 − 12.0 = 0.2 cm. The small spread is consistent with readings being fairly close to one another, though we still do not know whether the ruler was positioned correctly.
A second fictional set for what is claimed to be the same rod reads 12.1 cm, 12.0 cm and 15.8 cm. The arithmetic mean is (12.1 + 12.0 + 15.8) ÷ 3 = 39.9 ÷ 3 = 13.3 cm. The range is 15.8 − 12.0 = 3.8 cm. The large spread raises a question. Did somebody read the wrong scale, measure a different object, make a transcription error or encounter another genuine issue? We cannot know from the three numbers alone.
It would be tempting to delete 15.8, average the other two and report 12.05 cm. That is not justified merely because 15.8 looks inconvenient. Preserve it, check the record and repeat the measurement if feasible and safe. If a documented copying error is found, record the correction and why. If the cause is unknown, report the anomalous reading and uncertainty honestly rather than inventing a reason.
Repeatability is not the same as accuracy
Readings close together show consistency under the tested conditions; they may still be consistently wrong. A ruler starting at a worn edge or a constant zero offset could produce closely grouped measurements that miss the true length. Conversely, a single surprising value can reflect a real change in the object or method, not laziness.
The US National Institute of Standards and Technology describes repeatability as agreement among measurements made under the same specified conditions. Its technical treatment distinguishes repeatability from reproducibility under changed conditions. For schoolwork, the immediate lesson is simpler: explain what stayed the same, keep all recorded observations and never equate a small spread with guaranteed truth.
A safe order of operations for student scientists
- Record: write every raw reading with its unit, trial number and any necessary time or condition.
- Verify: make sure the readings concern the same intended quantity; check the instrument, scale, labels and transcription.
- Compare: look at the maximum, minimum, range and pattern before computing a summary.
- Investigate: circle a surprising value and ask what could have caused it; do not silently erase it.
- Summarise: calculate a mean only when a mean answers the question, showing what readings it includes.
- Explain: state what the data support and what the limited trials cannot establish.
In a basic classroom measurement, the range is a useful first look at spread. It is not a formal uncertainty estimate or a universal rule for removing outliers. Advanced laboratory analysis may use standard deviations and calibrated uncertainty methods. At school level, correct records and proportionate claims are the priority.
A second check: do not average away a trend
Suppose water cools over time and readings are 70 °C, 65 °C and 60 °C at successive minutes. Their mean is 65 °C, but that single number hides the important downward pattern. Those readings are a time series, not interchangeable repeats of one unchanging condition. A table and a graph may answer the question better. Planning what to record first makes such differences visible; see how to design a results table.
What usually goes wrong
Averaging mismatched measurements
Students sometimes combine centimetres with millimetres, different instruments or readings taken at different temperatures without documenting the differences. Standardise compatible units where appropriate and keep the conditions visible; do not pretend changing conditions were identical.
Throwing away the odd value
A repeated experiment that disagrees with expectations is not automatically a failure. A strange result may reveal a procedural mistake or a real phenomenon. Check it, repeat if justified and report it transparently. Deleting it solely to improve the graph is poor scientific practice.
Mistaking a precise display for a precise experiment
A digital device may display many decimals, but that does not mean all those digits are reliable. Measurement resolution, calibration, method and variation matter. Read our related accuracy versus precision guide.
How parents and teachers can observe genuine progress
Ask the learner to explain why a reading belongs to the same trial set and what they would do if a measurement changed unexpectedly. Teachers can provide two short invented data sets and ask students to decide which can be sensibly averaged and which needs a further check. The assessment should reward reasoning and transparent records, not only a correct mean.
Progress becomes visible when the student labels units without prompts, keeps raw data, checks a surprising entry before calculating and explains why a conclusion is cautious. An independent transfer task could involve timing the same event several times: can the pupil distinguish measurement error from genuine variation between events, and say what their mean represents?
Limits of this classroom method
Three repeats do not prove reliability. The needed number of measurements and suitable statistical method depend on the process, purpose and acceptable uncertainty. Some investigations cannot safely be repeated at all. Never rerun a hazardous experiment without the teacher’s approval and proper safety controls. A school worksheet may also prescribe exactly how a class should calculate and report data; follow those instructions while learning the underlying idea.
Frequently asked questions
Should students always take the average of three results?
No. Check whether the measurements are comparable, whether a trend is the real finding and whether the exercise actually asks for a mean. There is no universal three-reading rule.
Is a result wrong because it is far from the others?
Not necessarily. It deserves investigation. An unusual result may arise from an error, a changed condition or genuine variation. Preserve it unless there is a documented, defensible reason to correct or exclude it.
Can consistent readings still be inaccurate?
Yes. A constant measurement bias can make repeated readings agree while all are displaced from the true value. Consistency alone does not establish accuracy.
The core aim: learn to respect the evidence before reducing it
An average is useful only when the student knows what was averaged and why. Diagnose the setup, model a careful comparison, practise on two new data sets and then ask the learner to check an unfamiliar table alone. The same explanation-first progression appears in eduKateSG’s Clementi small-group teaching benchmark and the wider Education library.