Did you know? A student can write ‘I repeated the experiment nine times’ after taking nine measurements, yet may have tested only three actual objects. That difference matters. Nine thermometer readings from one cup are not nine independent cups, and nine height readings from three plants are not nine independent plants. Repeating a measurement improves information about a particular object; repeating a complete experiment across separate experimental units tells us something different about variation across objects.
To distinguish them, identify the experimental unit—the smallest independently assigned object or group that receives a condition—and count that unit separately from the measurements taken on it. A repeated measurement is another reading of an existing unit. An independent trial, in the relevant sense, involves a distinct unit or independently reset experimental run that genuinely provides a separate test of the condition. The count of readings and the count of independent units must not be silently substituted for one another.
This lesson is about honest evidence, not a fancy statistical label. It helps students design fair science investigations, write sensible conclusions, and explain to parents and teachers why ‘more measurements’ does not automatically mean ‘more independent evidence’. The plant measurements below are deliberately invented, and no real student outcomes or school experiment results are claimed.
A quick diagnosis before any experiment is performed
Put one cup of water on a table and read its temperature three times with the same thermometer. Ask: ‘How many cups have we tested? How many readings have we made?’ The answers are one cup and three readings. Now place three separate cups on the table and measure each once. We have three cups and three readings. The identical total of three numbers hides different designs.
A stronger diagnostic asks what was changed and what was controlled. If one window-side pot is compared with one shaded pot, a student may read each pot every morning for a week. There are fourteen daily readings but only two pots. Those readings can describe how each particular pot changed over time. They cannot, on their own, describe normal variation across many pots or demonstrate that the light condition alone caused the difference.
What is an experimental unit, and why does it matter?
An experimental unit is the object or collection that can receive the condition independently. For a plant experiment, it could be an individual pot when each pot is treated and placed independently. If several pots share a single sealed chamber and the only treatment is applied to the whole chamber, the chamber may be the relevant experimental unit for that chamber-level treatment. Counting the pots alone would overlook their shared exposure.
The right unit depends on how the experiment is arranged—not merely on which objects look separate. A teacher who asks ‘What could be assigned to the other condition independently?’ is helping the learner identify the unit. This protects against pseudoreplication: treating dependent readings as though they were independent experiments.
The statistical software publisher GraphPad’s explanation of replication distinguishes independent experimental units from multiple readings of the same sample. Its guidance on replicates and repeated measures also shows why measurements of the same subject at different times are not interchangeable with measurements of different subjects. The logic carries to school investigations, although formal statistical testing is not required for the examples below.
Worked classroom investigation: three pots, three readings per pot
Imagine a fictional demonstration using two sets of separately potted seedlings. Group A has three pots at location A; Group B has three at location B. Each pot contains one seedling. A pupil measures each seedling’s height on the same day three times to check ruler placement, using centimetres to the nearest 0.1 cm. No plant is uprooted or damaged. The numbers are invented to teach the distinction, not measured data.
- Group A, pot A1: 8.9, 9.0, 9.1 cm. The pot’s mean measured height is 9.0 cm.
- Group A, pot A2: 9.9, 10.0, 10.1 cm. Pot mean 10.0 cm.
- Group A, pot A3: 7.9, 8.0, 8.1 cm. Pot mean 8.0 cm.
- Group B, pot B1: 11.9, 12.0, 12.1 cm. Pot mean 12.0 cm.
- Group B, pot B2: 10.9, 11.0, 11.1 cm. Pot mean 11.0 cm.
- Group B, pot B3: 9.9, 10.0, 10.1 cm. Pot mean 10.0 cm.
There are 18 individual height readings, taken from six separate pots: three pots in each group. The nine readings from Group A do not represent nine seedlings. First average the three repeat measurements within each pot; then describe the variation between pots. For Group A, the pot-level mean is (9 + 10 + 8) ÷ 3 = 9.0 cm. For Group B, it is (12 + 11 + 10) ÷ 3 = 11.0 cm. The observed difference between these two fictional sample means is 2.0 cm.
Check the arithmetic: Group A pot means sum to 27 cm and 27 ÷ 3 = 9 cm; Group B sum to 33 cm and 33 ÷ 3 = 11 cm. Subtract 11 − 9 = 2 cm. Now check the design: the groups had three distinct plant units each, not nine. The 0.1-cm spread within a pot’s measurements describes small differences in repeated ruler readings. The larger variation among pot means (8–10 cm in A, 10–12 cm in B) describes differences among the plants as well as possible measurement and environmental influences.
Why the conclusion must be narrower than ‘location B makes plants grow faster’
The worked example reports one height measurement session, not a change in height over time. Therefore it cannot tell us which plants grew faster. To study growth rate we would need comparable earlier and later heights for the same plants and a defined elapsed period. Even a difference in final heights does not, by itself, establish that the location caused the difference: the seedlings may have started at different sizes, been watered differently, experienced different temperatures, or received different handling.
A careful school report might say: ‘In this invented example, the three seedlings in Group B had a higher mean measured height than the three in Group A by 2.0 cm on the measurement day.‘ It might add that small groups, environmental differences and unmeasured starting heights limit inference. That sentence is both more useful and more defensible than ‘Group B always grows plants better’.
What goes wrong when repeated readings are counted as independent trials
Suppose a report states ‘nine plants were tested in each condition’ because it contains nine measurements per group. A reader could wrongly believe the class observed 18 different plants. This exaggerates the amount of independent biological variation sampled. Repeating a ruler measurement on one plant may make that plant’s estimated height steadier, but it does not reveal how a new plant would respond.
