G3 Science practical and data questions become much easier when controls are treated as reasoning tools rather than memorised vocabulary. This workshop trains the learner to identify what a control actually rules out, what a confounder leaves unresolved, and how that changes the conclusion.
It follows Vol 0067 and the evidence discipline in Vol 0065. Use the G3 SEC learner route for the wider examination sequence and the Science Hub for underlying concepts.
For 2027 school candidates, the official K326/K327/K328 combined Science syllabus and SEAB G3 syllabus directory are the authorities for the registered combinations and practical assessment. The cases below are original teaching material; practical work must follow school and examination safety instructions.
A control is there to answer a specific alternative explanation
A good control is not simply another group kept the same. It helps answer a particular question: could the observed result have happened without the factor being tested? The learner should always connect a control to the alternative explanation it is meant to rule out.
Controlled variable and control condition are different ideas
A controlled variable is a quantity kept constant across compared conditions, such as time or starting temperature. A control condition is a comparison condition designed to show what happens without the tested factor or under a reference state. The words sound similar, but their jobs are different.
A confounder changes with the factor you care about
A confounding factor is another difference between conditions that could explain the result. If two plants receive different light levels and also different amounts of water, an observed growth difference cannot confidently be attributed to light alone.
Alternative explanation does not mean proven explanation
If water amount differed, you may say it is an alternative explanation for the growth difference. You cannot say water definitely caused the result unless the evidence supports that stronger claim. Good evaluation limits certainty without inventing a new cause.
Paper 5 makes this reasoning practical
The 2027 G3 combined Science practical assesses more than following steps. Learners need to measure, record, process and evaluate evidence. Control logic matters because conclusions depend on whether the comparison actually isolates the factor of interest.
Worked case 1: cooling containers
Two insulated cups are compared. Cup A starts at 80°C with 200 mL of water. Cup B starts at 70°C with 200 mL. After ten minutes, A is 63°C and B is 57°C. A learner claims A is the better insulator because its final temperature is higher.
Identify the confounder before calculating more
The starting temperature differs. That changes the initial temperature difference between the water and surroundings. The comparison does not isolate insulation quality. More decimals in the temperature loss will not repair the design.
Repair the comparison
A stronger test would use matched starting temperatures, equal water mass or volume, equal duration and relevant matched environmental conditions, then compare the temperature changes. The repair targets the confounder rather than simply adding more repeats of the flawed setup.
Worked case 2: a control condition for germination
Imagine a school-supervised investigation testing whether a treatment affects seed germination. One group receives the treatment; a comparison group receives the same relevant conditions except for that treatment. The control condition helps show whether germination occurs without the tested factor.
Do not call every unchanged variable the control
Temperature, water and seed type may be controlled variables. The untreated group is the control condition. Naming every constant as “the control” makes it harder to explain what each part of the design is doing.
Worked case 3: exercise and heart rate
A fictional class compares heart rate before and after exercise. One learner also drinks a caffeinated beverage between measurements. If the aim is to attribute the change to exercise, the additional intake creates another factor that could affect heart rate.
Evaluation should identify the mechanism of the confounder
Do not write only “caffeine is unfair.” Explain why the extra factor matters: it could affect heart rate independently of exercise, making the observed change harder to attribute to the intended variable.
Worked case 4: dissolving rate
Two beakers contain equal masses of the same solid. One uses warmer water and is stirred; the other uses cooler water and is not stirred. A faster dissolving rate in the first beaker cannot be attributed specifically to temperature because stirring changes too.
One change at a time is a useful design principle
To test temperature, keep stirring condition, particle size, amount of solute and other relevant factors matched. To test stirring, match temperature. The point is not that every imaginable factor must be identical, but that competing explanations relevant to the conclusion must be controlled.
Worked case 5: surface area and reaction rate
Two equal masses of the same solid react under matched conditions, but one is finely divided and one is in large pieces. If the finer sample reacts faster, the design is more informative because surface area is the intended difference. Keep temperature, concentration and amount of reactants appropriately matched.
