Did you know that a control sample can change during a perfectly sensible experiment? A pupil might look at an uncovered dish losing water and exclaim, ‘Our control is broken because it didn’t stay the same!’ Another pupil might make the opposite mistake: accept every difference as an effect of the experimental treatment, even after the control was moved next to a fan. Both misunderstand what a control is supposed to do.
A control condition is a reference setup used to show what happens without the particular treatment being investigated. Its measured outcome may naturally change over time; what must be kept comparable are the important controlled variables, such as timing, initial amounts, room conditions and measuring method. If a new outside influence affects one condition differently, record it and question the causal conclusion. A changing control result is not automatically a failed test, but an uncontrolled change in its circumstances may be.
The practical diagnostic: what changed, and what was supposed to remain controlled?
Ask the learner to distinguish three things before looking at data. The independent variable is what the experimenter deliberately varies, such as adding a loose cover. The dependent variable is what they measure, such as water remaining after a fixed interval. The controlled variables are important conditions kept comparable between groups, such as equal starting water mass, identical dishes, the same room and the same weighing times. A control condition is one of the groups, usually the one without the treatment.
These names sound similar but do different jobs. A control condition is not a magical object that must never change. An uncovered water dish naturally loses water even if no special treatment is applied. That background change is exactly why a reference condition can be useful. The reason for comparing groups is to distinguish an ordinary change that may happen anyway from an additional difference associated with a treatment.
An invented classroom case with hand-checked measurements
Imagine a safe, illustrative classroom investigation using two identical shallow dishes at room temperature. Both begin with 50 g of water, measured after subtracting each empty dish’s mass. Dish A is uncovered: the control condition. Dish B has a loose cover: the treatment condition. Both are monitored for the same two hours, with the cover removed only as needed for a consistent weighing procedure. These numbers are designed for teaching, not reported observations of actual pupils.
- At time zero, dish A contains 50 g of water; at two hours, it contains 43 g. The loss is 50 − 43 = 7 g.
- At time zero, dish B contains 50 g of water; at two hours, it contains 47 g. The loss is 50 − 47 = 3 g.
- The observed difference in water lost is 7 − 3 = 4 g, with the covered dish losing 4 g less in this illustrative observation.
- Reverse-check the masses: 50 − 7 = 43 g and 50 − 3 = 47 g. Both remain non-negative and below their 50 g starting amount.
If the only relevant difference had been the intended cover, this result would be consistent with the cover reducing water loss over those two hours. It would not, from two single dishes alone, establish how large the effect is generally or prove what would happen in every room. Notice that the control lost 7 g. That does not invalidate it: the normal evaporation of water from an open dish is the comparison we wanted.
Now discover the hidden change
Halfway through the imaginary experiment, someone moves dish A nearer a running fan while dish B stays on the original table. This is a crucial new fact. Air movement can affect evaporation, so the dishes now differ in both cover and airflow. A 4 g difference cannot safely be attributed to the cover alone. A student who merely writes ‘the cover prevented 4 g of evaporation’ has converted an observed association into a causal claim the test no longer isolates.
The numerical calculation remains correct. Nothing about 50 − 43 or 50 − 47 changed. The interpretation did. Good science requires checking the conditions under which numbers were obtained, not just recomputing them. A result can be accurate as a measurement yet insufficient as evidence for the intended explanation.
Should we subtract the control change to fix the problem?
Not automatically. Subtracting 7 g from 3 g or reporting the 4 g gap does not ‘correct’ the unknown fan effect. We do not know how much of dish A’s 7 g loss was caused by its new location. An invented mathematical adjustment cannot recover the missing counterfactual—the mass dish A would have had if it had stayed in the same airflow as dish B. If the accident is known, document exactly when it occurred, retain the observations truthfully and describe the causal conclusion as compromised.
In more sophisticated research, models can sometimes account for measured nuisance factors, but those methods need assumptions and evidence. In a simple school practical, the cleanest repair is usually a properly documented new trial under consistent conditions, not invented corrected masses. The Core Aim article on comparing experiments with unequal durations shows a different case where choosing a suitable calculation is legitimate; uncontrolled airflow is not merely a units problem.
How a careful repeat could look
Suppose a second entirely hypothetical trial is performed with matched dish locations and a procedure that keeps measurement intervals and handling consistent. Both dishes again begin with 50 g water. After two hours the uncovered dish has 46 g and the covered one has 48 g. The uncovered dish loses 4 g, the covered one loses 2 g, and the observed difference is 2 g. Check: 50 − 46 = 4, 50 − 48 = 2, and 4 − 2 = 2.
