A fair test is not simply an experiment with many things kept the same. It is a comparison designed so that the changed factor can be linked meaningfully to the measured outcome. Primary 6 pupils often know the words “changed variable”, “measured variable” and “controlled variable”, yet still struggle when two conditions change together, a control is irrelevant, a measurement does not match the question or the method cannot support the conclusion.
This guide develops experimental validity, confounding variables, control conditions, fair comparisons and targeted method improvement for Primary 6 and PSLE Science.
Return to the Primary 6 Science Learning Hub.
The validity rule
QUESTION → ONE INTENDED CHANGE → MEASURED OUTCOME → RELEVANT CONTROLS → PROCEDURE → DATA → CONCLUSION → LIMIT.
This is an eduKate reasoning routine, not an official SEAB formula.
Part I — Start from the scientific question
Before identifying variables, state the relationship being tested.
Example: “How does surface type affect the distance travelled by the same toy car after the same release?”
This reveals:
- changed variable: surface type;
- measured outcome: distance travelled;
- important controls: same car, same release condition, same starting point and comparable measurement method.
Variables are easier to identify when the scientific question is clear.
Part II — A confounding variable changes with the intended variable
A confounding variable is another relevant condition that changes at the same time and can also affect the outcome.
Example: Plant A receives bright light and 100 mL of water. Plant B receives dim light and 50 mL of water. Plant A grows more.
The experiment cannot isolate the effect of light because water also changed.
A strong evaluation answer says which extra variable changed and why that prevents a clean conclusion.
“Not fair” is too vague
Weak: “The experiment is not fair.”
Stronger: “Both light intensity and water volume differed between the plants, so the growth difference cannot be attributed confidently to light intensity alone.”
The second answer identifies the confounding variable and the consequence for the conclusion.
Part III — Controlled variables must be relevant
A controlled variable should be a factor that could reasonably affect the measured outcome.
For a friction experiment, relevant controls may include:
- same car or block;
- same mass/load;
- same release method;
- same starting position;
- same measurement method.
“Use the same colour ruler” is not scientifically useful unless colour somehow affects measurement, which it normally does not.
Do not over-control the changed variable
If surface type is the variable being tested, it must change. A pupil who writes “keep the surface type the same” has removed the investigation itself.
Part IV — Control setup versus controlled variable
A controlled variable is kept comparable across setups.
A control setup is a comparison setup used to show what happens without a treatment or tested condition, where appropriate.
Example: if testing whether light is required for a process, a light-exposed setup may be compared with a light-restricted setup. The comparison arrangement acts as evidence about the changed condition.
Do not use these two terms interchangeably.
Part V — The measured outcome must answer the question
An experiment can control variables carefully yet still be invalid if it measures the wrong outcome.
Question: “How does lamp distance affect photosynthesis-related gas production?”
Measuring plant height after five minutes would not directly answer the short-term gas-production question.
Bubble count or collected gas volume may be more relevant indicators, depending on the setup and its limitations.
Proxy measurements have limits
A proxy is an indirect indicator.
Bubble count may represent gas production, but bubble sizes can differ.
Leaf colour may indicate plant condition, but it does not directly measure every aspect of photosynthesis.
When evaluating a method, ask whether the proxy tracks the intended process closely enough.
Part VI — Starting conditions matter
Two trials may use the same procedure but begin from different baselines.
Example: Cup P begins at 80°C while Cup Q begins at 65°C. Comparing their final temperatures after ten minutes does not isolate insulation quality cleanly because starting temperature differs.
A fair comparison often requires comparable starting conditions.
Part VII — Same time versus same stage
Experiments may need comparison at:
- the same elapsed time;
- the same developmental stage;
- the same starting temperature;
- the same initial length;
- the same release condition.
Which one matters depends on the scientific question.
Part VIII — Repetition improves reliability, not variable isolation
Repeating an unfair experiment does not make it fair.
If light and water both change, ten repeats still leave the cause confounded.
Repetition helps assess consistency only after the design is valid enough to answer the question.
Part IX — Sample size in living systems
One plant, one animal observation or one small environmental sample may be unusually affected by natural variation.
Using several comparable specimens or locations can strengthen the evidence.
But sample size does not replace control. Ten differently watered plants under differently lit conditions still produce a confounded comparison.
Part X — Order effects
Sometimes the order of testing changes the system.
Example: the same spring is repeatedly stretched with increasing loads. If the spring becomes permanently changed, later trials may not be directly comparable with earlier ones.
Example: a surface becomes wet after one trial and is reused without drying.
Ask whether the procedure itself changes the test object over time.
Part XI — Carryover effects
A previous condition can influence the next trial.
