How Scientific Experiments Fail | Variables and Controls

Did you know? A student can conduct an experiment carefully, record real observations and still reach a conclusion that the experiment cannot support. Nothing needed to be fabricated. The issue is that two things changed together, so the result could have more than one explanation.

That is a central way scientific experiments fail. The purpose of a fair test is not simply to produce a table or prove a prediction correct. It is to collect evidence that can meaningfully distinguish between competing explanations.

Quick answer: experiments can mislead when the independent and dependent variables are confused, relevant conditions are not controlled, samples are insufficient or unrepresentative, or data are interpreted beyond the study design. A useful repair is to define the question, the comparison and the measurement before beginning.

The first question: what are we trying to find out?

Imagine a classroom studying plant growth. Someone asks: ‘Does more sunlight help seedlings grow?’ The class sets one plant on a sunny windowsill and another in a shaded corner.

Two weeks later, the sunny seedling is taller. Does that prove sunlight caused the difference? Not yet. The two places may also have differed in temperature, watering, pot size, soil, airflow or other conditions.

The measured growth may be genuine, but the comparison has not isolated the possible effect of sunlight. The first step toward a better experiment is to ask which alternative explanations could produce the same observation.

OpenStax Concepts of Biology: The Process of Science explains how testable hypotheses, variables, controls and repeated observations support scientific reasoning. It also notes that actual science is more complex and iterative than a rigid classroom flowchart.

Independent, dependent and other variables

The independent variable is the condition the experimenter deliberately changes or assigns in the comparison. The dependent variable is the outcome observed or measured to examine the response.

In our simplified seedling example, assigned light exposure might be the independent variable. Growth over the period might be the dependent variable, provided ‘growth’ is clearly defined.

Other factors—plant species, initial size, water, soil, pot type and temperature—could influence the outcome. The study needs to handle those factors appropriately rather than pretending they do not exist.

In some scientific designs, several factors are varied deliberately. In others, researchers use statistical methods rather than making every condition literally identical. The classroom rule ‘change one thing at a time’ is a useful starting simplification, not a complete description of modern experimental science.

Six reasons an experiment can produce a misleading answer

Question confusion: the study begins without a precise outcome or comparison.
Variable confusion: the researcher changes one condition but measures another poorly defined quantity.
Confounding: another factor changes alongside the intended intervention and offers a plausible explanation for the observed result.
Weak controls: reference conditions do not demonstrate whether the test can distinguish the proposed effect from background processes.
Insufficient or biased observations: the sample or repeated trials do not capture relevant variation.
Overinterpretation: the conclusion claims a universal rule or a causal effect that the actual design cannot establish.

These mechanisms can overlap. An unclear question often leads to poor measurement. Weak measurement makes results noisy. A small sample then magnifies uncertainty, while a dramatic headline hides the design’s limitations.

A fair-test example: the seedling that grew taller

Illustrative classroom experiment. Suppose two seedling groups start at similar stages and the learner wants to investigate how assigned light conditions are related to growth.

The class might use the same plant species and similar pots, planting medium and watering schedule, while carefully documenting the different light conditions. They would include an appropriate number of plants for the learning task, record relevant conditions and define growth consistently.

Even after these improvements, the experiment has limitations. Position can still influence temperature, and individual seedlings can differ. A responsible report explains those limitations rather than declaring the result unquestionable.

It is also important to distinguish a school demonstration from research meant to establish a reliable biological effect. The level of replication, randomisation, equipment and analysis needed grows with the claim and its consequences.

What is a control group, and what does it actually control?

A control condition provides a meaningful comparison. It may represent standard practice, no exposure to an intervention or another reference state suited to the question.

A negative control can help show that a method does not report a positive response when the target factor is absent. A positive control can help demonstrate that a test is capable of detecting a response under known conditions. These terms have particular uses depending on the discipline and experiment.

OpenStax AP Biology Lab Manual, Activity 2 introduces independent and dependent variables as well as positive and negative controls.

A control is not a magic element that automatically makes a study valid. If the control and treatment groups differ systematically in other important respects, the intended comparison may remain difficult to interpret.

What is a confounding variable?

A confounder is a factor that can distort the relationship between the intended exposure or intervention and the observed outcome.

Imagine comparing the examination scores of students who joined a voluntary tutorial and those who did not. Students who enrolled might have had different prior attainment, family support or motivation. The outcome difference alone would not establish that the tutorial caused all the difference.

That does not mean tuition cannot help. It means the evidence needs a design that makes alternative explanations less plausible.

In education, controlled experiments may use appropriate group assignment and outcome measures. In many settings, randomisation is not feasible or appropriate, so researchers use other designs and explicitly discuss their remaining limitations.

See How Tuition Fails for a related teaching-side question: whether more lessons actually address a specific gap.

A result is not the same as an explanation

A student may measure taller plants, a faster chemical reaction or a difference between two groups. That is an observed result.

The explanation answers why it happened. A credible explanation considers whether the measurement is suitable, whether the intended comparison was fair and which other causes could account for the difference.

It is useful to write findings in three layers: observation, interpretation and limitation. For example: ‘In this setup, the seedlings under the assigned light condition grew more over two weeks. These results are consistent with a possible effect of light, but temperature and small sample size remain potential limitations.’

