Randomized Controlled Trial

A randomized controlled trial (RCT) is a study design in which participants are randomly assigned to receive an intervention (the treatment group) or not (the control group), and outcomes are subsequently compared. Randomization is the methodological key: by distributing both known and unknown confounders equally across groups by chance, it eliminates confounding bias — the primary threat to causal inference in observational studies.

The RCT is the gold standard of evidence-based medicine because it is the study design that most directly implements the logical structure of a causal test: hold everything constant except the cause of interest, vary the cause, and observe the effect. Observational studies can approximate this ideal through statistical adjustment, but the adjustment is only as good as the researcher's knowledge of which confounders exist — and unknown confounders cannot be adjusted for. Randomization sidesteps the problem entirely.

RCTs have important limits. They cannot be used where randomization is unethical (we cannot randomly assign people to smoke). They are expensive, time-limited, and often conducted on populations that do not represent real-world patients. Their results may not generalize to different populations, doses, or contexts. And they answer only the question they were designed to ask: average causal effects in the study population, not individual effects or mechanisms.

The deeper philosophical point: the RCT is not gold because it is elegant, but because it minimizes the assumptions required to make a causal claim. The entire apparatus of observational causal inference — causal graphs, instrumental variables, regression discontinuity — exists to approximate the RCT ideal when randomization is impossible. Understanding the RCT illuminates what those approximations are approximating.