Skip to content
Back to search results

Control Group Experimental Design — Controlled Experiment

Introduced by
Ronald A. Fisher; systematised by Donald T. Campbell & Julian C. Stanley
Year
1935 (Fisher); 1963 (Campbell & Stanley codification)
Type
Experimental research design
Family
Experimental design
Font size

What does it do?

Control group experimental design is a fundamental experimental structure in which participants are assigned to at least two groups — a treatment group that receives the intervention and a control group that does not — so that the effect of the intervention can be isolated by comparing outcomes across groups.

Randomisation of assignment strengthens causal inference by balancing known and unknown confounders.

The intuition behind the method

Think of the control group as the baseline against which the effect of your intervention is measured. Without a control group you cannot tell whether any change you observe is caused by the treatment or by time, maturation, history, or other factors. By keeping one group untreated under otherwise identical conditions and then comparing the two groups at the end, you isolate what the treatment specifically contributed. The control group is not an afterthought; it is the logical anchor of causal reasoning in experiments.

History

The logic of a comparison group in experiments was formalised by Ronald A. Fisher at Rothamsted Experimental Station in the 1920s–30s, culminating in his 1935 book The Design of Experiments, which introduced randomisation as the basis of experimental inference. Donald Campbell and Julian Stanley's 1963 monograph Experimental and Quasi-Experimental Designs for Research systematised the typology of experimental designs and catalogued the internal validity threats that a properly randomised control group design addresses. Their framework remains the dominant reference for design validity in social and behavioural sciences.

When to use it

Use a control group experimental design when you need to establish a causal relationship between an intervention and an outcome, not merely an association. It is the standard design for evaluating educational interventions, clinical treatments, behavioural programmes, and product features. It requires that (a) you can manipulate the independent variable, (b) random or at least systematic assignment of participants to groups is feasible, and (c) outcome measurement is possible in both groups under comparable conditions. Do not use this design when randomisation is ethically or practically impossible — a quasi-experimental design or natural experiment is more appropriate in that case. Do not use it when the research question is exploratory, descriptive, or asks about lived experience rather than the effect of a treatment.

Limitations and considerations

  • Requires that random (or otherwise systematic) group assignment is ethically and logistically feasible.
  • Does not automatically prevent attrition bias if dropout rates differ between treatment and control groups.
  • Internal validity is high, but generalising results to different populations or settings requires additional studies.
  • Blinding is often difficult outside clinical contexts, leaving expectancy and placebo effects as residual threats.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Provides the strongest evidence for causal inference among observational and experimental designs.
  • Random assignment controls for both known and unknown confounders simultaneously.
  • Straightforward conceptual logic that maps directly onto most statistical comparison frameworks.
  • Widely accepted standard for intervention evaluation in clinical, educational, and social research.
  • Adaptable: the basic structure can be extended with factorial arrangements, pretest measures, or repeated assessments.

Applications

  • Evaluating the efficacy of a new drug or therapy by comparing treated patients with a placebo-receiving control group in a clinical trial.
  • Testing whether a new pedagogical intervention improves student learning outcomes relative to a standard-instruction control class.
  • Assessing the effect of a workplace wellbeing programme by comparing employee productivity and absenteeism between enrolled and non-enrolled groups.
  • Measuring the impact of a public health campaign on behaviour change by comparing regions exposed to the campaign with unexposed control regions.
  • Validating a new software feature by comparing user task-completion rates in groups with and without the feature enabled.

How it works

How it works

1

The researcher defines at least two groups: one or more treatment groups that will receive the experimental intervention, and a control group that will not receive it (or will receive a placebo or standard-of-care condition). The number and nature of groups are determined by the research question and the factorial structure of the design.

2

Eligible participants are randomly assigned to treatment or control conditions. Random assignment — as opposed to self-selection or convenience allocation — is the mechanism that equates groups on both measured and unmeasured confounders at baseline, making the design a true experiment. Without randomisation the design is quasi-experimental.

3

The treatment group receives the experimental intervention under standardised, controlled conditions. The control group experiences conditions that are as identical as possible except for the absence of the active treatment. Blinding of participants and/or assessors is applied where feasible to prevent expectancy effects from biassing outcomes.

4

Primary and secondary outcomes are measured using the same instruments and procedures in both groups at the same time points. Measurement should be objective or, when subjective, blinded to group assignment. Any pre-specified outcome measures and time points should have been declared before data collection begins.

5

Outcome scores between groups are compared using appropriate statistical tests — typically t-tests or ANOVA for continuous outcomes, chi-square or logistic regression for categorical outcomes. The between-group difference, its confidence interval, and the effect size together convey whether the treatment had a practically and statistically meaningful impact.

Common pitfalls

  • Omitting a true control condition and instead comparing two active treatments — this tests relative efficacy, not absolute effect, and requires explicit acknowledgement.
  • Assigning groups by convenience rather than randomisation and then treating the design as if it were a true experiment, which overstates causal confidence.
  • Insufficient sample size in either group, resulting in underpowered tests that cannot detect realistic effect sizes — always conduct a power analysis before data collection.
  • Differential attrition between groups left unaddressed in analysis — if more participants drop out of one condition, baseline equivalence is compromised and intent-to-treat analysis should be reported.
  • Contamination between groups when participants in the control condition learn about or adopt elements of the treatment, attenuating the observed effect.

Frequently asked questions

What is the difference between a control group and a comparison group?

A control group strictly receives no treatment or a placebo and is formed by random assignment. A comparison group is any group used for contrast but formed by non-random means — for example, a pre-existing class or a historical cohort. The distinction matters for causal inference: random assignment supports causal claims; comparison groups support only associational ones.

Is a waitlist control an acceptable control condition?

Yes, in many applied settings a waitlist control — where the control group receives the intervention after the study ends — is an ethical compromise that preserves the logic of comparison. The limitation is that participants know they are waiting, which can introduce expectancy effects and reduce internal validity relative to a true no-treatment or placebo control.

Do I need a pretest if I have random assignment?

Random assignment theoretically produces equivalent groups at baseline without a pretest, especially in large samples. However, a pretest is still strongly recommended because it allows you to verify baseline equivalence, to use change scores or ANCOVA for more powerful analyses, and to detect any differential attrition over the course of the study.

How large should each group be?

Group size should be determined by a formal power analysis specifying the minimum effect size of practical interest, the desired power level (conventionally 0.80), and the alpha threshold (commonly 0.05). Rules of thumb such as 'at least 30 per group' are starting points only and are inadequate when expected effects are small or outcomes are costly to measure.

Can I use a control group design with a single subject?

Not in the between-subjects sense. Single-subject experimental designs use the individual as their own control across time (e.g., AB or ABAB designs), which is a conceptually distinct approach. A conventional control group design requires multiple participants distributed across conditions.

Related methods

Sources

  1. Campbell, D. T., & Stanley, J. C. (1963). Experimental and Quasi-Experimental Designs for Research. Rand McNally.

    Open publication reference
    Copied.
  2. Fisher, R. A. (1935). The Design of Experiments. Oliver and Boyd.

    Open publication reference
    Copied.