Randomized Controlled Trials: Causation Through Randomization
The randomized controlled trial (RCT) is the gold standard for causal inference because random assignment to conditions (treatment vs. control) distributes all extraneous variables — both measured and unmeasured — approximately equally across groups. This balancing act eliminates confounds, allowing any observed difference in outcomes to be attributed to the treatment. Elements of a well-designed RCT: (1) Random assignment: each participant has an equal probability of being assigned to any condition — use a random number generator, blocked randomization, or stratified randomization. Simple random assignment with small samples can produce imbalanced groups; blocked randomization (randomizing within blocks of 4, 6, or 8) guarantees balanced group sizes. (2) Control group: provides the counterfactual — what would have happened without the treatment. Active control (existing treatment), placebo control (inert treatment), or no-treatment waitlist control. (3) Blinding: participants blinded to condition (single-blind) prevents placebo effects and demand characteristics. Both participants and assessors blinded (double-blind) prevents assessor bias. (4) Pre-specified analysis plan: registered before data collection; prevents post hoc flexibility. (5) Intention-to-treat (ITT) analysis: analyze all participants as randomized, regardless of compliance — preserves the causal validity of randomization. Per-protocol analysis (only compliers) risks re-introducing selection bias. Internal validity (causal validity within the study) vs. external validity (generalizability): RCTs can have high internal validity but low external validity if the sample is unrepresentative of the target population.