This second illustration is easier to interpret because the documented fan disturbance is absent. But it is still only one observation per condition and could be affected by other small differences. A sensible student might propose repeats, swap equivalent positions between trials, label vessels clearly and record the room conditions. These changes aim to reduce alternative explanations; they do not guarantee a particular outcome.
A three-column checking notebook
Have pupils create a short log with headings time and measured outcome, intended experimental condition, and unexpected event. At zero minutes, write both water masses and the location of each dish. At 60 minutes, note if a dish moved or airflow changed, even if no mass was recorded. At 120 minutes, enter the measured final values. A notebook that admits a disturbance is better science than a flawless-looking table that silently hides it.
The Science Buddies guide to independent, dependent and controlled variables distinguishes the deliberately changed factor from important conditions to keep constant. It explains why two simultaneous differences can undermine a cause-and-effect interpretation. In our case the experimenter chose the cover, but the fan shift was an uncontrolled extra difference.
Four mistakes a learner might make
- Control-must-be-flat: declaring a control invalid because its outcome changes normally with time. Ask what the dependent variable should do even without treatment.
- Any-difference-means-cause: crediting all of a group difference to the cover without checking placement and handling.
- Repair-by-invention: changing measured data to make the table look like an ideal experiment rather than preserving what was actually observed.
- One-repeat-proves-everything: treating one cleaner trial as a universal guarantee instead of limited evidence that could be checked further.
Progressive guided-to-independent practice
Guided: separate the words
Give the learner four labels: ‘loose cover’, ‘water mass at two hours’, ‘dish material’, ‘uncovered dish’. Ask for the independent variable, dependent variable, controlled variable and control condition. The answers are, respectively, cover status, remaining water mass, dish material and uncovered dish. Then ask, ‘Must the uncovered dish’s water mass stay at 50 g?’ The expected answer is no, because evaporation can occur in the absence of a cover.
Partner work: calculate and qualify
Use the first 50 g example. One student calculates losses and the 4 g difference; the other reads the fan note and writes a careful conclusion. Together they should produce: ‘The uncovered dish lost 7 g and the covered dish lost 3 g, but the uncovered dish was moved nearer a fan, so the observed difference cannot be assigned solely to the cover.’ Have them identify exactly which sentence would be misleading if the fan note were omitted.
Independent transfer: a control that changes normally
Present an unrelated invented seedling investigation in which control seedlings grow from 4 cm to 6 cm and treated seedlings grow from 4 cm to 7 cm during the same week. Ask whether the control’s 2 cm growth is automatically an error; it is not. The observed increases are 2 cm and 3 cm, respectively, with a 1 cm difference in growth. The learner must list other conditions to check before attributing that difference to the treatment, such as light, water and starting plant characteristics.
Parents, teachers and observable progress
Parents can ask, ‘What did you change on purpose, what did you measure, and what else might have changed?’ These three questions are more useful than asking whether the practical got the ‘right’ result. Teachers can record whether learners distinguish a changing dependent variable from a changed controlled condition; independently calculate differences; describe unexpected events honestly; and propose an improved fair comparison. Recheck after a new task to see whether the student can make these distinctions without prompts.
The Core Aim investigation of germination controls develops the same fundamental concept in a different biological setting. The subject-specialist guide to control experiments, the eduKate education learning hub and the Clementi explanation-first reference provide additional practice and ways to diagnose mistakes instead of hiding them.
Frequently asked questions
Is a control sample allowed to change?
Yes. A control condition may exhibit the natural or background process being measured. What matters is that its relevant conditions are comparable to those of the treated condition apart from the planned difference.
Does an uncontrolled event mean all the data are useless?
No. The measurements and event record can still describe what happened. The uncontrolled event limits which explanations the data can support. Preserve the information and state that limitation rather than rewriting the history of the experiment.
Can a larger number of repeated trials fix a systematic problem?
Repeats can reveal variability, but repeating the same biased placement will not remove that confound. Improve the design, then repeat under documented comparable conditions. This distinction between reliability and validity is central to practical science.
The aim behind a fair experiment
The point of a control is not to produce a motionless number. It is to create a meaningful reference for what would happen without the treatment. When a learner checks which conditions truly stayed comparable—and records when they did not—science becomes a habit of honest explanation instead of a hunt for the prettiest result.