Do not confuse rate with final amount
A faster reaction can reach the same final amount when the limiting quantities are unchanged. Rate and extent are different targets. A control design may isolate the reason for a rate difference without changing the theoretical final yield.
Worked case 6: plant growth and location
Two sets of plants receive different fertilisers, but one set is kept near a window and the other in a shaded corner. Light is a confounder because it changes with fertiliser condition and can affect growth.
Random placement can reduce location bias in some designs
If appropriate to the task, randomising positions or rotating comparable setups can help prevent systematic location differences from aligning with one treatment. In an examination, describe only methods justified by the context and official practical instructions.
Repeats do not fix confounding
Repeating a flawed comparison many times can improve information about random variation while leaving the same systematic design problem intact. If every A plant receives more light than every B plant, more plants do not isolate the fertiliser effect.
A control can fail if it differs in another important way
A control group is useful only if the relevant comparison is fair. If the control uses a different container material, different starting temperature or different measurement method, it may introduce its own alternative explanation.
Negative control and positive control have different purposes
In some scientific contexts, a negative control checks what happens without the target effect, while a positive control checks that the system can produce a known response. Use these terms only when the task or scientific context supports them; do not force specialised labels into every school experiment.
Calibration checks a measuring system, not the biological or physical hypothesis
If an instrument reads zero incorrectly, the measurement system may bias every result. A calibration or zero check can address the instrument, but it does not replace the experimental control needed to test the causal question.
Worked case 7: two thermometers disagree
Thermometer A reads 24.0°C and B reads 24.8°C in the same reference condition. Averaging to 24.4°C does not automatically solve the problem. First inspect calibration, zero error, resolution and whether both instruments truly measured the same location.
Systematic error behaves differently from random variation
A systematic offset can shift every reading in the same direction. Repeats alone may not remove it. Random variation can cause readings to scatter around a value; repeats and averaging may help estimate the underlying quantity when the method is otherwise sound.
Worked case 8: measurement order as a confounder
Suppose every low-temperature trial is measured first and every high-temperature trial later, while the apparatus gradually warms. Order is now linked to condition. The learner should ask whether the sequence itself could influence the result.
Counterbalancing can separate order from condition
Where a school task allows and the design requires it, varying order across repeats can prevent one condition from always occurring first. Again, this is a design principle for reasoning, not permission to change an official examination procedure.
Worked case 9: survey evidence in Science communication
A school Science project asks students whether they find a display easy to read. Positive opinions can evaluate usability, but they do not measure the display’s physical brightness or sensor accuracy. Different evidence types answer different questions.
Do not let a questionnaire stand in for a measurement
If the claim concerns temperature, mass, time, brightness or concentration, use the appropriate measurement. Perception data may still matter, but it should not silently replace the physical quantity.
Worked case 10: correlation and alternative explanation
A dataset shows that students who sleep longer also report better concentration. This association does not isolate sleep as the cause. Other factors such as schedule, health or workload may differ too. The learner should distinguish association from a controlled causal comparison.
A mechanism can strengthen a causal explanation but does not automatically prove it
Knowing a plausible mechanism helps explain how one factor could affect another. But if the observational data are confounded, the mechanism alone does not establish that it caused the observed pattern in that dataset.
Control logic in Physics
In a Physics comparison, keep relevant dimensions, masses, distances, initial conditions or circuit components matched depending on the question. The control choice should come from the relationship being tested, not from a memorised checklist.
Control logic in Chemistry
In Chemistry, concentration, temperature, surface area, amount of substance and catalysts can all affect observed behaviour in different questions. Identify which one is the intended variable and which could provide an alternative explanation.
Control logic in Biology
Biology often adds natural variability among organisms. Matching species, age, size or starting condition may matter depending on the investigation. Repeats can help reveal biological variation, but the main comparison still needs relevant controls.