That statement is more defensible than claiming the class has discovered a universal biological law.

Why repeating an experiment matters

A single trial might be affected by an unusually healthy seedling, an unnoticed measurement slip or a changing environment. Repetition and appropriate replication help distinguish a repeatable pattern from an isolated outcome.

But repeating a flawed method can reproduce the same mistake. If every repetition uses a biased measurement or an unsuitable control, more observations alone will not correct the underlying design.

The US National Academies’ Reproducibility and Replicability in Science separates repeating computations on the same evidence from independently testing a question with new evidence. Both ideas matter, and neither is a substitute for a study that addresses the intended cause.

What about sample size?

A very small number of observations may be inadequate to detect a meaningful difference reliably or reflect variation across cases. But the appropriate sample size depends on the expected effect, variability, design and purpose.

There is no universal rule that ten samples are always enough or that a sample of one is always useless for every scientific question. Some investigations study unique systems or intensive time series; others need much larger comparative groups.

Students can start by asking how much natural variation exists, whether the chosen observations are representative of the question and whether uncertainty is acknowledged in the conclusion.

The goal is thoughtful evidence, not an invented number that makes the result sound professional.

Measurement errors can make a fair design unfair

Even with a well-chosen comparison, poor measurement can spoil the result. Two students may measure plant height from different starting points; one may include a bent leaf, while another measures only the stem.

Before beginning, define where to measure, which tool to use, when measurements are taken and how results are recorded. Use suitable precision rather than adding impressive decimal places.

The companion How Scientific Measurement Fails explains why calibration, sampling and uncertainty are part of the evidence behind every number.

A classroom procedure worth teaching before the experiment

State one precise, testable question.
Define which condition will vary and which outcome will be recorded.
Identify other relevant influences and how the procedure will handle them.
Choose a meaningful control or comparison.
Decide how observations will be measured and recorded.
Plan enough observations or repetitions for the educational purpose.
Explain what alternative findings would mean for the original hypothesis.
Follow laboratory safety instructions and obtain required supervision or approval.

Students should see this as a checklist for clearer thinking, not an inflexible recipe that replaces scientific creativity.

When a negative result is actually successful science

A result that does not support the prediction can still be informative. Perhaps the expected effect does not exist under the tested conditions, is smaller than anticipated or depends on a missing variable.

The scientific failure is not that the data disagreed with the student. The failure would be altering the question after seeing the outcome, hiding inconvenient observations or claiming more certainty than the study can support.

A careful researcher asks what was learned and what better investigation would follow. Sometimes a negative finding is precisely what narrows the set of plausible explanations.

Six warning signs the conclusion is too strong

The statement claims cause and effect from a comparison with obvious unresolved confounding.
The measured outcome is vague or changes halfway through the exercise.
Several relevant conditions varied together without acknowledgement.
The report presents one unusual observation as if it represented a general rule.
The written conclusion is more confident than the quality and quantity of evidence justify.
No one can explain what result would have challenged the original hypothesis.

These warning signs deserve review. They are not proof that students or researchers acted dishonestly.

A fifteen-minute experiment-design repair

Minutes 1–3: rewrite the question so that the outcome and comparison are explicit.

Minutes 4–6: name the intended independent and dependent variables and list potential confounders.

Minutes 7–9: examine whether the controls, measurement procedures and comparisons suit the question.

Minutes 10–12: identify the strongest alternative explanation that would remain after the planned experiment.

Minutes 13–15: revise one design decision, then write a conclusion template that distinguishes observation from explanation and limitation.

The timing is illustrative. High-stakes scientific work requires specialist review, ethics and safety controls where relevant.

Frequently asked questions

Is a controlled experiment always necessary for science?

No. Observational, modelling, descriptive and other scientific approaches can answer important questions. A controlled experiment is valuable for some causal questions when it is feasible and appropriate.

Does changing one variable at a time guarantee a fair test?

No. It is a helpful classroom starting rule, but uncontrolled conditions, measurement problems and sampling can still affect the comparison. Some valid experimental designs intentionally vary multiple factors.

What happens if the hypothesis is wrong?

That can be a useful result. Assess the quality of the design and data, then revise the explanation and decide what further testing is warranted.

Why do we need controls if we have repeated measurements?

Controls and repetition answer different questions. Controls provide an appropriate comparison or test a method’s behaviour; repetition helps assess the pattern and variation.

The final idea

Scientific experiments fail when a result is mistaken for evidence of a cause without checking the comparison that produced it. Clear questions, appropriate controls, careful measurement, replication and honest limits make experiments useful even when their outcomes surprise us.

Sources and further reading

Educational science guide. Laboratory work must follow the appropriate teacher, institutional and safety procedures.

Other guides in the How Things Fail series

How Career Guidance Stalls | When Qualifications and Opportunities Disconnect
How Scientific Experiments Fail | When Variables and Controls Are Confused — current article
How Public Procurement Breaks | When Tenders Do Not Deliver Public Value
How Maps Mislead | When Scale, Projections and Context Are Lost

For a wider look at setbacks in learning, see How Education Fails and How Strategy Fails. The unchanged Secondary 1 Mathematics Tutor Clementi | Small Groups Tutorials offers an example of identifying the learner’s starting point before choosing support.

How Scientific Experiments Fail | When Variables and Controls Are Confused