Randomised controlled trialRCT
A study that assigns people to treatment or control at random: the design that can show cause rather than association.
A randomised controlled trial assigns participants to an intervention or a comparison group by chance. That single feature is what makes it powerful: randomisation distributes both the confounders you thought of and the ones you did not, so differences in outcome can reasonably be attributed to the intervention.
Several other features strengthen it. Blinding participants and assessors prevents expectation from shaping results. Analysing by intention to treat, meaning by the group people were assigned to rather than what they ended up doing, prevents dropouts from quietly biasing the answer. Pre-registration of the outcome stops the analysis being chosen after the results are seen.
It sits at the top of the evidence ladder for testing whether something works, with meta-analyses of multiple trials above it. That is why our evidence tiers reserve the strongest rating for claims supported at this level.
It is not automatically better than every other design. Trials are usually short, often small, frequently conducted in narrow populations, and they answer the question they asked rather than the one you have. A twelve week trial cannot tell you about twenty years.
Exercise and diet trials carry a specific difficulty: you cannot blind someone to whether they are lifting weights. That is why confidence in exercise findings is high rather than absolute, and why the 2024 network meta-analysis on exercise for depression flagged that only one of 218 included studies met the Cochrane criteria for low risk of bias.
Effect sizes also shrink as trials get bigger and better controlled, which is a useful prior when a small dramatic trial appears.
In practice, ask how many people, for how long, compared with what, and how large the effect was in absolute terms.