A fair test is not a slogan
Writing “keep everything the same” is usually too vague and sometimes impossible. Name the relevant variables and explain how they affect the conclusion. Strong practical evaluation is specific enough that the reason for each control is visible.
Worked evaluation: vague versus specific
Vague: “Keep the experiment fair.” Specific: “Use equal volumes of water at the same starting temperature so any difference in cooling can be linked more confidently to the container rather than to different thermal starting conditions.” The second answer connects method to inference.
Alternative explanation checklist
When a result appears, ask what else differed between the conditions, whether the measurement method changed, whether time or order changed, whether the groups started differently, and whether an outside factor could affect the outcome. Then choose only the alternatives relevant to the case.
Do not list every possible confounder
An examination answer should not become a catalogue of imaginable problems. Identify the most relevant design weakness supported by the scenario. Precision scores better than generic scepticism.
A confounder must be able to affect the outcome
If a factor differs but has no plausible relation to the measured outcome, it may be irrelevant. The learner should connect the alternative factor to a possible effect instead of naming differences mechanically.
A controlled variable may not need perfect numerical identity
Some tasks require keeping a quantity within a suitable range or using the same method rather than matching an impossible exact value. Read the practical context. The principle is to prevent a relevant competing explanation.
Worked case 11: container surface area
Two cooling containers have different materials and different exposed surface areas. If one cools faster, material and surface area both differ. The learner cannot isolate material without controlling or accounting for the geometry.
Worked case 12: light intensity and distance
A lamp experiment compares readings at different distances, but the sensor angle also changes. If angle affects the reading, distance is no longer the only systematic difference. The design should maintain the relevant orientation.
Worked case 13: timing a pendulum
A school-supervised timing exercise compares pendulum lengths. If one length is timed for five oscillations and another for twenty, the measurement procedure differs. The learner should compare equivalent derived quantities or use a consistent method.
Worked case 14: reaction endpoint judgement
If one trial uses a colour endpoint judged by eye and another uses an instrument threshold, the detection method differs. A difference in recorded time may partly reflect measurement method rather than reaction behaviour.
Controls and validity
A design has better validity when the evidence actually addresses the intended question. Controlling relevant alternative explanations supports validity because the observed difference can be linked more closely to the variable being tested.
Repeats and reliability
Repeats can improve information about consistency and random variation. They support reliability, but they do not by themselves guarantee validity or accuracy. Keep these ideas separate.
Accuracy and calibration
Accuracy concerns closeness to the relevant true or accepted value where that concept applies. Calibration and systematic error matter. A very repeatable instrument can still be consistently wrong.
Precision and resolution
Precision in measurement language can refer to consistency or to how finely a quantity is reported, depending on context. Instrument resolution also matters. Use the terms as the syllabus or question defines them rather than relying on casual everyday meaning.
Independent task A
Two cups are compared for insulation. Cup A begins at 85°C, Cup B at 75°C. Both are measured after ten minutes. Identify the main confounder and write one specific repair.
Independent task B
Two plant groups receive different fertilisers. Group A is also watered daily; Group B every two days. Explain why fertiliser effect is not isolated and identify the variable that must be matched.
Independent task C
A reaction-rate comparison changes both temperature and particle size. Write a sentence explaining why a faster rate cannot be attributed to temperature alone.
Independent task D
An instrument gives nearly identical readings on five repeats, but a known zero error has not been corrected. Explain what the repeats show and what they do not show.
Independent task E
A dataset shows a strong association between study time and test score. Give one reason this does not automatically prove that increasing study time by itself caused the higher scores.
Independent task F
A practical report says, “Repeat the experiment to make it fair.” Explain why this is incomplete and rewrite it as a matched improvement for a stated random-variation problem.
Worked feedback A
The starting temperature differs and can affect cooling behaviour. A specific repair is to use the same starting temperature, with other relevant conditions matched, then compare the temperature change over the same duration.