Another frequent problem occurs when unequal numbers of readings are pooled without accounting for the object. Imagine one plant was measured once and stood 8 cm, while a second was measured four times and each reading was 10 cm. Averaging the five readings gives (8 + 10 + 10 + 10 + 10) ÷ 5 = 9.6 cm, heavily weighting the second plant. If the question is the mean of the two plant-level heights, the unweighted mean is (8 + 10) ÷ 2 = 9.0 cm. Both arithmetic operations are possible, but they answer different questions. Name the unit before choosing the denominator.
A fairer plan for a school-level investigation
- State the question precisely: for example, ‘How does a defined difference in light exposure affect change in plant height over 14 days under these conditions?’
- Identify independent units: separately potted plants assigned independently to conditions, provided the treatment is applied at the pot level.
- Record baseline values: measure each plant before comparing later heights; do not assume identical starting size.
- Specify what stays consistent: plant type, pot size, soil, watering procedure, measurement method and intended observation times as feasible.
- Reduce shared confounders: consider whether location also changes temperature, airflow or watering convenience.
- Keep two counts: the number of pots and the number of readings per pot; never label the latter as the former.
- Preserve raw data: make one row per pot and columns for readings, dates, conditions and any problems.
In real settings, plant experiments can be more complicated than they appear. Conditions such as sunlight are not easily isolated from the rest of a windowsill environment. A careful student describes the limits of the apparatus rather than claiming laboratory-level control. Safety also matters: use ordinary classroom-safe plants, avoid chemicals, wash hands after handling soil, and follow a teacher’s instructions.
Three stages of practice, with increasingly independent thinking
Stage 1 — sort the counts
A class takes five measurements of the same paper bridge before adding any weight. Ask: how many bridge specimens and how many measurements? Answer: one specimen, five measurements. If the bridge was permanently bent between readings, ask whether its state has changed; the later readings may no longer be simple repeats of the same unchanged condition. The goal is to say what physically happened.
Stage 2 — calculate first by unit
Three separately prepared samples each receive two readings. Sample X: 18 and 20 units; sample Y: 21 and 23; sample Z: 17 and 19. The unit-level means are 19, 22 and 18 units. Their average is (19 + 22 + 18) ÷ 3 = 59/3 ≈ 19.67 units. The report should say three samples, six readings, not six independent samples.
Stage 3 — diagnose a flawed explanation
An independent exercise says: ‘Four seedlings were each measured at 9 a.m., 10 a.m. and 11 a.m. The experiment had twelve independent seedlings.’ Ask the student to correct it. A sound answer states four seedlings and twelve observations, with repeated measurements of the same four seedlings at three times. If temperature and light change over those hours, the readings may reflect genuine within-plant change as well as measurement variation. They should not be treated automatically as three interchangeable technical readings.
How students can display the result without hiding dependence
A table should identify each pot or sample in a separate row and each repeat reading in separate columns. A simple dot plot of the six pot-level means can reveal variation across pots; a second display of the measurement spread within each pot can reveal repeatability. A chart that plots all 18 readings as though they came from 18 unrelated plants conceals the study’s structure.
When a pupil is more advanced, introduce the idea that repeated observations can be correlated. Measurements from one plant often resemble one another more than measurements from unrelated plants. Formal analysis may use paired or hierarchical methods, but the foundation is understandable much earlier: remember which readings came from the same thing.
Parent and teacher checkpoint: reward explanation, not a larger sample number
Instead of merely asking ‘How many trials?’, ask three questions: ‘What object got the treatment?’, ‘How many different objects were involved?’, and ‘How many times was each one measured?’ A learner who answers all three and gives a cautious conclusion has made observable progress. At the next lesson, use a wholly different setting—a class experiment with paper spinners, soil samples or cooling cups—and check whether the learner can identify the experimental unit again.
Teachers can mark lab records on two dimensions: accurate recording of observations and accurate description of independent replication. A child may keep an excellent table but still overstate independence. The correction should target that reasoning gap. Families can rehearse the same skill by distinguishing three attempts on one quiz from tests completed by three different pupils; the two designs reveal different sorts of information.
Frequently asked questions
Are repeated measurements useless?
No. They help identify measurement variability, mistakes and changes within an object over time. They simply do not create new independent objects. Good experiments often need both repeated readings and well-chosen independent units.
If I use three different thermometers on one cup, have I made three trials?
You have three instrument readings of one cup, potentially useful for comparing instruments. You have not tested three independent cups. If the research question concerns thermometer agreement rather than variation among cups, the relevant unit and analysis will be different. Begin with the actual question.
Can I average every measurement together?
Sometimes a pooled average is mathematically computable, but it may weight some experimental units too heavily or conceal dependence. When samples have unequal repeat counts, calculate and inspect unit-level summaries, and describe any chosen weighting rule explicitly.
Does three pots per group prove a causal effect?
No. It may be a feasible classroom demonstration, not strong general evidence of causality. Random assignment, genuine independent treatment, sensible control of other influences and adequate replication all matter. Interpret the evidence within the limits of the design.
The core aim: report exactly what was really tested
A useful companion is the Core Aim guide to fair paper-bridge load-position tests, alongside the Core Aim lesson on cooling rates with unequal observation times. Both encourage attention to what actually changed. The eduKate Learning System and immutable Clementi teaching reference model diagnosis first, clear reasoning, progressive practice and independent checking. The next time a student sees ’18 readings’, the real question is not how impressive that number sounds. It is: eighteen readings of what?