Randomised controlled trials are the reference standard for showing cause and effect.
They solve confounding by design rather than by adjustment after the fact.
Because chance decides who is treated, nothing about the participants can influence who gets what.
A trial defines its population, treatment, control, outcome and sample size before it begins.
The control can be no treatment, a placebo or the current standard treatment.
Single blinding hides the assignment from participants, and double blinding hides it from the researchers too.
Blinding matters because expectations change behaviour and how outcomes are judged.
The analysis compares the groups' outcomes, usually with a hypothesis test and a confidence interval.
Trials can be too small to detect real effects, so power calculations are part of their design.
They often recruit unusual participants, such as healthier or more motivated volunteers, which limits how far results apply.
Some questions cannot be randomised for ethical reasons, such as whether smoking causes cancer.
Others are too slow or costly, such as the effect of a policy over decades.
A/B tests are randomised controlled trials run on products, with users in place of patients.
In medicine, trials are registered in advance, so that unfavourable results cannot quietly disappear.
The common misconception is that a trial's result applies to everyone, when it applies most clearly to people like those who took part.