Worked feedback B
Watering frequency changes with fertiliser condition and can affect growth. Match the watering amount and schedule appropriately so the fertiliser comparison is not mixed with water availability.
Worked feedback C
Temperature and particle size both affect reaction rate under many relevant contexts. Because both changed, the observed difference has more than one plausible explanation. Match particle size when testing temperature, or match temperature when testing particle size.
Worked feedback D
The repeated readings show consistency under the same biased method. They do not establish accuracy if a systematic zero error shifts every reading. Correct or account for the known offset according to the task.
Worked feedback E
Students who study longer may differ in prior attainment, sleep, motivation, subject difficulty or other factors. The association is real in the dataset, but causation needs stronger evidence about alternative explanations.
Worked feedback F
Repeats can help estimate random variation if the method is otherwise valid. A better statement is: “Repeat the measurement under the same controlled conditions and compare the readings to assess consistency; this does not by itself remove a systematic bias.”
How to review a practical evaluation
Underline the limitation. Draw an arrow to its likely effect on the data or conclusion. Then check whether the proposed improvement directly interrupts that effect. If the arrow cannot be drawn, the improvement may be generic.
How to review a causal conclusion
Ask whether the compared conditions differed only in the intended variable among the relevant factors. If not, name the alternative explanation and reduce the claim. Do not replace it with a new untested cause.
How to review a control condition
State what would be expected in the control if the tested factor were absent or at the reference level. This makes the control’s purpose explicit and prevents it from becoming a decorative extra group.
Repair route
Begin with simple two-condition comparisons. Ask which factor changed intentionally, which outcome was measured and what other relevant difference remained. Repair one confounder at a time.
Stabilisation route
Mix experiments so that some are already well controlled and others contain one hidden design problem. The learner should decide when no repair is needed instead of assuming every question must contain a flaw.
Extension route
Use cases with two plausible alternative explanations and ask what additional comparison would separate them. The goal is not to list possibilities but to design evidence that changes what can be concluded.
What progress should look like
Progress is visible when the learner distinguishes controlled variables from control conditions, identifies confounders that actually matter, explains why repeats do not fix systematic flaws, and proposes improvements tied to specific limitations.
Frequently asked: should everything be kept constant?
No. The independent variable must change. Other variables are controlled when they are relevant to the causal question. The phrase “keep everything the same” hides the reasoning.
Frequently asked: is more data always better?
More data can help, but not if the same confounder remains in every trial. Data quantity does not automatically repair design quality.
Frequently asked: does a control prove causation?
A well-designed control strengthens causal interpretation by ruling out a specific alternative. Strong causal conclusions usually depend on the whole design, measurements and consistency of evidence, not on the word control alone.
Frequently asked: can I suggest several improvements?
Yes if the question asks for them, but each should address a real limitation. One precise matched improvement is often stronger than several generic suggestions.
Final operating rule
When evaluating an investigation, name the intended variable, the measured outcome, the relevant controlled variables, the purpose of any control condition and the strongest alternative explanation still left. Then state exactly how that remaining alternative limits the conclusion.
G3 Science control-and-confounder checklist
- name the intended independent variable
- state the measured dependent variable
- identify only relevant controlled variables
- explain what the control condition rules out
- find any alternative factor that changed with the treatment
- separate random variation from systematic error
- match every improvement to a specific limitation
- keep the conclusion proportional to the design
Advanced control-logic laboratory
Advanced case: a control can answer the wrong question
Suppose a learner includes an untreated group but the real concern is whether two measurement methods agree. The untreated group does not address the method-comparison question. A control is useful only when its design matches the alternative explanation being tested.
Advanced case: one control may not rule out every alternative
A comparison may control starting temperature but still use different container sizes. The first repair improves the design without making it perfect. Scientific evaluation should identify what has been ruled out and what remains unresolved.
