A controlled experiment is an experiment designed so that the effect of a specified independent variable or treatment can be evaluated by comparing conditions while other relevant factors are controlled, standardised or balanced as far as practical. A controlled experiment commonly includes an experimental group receiving a treatment and a control group that does not receive that treatment or receives a comparison condition.
The purpose is to isolate the effect of the factor under investigation and reduce alternative explanations for observed differences. Control does not mean that every variable in the entire environment is perfectly constant, which is often impossible. Instead, researchers identify important sources of variation and manage them through standardisation, randomisation, blocking, matching or statistical adjustment. Controlled experiments are especially useful for causal questions because deliberate manipulation combined with comparison can reveal whether changing one factor produces a measurable outcome.
The control group provides the baseline against which the treatment condition is interpreted. A negative control may receive no active treatment and establishes what happens without the experimental intervention. A positive control receives a condition known to produce an expected response and can demonstrate that the experimental system and measurement procedure are capable of detecting an effect. A placebo control is used in some human studies to separate treatment effects from expectations and other contextual effects.
Sham controls can mimic a procedure without delivering its active component. The appropriate control depends on the research question and experimental system. A control group is not necessarily identical to the experimental group in every respect because the treatment itself is intentionally different; the objective is to make treatment assignment the principal systematic difference relevant to the causal comparison.
Randomisation is a major tool in controlled experiments because it helps distribute known and unknown confounding factors across treatment groups. If experimental units are randomly assigned, differences in characteristics between groups are less likely to be systematically associated with treatment. Randomisation does not guarantee perfectly identical groups, especially with small samples, but it provides a formal basis for statistical inference under appropriate assumptions.
Blocking can be used when researchers know that a variable, such as age, location or batch, strongly affects the outcome. Units are grouped into similar blocks and treatments are then randomised within those blocks. Matching can serve a related purpose in some designs. These procedures improve comparability and reduce the likelihood that observed treatment effects are actually consequences of pre existing differences.
Blinding can strengthen controlled experiments when expectations could influence behaviour, measurement or interpretation. In a single blind design, one relevant party, often participants, does not know the treatment assignment. In a double blind design, both participants and relevant researchers or assessors are unaware of assignments until specified procedures permit unblinding. Blinding is particularly important when outcomes involve subjective assessment, symptoms, behaviour or researcher judgement. It is less relevant or impossible in experiments where the physical treatment is obvious, such as comparing visibly different environmental conditions. When blinding cannot be implemented, objective measurements, automated instruments, prespecified analysis procedures and independent assessment can reduce bias.
Controlled experiments can take many forms. A simple parallel group experiment assigns separate experimental units to different conditions. A crossover design exposes the same participant or experimental unit to multiple treatments at different times, potentially reducing between subject variability but requiring attention to order effects and carryover. Factorial experiments investigate multiple factors simultaneously and can determine whether factors interact. Repeated measures designs collect several observations from the same units, improving efficiency but requiring appropriate treatment of within unit dependence. Laboratory controlled experiments can provide high environmental control, while field controlled experiments deliberately manipulate conditions in realistic environments. Each design involves trade offs among control, realism, efficiency, ethical feasibility and generalisability.
Control variables should be distinguished from the control group. A control variable is a condition that researchers attempt to keep constant or account for, such as light intensity, temperature or experimental duration. A control group is a comparison group representing a specified baseline or alternative condition. Confounding occurs when another factor varies systematically with the treatment and could explain the observed outcome.
For example, if one treatment group is tested in the morning and another in the afternoon while time of day affects the outcome, treatment and time are confounded. A controlled experiment seeks to prevent such confounding through design rather than attempting to eliminate every source of natural variation. Statistical analysis can account for some variation, but strong experimental design generally provides more reliable causal identification than statistical adjustment alone.
The concept of control has limitations. Perfect control is rarely possible in complex biological, environmental or social systems. Excessive laboratory control can create conditions that differ from real world settings, limiting external validity. Participants may change behaviour because they know they are being studied, and experimental animals may respond differently under laboratory conditions from natural environments. Some variables cannot be manipulated for ethical or practical reasons. Researchers may therefore combine controlled experiments with observational research to determine whether experimentally demonstrated mechanisms operate under real world conditions. Replication across laboratories, populations, environments and time periods can help determine how general a controlled experimental result is.
A controlled experiment provides strong evidence for causal relationships when treatment assignment is appropriate, groups are comparable, measurements are reliable, relevant confounders are addressed and statistical analysis matches the design. It does not automatically establish universal causation because the result may depend on the population, treatment intensity, environmental conditions, duration and measurement method. A controlled experiment should therefore be interpreted within its specified scope.
Its central advantage is that deliberate manipulation and comparison can make causal inference substantially stronger than simple observation when the design is appropriate. Its central requirement is not merely the existence of a group called the control group, but a carefully structured comparison in which alternative explanations for differences in outcomes have been systematically reduced.
