An experimental study design in which eligible units (individuals, groups, or clusters) are assigned to one or more intervention (treatment) arms or to a control arm by a known random mechanism so that, under the maintained assumptions (randomization executed as assigned, no interference between units, well-defined treatments and outcomes), differences in observed outcomes can be interpreted as causal effects of the intervention for the study population or randomized sample.

Principle

Random assignment ensures that, in expectation, pre-treatment characteristics (observed and unobserved) are balanced across arms, so differences in post-treatment outcomes can be attributed to the assigned intervention conditional on preserved assumptions (compliance, no interference, correct implementation).

Demonstration

Illustrative scenario: Among applicants who meet eligibility, administrators randomly assign half to receive a job-skills training and half to a control condition. Recognition: randomization is implemented and documented. Action: compare average employment outcomes after follow-up using intention-to-treat and, if necessary, complier analyses. Consequence: differences in employment rates near follow-up time provide an unbiased estimate of the program's average effect for the randomized sample, subject to attrition and compliance considerations.

Misapplication

Interpreting an RCT estimate as the population average without considering sample selection, or ignoring bias introduced by differential attrition, noncompliance, or protocol deviations; treating a failed randomization check as proof of no causal effect rather than an indication of compromised internal validity.

Consequence

When assumptions hold and implementation is faithful, an RCT provides high internal validity for causal inference within the study sample; practical consequences include requirements for careful ethical review, sample size planning, pre-specification, and attention to compliance and attrition which affect estimands and inference.

Reversal

If randomization is compromised (systematic assignment errors, differential attrition, or post-randomization contamination between arms), if interference between units (violations of SUTVA) is present, or when noncompliance is widespread without suitable analysis, the causal attribution fails or must be reinterpreted (e.g., as local or complier effects, not average effects).

Boundary

Clearly within: prospectively designed studies with documented random assignment of units to arms. Boundary case: cluster-randomized designs or stratified randomization require different analytic adjustments. Clearly outside: observational studies lacking deliberate random assignment or relying solely on as-if randomness without verification are not RCTs.

Semantic Tension

Internal validity (causal identification within the study) versus external validity (generalizability to other populations, settings, or times), and causal inference versus ethical constraints that may limit feasible assignments.

Synthesis

An RCT is primarily an internally identifying device: it trades external generality and some practical feasibility for a design that, when correctly implemented and analyzed, isolates causal effects for the randomized sample; credible application requires attention to implementation details (randomization fidelity, attrition, noncompliance, interference) and careful translation of estimands to policy questions.