Advanced case: controls and baseline drift
If an instrument reading slowly changes over time even when the input is constant, a reference reading taken only at the beginning may not reveal later drift. A suitable control or repeated reference can help detect the measurement-system change, depending on the task.
Advanced case: background effects
In measurement work, a background signal may be present even without the target source. A background or blank condition can help estimate that contribution. The learner should explain what background correction addresses rather than treating it as a universal ritual.
Advanced case: matched controls can still have poor measurement
A perfectly matched comparison can still produce weak evidence if the instrument is too coarse, the endpoint is subjective or the recording method is inconsistent. Experimental control and measurement quality are different parts of validity.
Advanced case: more precise language about fairness
Instead of writing “make it a fair test”, name the variable and consequence: “Keep the exposed surface area the same so the cooling difference is not explained by geometry.” This sentence shows what fairness means in the actual investigation.
Advanced case: partial control and cautious conclusion
If a study controls temperature and mass but not stirring, the conclusion may still support an association under the tested conditions without isolating one cause fully. The learner should reduce the causal strength rather than discarding every observation.
Advanced case: a control can reveal contamination
If a condition expected to show no response unexpectedly shows a strong response, the result can signal contamination, background effect or method failure. The correct action is investigation, not automatic subtraction or deletion.
Advanced case: repeated controls can reveal instability
If control readings vary widely across repeats, the experimental system itself may be unstable. This weakens confidence in small treatment differences. The learner should connect control variability to interpretation of the main comparison.
Advanced case: ceiling and floor effects
A measurement system may saturate at the top or bottom of its range. Two conditions can look identical because the instrument cannot resolve further change. The learner should consider whether the measurement range can reveal the expected difference.
Advanced case: confounding by starting level
If two groups start at different baseline values, comparing only final values can be misleading. Depending on the task, change from baseline or matched starting groups may provide a fairer comparison. The correct approach follows the question and design.
Advanced case: confounding by duration
A treatment tested for ten minutes should not be compared casually with another tested for twenty minutes if duration can affect the outcome. Time is a variable, not merely a scheduling detail.
Advanced case: confounding by measurement position
Temperature, light intensity or concentration can vary by location. If two conditions are measured at different positions, position may become an alternative explanation. Keep the measurement point consistent where relevant.
Advanced case: sample identity
Two materials labelled the same type may still differ in age, size, concentration or preparation. If those differences can affect the outcome, sample identity needs tighter control or cautious interpretation.
Advanced case: control logic in a graph
A graph can show treatment and control changing together. If both rise by similar amounts, the shared change may indicate an external factor affecting both. If only the treatment changes, the contrast supports—but does not by itself guarantee—the intended causal interpretation.
Workshop drill: name the alternative in one sentence
For every evaluation case, write: “Because ___ also differed, the result could be explained by ___ rather than only by ___.” This forces the learner to connect the design flaw to the conclusion instead of listing variables without meaning.
Workshop drill: match the repair
Then write: “Keep ___ the same / measure ___ consistently / add a comparison without ___ so that ___ is no longer an alternative explanation.” A repair is strong when its purpose can be stated explicitly.
Workshop drill: separate reliability from validity
Write one sentence about consistency and one about whether the method answers the intended question. Repeats can improve the first without fixing the second. This distinction is especially useful in practical evaluation.
Workshop drill: choose when no extra control is needed
Some questions already isolate the intended variable well enough. In practice sets, include well-designed investigations and ask learners to defend them. Evaluation skill includes recognising when the existing control structure is appropriate.
Final control logic
A control earns its place by changing what can be concluded. If removing or adding the control would not affect the interpretation of the tested factor, the learner should question whether that control is relevant to the stated scientific problem.
The final examination habit is to keep control language tied to the actual inference. Ask what you are trying to attribute, what else could produce the same observation, and which comparison would separate those possibilities. If the design already answers that question, do not invent a flaw. If it does not, name the remaining alternative precisely and propose the smallest relevant repair. That is stronger scientific evaluation than memorising a list of control variables.
