Wait, What? A perfectly careful experiment can still answer the wrong question.

You can measure every volume neatly, read every scale correctly and repeat every trial three times — and still produce weak science. Why? Because practical quality begins before the first measurement. It begins with whether your experiment is actually capable of testing the question you think it is testing.

This is the central idea of experimental design: change the thing you mean to change, measure the thing you mean to measure, and prevent other plausible causes from quietly changing the result.

The practical-science question behind every experiment

Suppose you want to investigate how temperature affects the rate of a reaction. The visible experiment may involve a flask, thermometer and stopwatch. The scientific structure underneath it is more important:

• Independent variable: the factor deliberately changed — temperature.
• Dependent variable: the response measured — perhaps time to produce a fixed volume of gas, or gas volume produced in a fixed time.
• Control variables: other factors that could change the response — concentration, volume, reactant mass, particle size, apparatus geometry and endpoint rule.

Students often memorise these three labels. Strong practical scientists go one level deeper: why must each control be controlled? If concentration changes, collision frequency changes. If particle size changes, surface area changes. If the endpoint rule changes between trials, you are no longer comparing like with like.

A fair test is not the same as a good experiment

A fair comparison is necessary, but not sufficient. A high-standard experiment also needs a sensible range, enough data, appropriate measurement resolution, repeatability, a method that can actually reveal the expected effect, and a risk plan.

Imagine testing the effect of temperature using only 20 °C and 25 °C. Even if everything else is controlled, the range may be too narrow for a clear trend to emerge above ordinary measurement scatter. Now imagine using 10, 20, 30, 40, 50 and 60 °C. The wider spread may reveal the relationship more clearly — provided the system remains safe and the mechanism is not changing in some new way at the extremes.

Variables: write them operationally

Weak planning says: “Change temperature.” Strong planning says: “Set the reaction mixture to 20, 30, 40, 50 and 60 °C using a water bath, checking the temperature immediately before mixing the reactants.”

Weak planning says: “Measure rate.” Strong planning says: “Measure the volume of gas collected every 10 s for 120 s using a gas syringe.”

This is called an operational definition: the variable is defined by what you will physically do or measure. It makes the plan reproducible and exposes hidden ambiguities.

Controls: ask what else could cause the change

For each control variable, complete this sentence: If this changed, it could affect the dependent variable because…

That causal sentence is the difference between memorising a list and understanding experimental validity. In biology, light intensity may need controlling because it changes photosynthetic rate. In physics, wire length may need controlling because resistance depends on length. In chemistry, total solution volume may matter because it changes concentration after mixing.

Repeats do not repair a bad design

Repeating measurements is useful because random variation can be detected and reduced by averaging. But repeats cannot rescue a systematic design flaw. If every temperature reading is taken with the same thermometer that is offset by +2 °C, repeating the measurement ten times does not remove the offset. If heat is always being lost through the same uninsulated cup, repeats do not eliminate that heat loss.

So ask two different questions: Is my result repeatable? and Is my method valid for the scientific question?

How many repeats?

There is no magical universal number. At school level, repeated trials are commonly used to reveal scatter and identify suspicious values. The important reasoning is to collect enough information to judge consistency, not mechanically write “repeat three times” without purpose.

If repeated values are 12.1, 12.0 and 12.2 s, the cluster is tight. If they are 12.1, 17.4 and 11.9 s, something requires investigation. Do not automatically delete the 17.4 s reading. First ask whether there is evidence of a procedural failure, recording mistake or identifiable disturbance.

Range and intervals

A good range should be wide enough to expose a relationship but remain within safe and scientifically meaningful limits. Intervals should normally give enough points to reveal shape, not merely two endpoints.

Five well-chosen values can show a trend that two values cannot. More points may reveal curvature, thresholds or saturation. At JC level, this becomes especially important because a model may only hold over a limited range. A straight-line model is not automatically valid merely because two points can always be joined by a line.

Sampling is experimental design too

When the object of study varies naturally — leaves, organisms, soil, populations, ecosystems — the question becomes not only “How precisely did you measure?” but “Did you sample representatively?” Measuring one unusually large leaf to describe a whole plant is a sampling problem, not an instrument problem.

Students should distinguish measurement repetition from biological replication. Re-reading the same leaf five times tells you about measurement consistency. Measuring many independently sampled leaves tells you something about variation among leaves.

Planning safety without turning it into a slogan

“Wear goggles” is sometimes correct, but risk assessment should connect hazard → possible harm → control measure. For example: concentrated acid can cause chemical burns; use dilute quantities where suitable, wear eye protection and rinse splashes immediately according to laboratory procedure. A hot water bath can scald; use appropriate temperature limits, stable containers and tongs or heat protection where needed.

The Singapore-Cambridge practical framework explicitly includes identifying risks and precautions as part of planning. Read the current SEAB Chemistry practical-assessment requirements here.

Secondary → JC progression

Secondary: identify independent, dependent and control variables; describe a workable method; choose sensible apparatus; repeat measurements; recognise errors; state precautions.

JC: justify ranges and transformations; distinguish random variation from systematic effects; recognise assumptions; test whether a model is supported by data; reason about uncertainty; understand when a control is insufficient; evaluate whether the method measures the intended quantity.

Checkpoint: can you rescue this experiment?

A student investigates how light intensity affects photosynthesis. She places a lamp at 10, 20, 30, 40 and 50 cm from pondweed and counts bubbles for one minute at each distance.

• What is the independent variable as actually manipulated?
• What is the dependent variable?
• Name two important control variables and explain why they matter.
• Why is distance from the lamp not identical to directly measuring light intensity?
• Why might counting bubbles be weaker than measuring gas volume?

Answer key and WHY reasoning

The manipulated variable is lamp distance, used as a proxy for light intensity. The dependent variable is bubble count per minute. Temperature should be controlled because the lamp may warm the water and temperature affects enzyme-controlled reactions. Pondweed size or exposed photosynthetic area should be controlled because more tissue may produce more oxygen. Lamp distance is only a proxy because actual irradiance also depends on lamp geometry, ambient light and absorption. Bubble counting is weaker because bubble size can vary, so equal bubble counts do not necessarily mean equal gas volumes.

How to study experimental design

Do not revise practical planning by rereading notes. Take unfamiliar scenarios and force yourself to produce five things from scratch: the question, the variables, a reproducible method, a data plan and an evaluation plan. Then interrogate every control with “what mechanism would make this matter?”

A strong student gradually stops seeing an experiment as apparatus. They see it as a causal argument built from controlled comparison.

Authoritative next steps

• SEAB 2026 O-Level syllabuses
• SEAB 2026 A-Level syllabuses
• Royal Society of Chemistry: evaluating experiments

Teaching Guide

For parents and teachers: practical improvement is fastest when students are asked to justify decisions rather than recite vocabulary. Give a short scenario and ask, “What would make this comparison unfair?”, “What would you measure?”, “What would count as convincing evidence?”, and “What result would make you doubt your own method?” Require causal explanations. Praise specific evaluation, not the phrase “human error.”