If a metal cup remains warm before a second cooling test, the second trial does not start from the same state.
Resetting the apparatus can be part of a valid method.
Part XII — Observer and measurement consistency
Two pupils may count bubbles differently or decide “bright” and “dim” differently.
A better method defines the measurement rule clearly and uses the same method across trials.
Where possible, quantitative measurements can reduce ambiguity.
Part XIII — Original case study: lamp and plant
A pupil compares two plants. Plant A is 10 cm from a lamp and receives 100 mL of water. Plant B is 30 cm from the lamp and receives 60 mL of water. After one week, A is taller.
Problem
Both lamp distance and water volume changed.
Improvement
Keep water volume and other relevant conditions comparable, changing only lamp distance.
Conclusion limit
The original result does not isolate lamp distance as the cause of the growth difference.
Original case study: friction car
Car A is released from a 20 cm-high ramp onto smooth plastic. Car B is released from a 30 cm-high ramp onto rough cloth. B travels farther.
Problem
Ramp height and surface type both changed.
Why it matters
The different starting condition changes the car’s motion before it reaches the test surface, so distance cannot be attributed to surface type alone.
Original case study: cooling material
Two cups use different insulating materials. Cup P contains 200 mL of water at 80°C. Cup Q contains 100 mL at 80°C.
Water volume is a confounding condition because it can affect cooling behaviour. To compare insulation material, volume should be kept comparable unless volume itself is being tested.
Original case study: food-web survey
Site A is surveyed in the morning after rain. Site B is surveyed at noon on a dry day. More insects are counted at A.
Location, time of day and recent weather all differ. The comparison cannot isolate location alone.
Part XIV — Validity versus reliability
Validity: does the design answer the intended question?
Reliability: do repeated measurements show consistent results?
A method can be reliable but invalid.
Example: repeatedly measuring leaf length when the question asks for gas production can give consistent numbers while measuring the wrong outcome.
Part XV — Fairness versus realism
A highly controlled experiment may simplify the real world to isolate one relationship.
An environmental observation may be more realistic but harder to interpret causally because many factors vary together.
Both types of evidence can be useful for different scientific jobs.
Part XVI — Method improvement should be surgical
Do not write generic improvements such as “repeat more” or “use better equipment” unless they fix the identified weakness.
| Weakness | Targeted repair |
|---|---|
| Two variables changed | Keep the unwanted variable constant |
| Starting values differ | Standardise starting condition |
| Measurement too subjective | Use a defined quantitative measure if suitable |
| One trial only | Repeat the valid procedure |
| One specimen only | Use several comparable specimens |
| Instrument too coarse | Use suitable finer resolution |
| Previous trial affects next | Reset apparatus between trials |
Part XVII — Evaluating conclusions
A conclusion should name the tested relationship and stay within the design.
Good: “Within the tested range, increasing the load increased spring extension for this spring.”
Too broad: “All springs extend more whenever any load is added.”
Good conclusions remember both validity and boundary.
Part XVIII — MCQ strategy
When several methods are offered, prefer the one that:
- changes the intended variable clearly;
- controls plausible alternatives;
- measures the relevant outcome;
- uses comparable starting conditions;
- provides repeatable measurement;
- supports the stated conclusion without overreach.
Part XIX — Open-ended evaluation structure
Use:
“[Variable/condition] was not controlled / was also changed. It can affect [measured outcome], so the effect of [intended variable] cannot be isolated confidently.”
Then add the targeted improvement if asked.
Validity checking routine
- What relationship is the experiment supposed to test?
- What is deliberately changed?
- What is measured?
- Which other factors can affect that measurement?
- Were those factors controlled?
- Were starting conditions comparable?
- Does the measurement represent the intended process?
- Was the valid method repeated?
- Does the conclusion stay inside the evidence?
Where to connect
- Investigations, Variables, Fair Tests & Method
- Models, Assumptions, Limits & Scientific Claims
- Measurement, Units, Resolution & Repeatability
Retrieval checklist
- I can state the scientific question before naming variables.
- I can identify a confounding variable.
- I can explain why a confound weakens the conclusion.
- I can choose relevant controlled variables.
- I can distinguish a controlled variable from a control setup.
- I can judge whether the measured outcome answers the question.
- I can recognise unequal starting conditions.
- I know repetition does not fix a confounded design.
- I can identify carryover and order effects in simple setups.
- I can propose a targeted improvement.
The quiet return
A fair test is a piece of causal architecture. Every control protects the pathway from the changed variable to the measured outcome. Every unnecessary change creates another possible explanation.
Ask one clear question. Change one intended condition. Measure the right outcome. Control plausible alternatives. Let the design determine what may be concluded.
Return to the Primary 6 Science Learning Hub.
