What a Randomised Controlled Trial Actually Buys You

22 October 2026

A randomised controlled trial, or RCT, is a study where people are assigned to a treatment or a comparison group purely by chance, like a coin flip. That single move, randomising, is what lets researchers say the treatment caused the difference, rather than just noticed one. But it only answers one question well, whether this treatment works in this population under these conditions, and it stays silent on almost everything else you might want to know.

Why chance does the heavy lifting

Imagine testing a new blood pressure drug by giving it to volunteers who ask for it, and comparing them to people who don't. The volunteers might be healthier, more motivated, or more likely to exercise. Any difference in outcome could be the drug, or it could be all the other ways those two groups differ.

Randomisation breaks that link. When a computer assigns people to groups by chance, every known and unknown trait, age, genetics, diet, income, optimism, gets spread roughly evenly between the groups. You don't need to measure these traits or even know they exist. The coin flip handles it for you.

This is the whole trick. With enough people randomised, the two groups become statistical twins on average. Whatever difference shows up afterwards is very unlikely to be caused by anything except the treatment itself.

What a control group adds

Randomising people into one group isn't enough. You need a second group that doesn't get the treatment, so you have something to compare against. This is the control group, and it might get a placebo, the standard existing treatment, or nothing at all.

The control group answers a question that's easy to forget to ask, what would have happened anyway. People often improve over time regardless of treatment, due to natural recovery, regression to the mean, or simply paying more attention to their health once they join a study. Without a control group, you'd credit the treatment for changes that were going to happen either way.

Good trials also try to stop people knowing which group they're in, a practice called blinding. If patients know they're getting the real drug, they might report feeling better simply because they expect to. If doctors know, they might treat patients differently without meaning to. Blinding both sides, where possible, removes this source of bias too.

What randomisation cannot fix

Randomisation guards against bias within the trial. It does nothing about who was allowed into the trial in the first place. Most RCTs recruit specific kinds of people, often younger, healthier, and more willing to show up for appointments than the general population. A result built on that group doesn't automatically carry over to everyone else.

It also says nothing about why a treatment works, only that it does. A trial can show that a drug lowers blood pressure without revealing the biological mechanism. That's a separate scientific question, answered by different kinds of research entirely.

And randomisation can't rescue a trial from a badly chosen outcome. If researchers measure the wrong thing, say, a lab value that doesn't actually predict health, or they measure it over too short a window to catch side effects, then a perfectly randomised trial can still produce a result that misleads you. Randomisation protects against one specific kind of error. It leaves plenty of others untouched.

The gap between trial and real life

Trials happen under controlled conditions. Patients get reminders to take their medication, regular check-ins, and close monitoring. Real-world use is messier, people forget doses, mix medications, or have health conditions that would have excluded them from the trial altogether.

This is the difference between efficacy, how well something works under ideal trial conditions, and effectiveness, how well it works when it's rolled out into ordinary life. A treatment can look impressive in a trial and underwhelm once it reaches the clinic, not because the trial was flawed, but because real life is a different test.

Trials also tend to run for a fixed, often short, period. Long-term effects, rare side effects, and what happens when a treatment is combined with others people are already taking, these often only show up once a treatment is used widely over years. No single RCT, however well designed, can speak to all of that.

Questions it was never built to answer

An RCT tells you whether a specific intervention produced a specific outcome in a specific group of people over a specific period. It is a narrow instrument, deliberately so, because narrowness is what gives it power. It is not built to tell you whether a policy will work across a whole country, whether an effect will hold in ten years, or whether it matters to someone whose life looks nothing like the trial participants' lives.

People often stretch a single trial's finding to cover questions it never asked. A trial showing a diet improves one blood marker in middle-aged men gets reported as if it proves the diet is good for everyone. The randomisation inside that trial was sound. The leap outside it is where things go wrong.

The honest way to use an RCT is to treat it as one careful answer to one narrow question, and then ask separately whether that question matches the one you actually care about. Often it's close enough to be useful. Sometimes it isn't, and no amount of statistical rigour inside the trial can close that gap.

Common questions

Why can't researchers just compare people who chose a treatment with people who didn't?

Because the choice itself carries information. People who choose a treatment often differ systematically from those who don't, in motivation, wealth, or health habits, and those differences can produce the result rather than the treatment. Randomisation removes this problem by taking the choice out of people's hands.

Is a bigger randomised trial always more trustworthy than a smaller one?

Size helps, because larger trials are less likely to produce a lopsided split between groups by chance and can detect smaller real effects. But size doesn't fix a poorly chosen outcome measure, a biased dropout pattern, or a recruitment pool that doesn't resemble the people you care about.

Can you randomise people for any kind of study?

No. Some questions can't ethically or practically be tested this way, such as the effects of smoking or childhood poverty, because you cannot randomly assign people to harmful conditions. For these, researchers rely on other methods that try to approximate randomisation's logic without actually using it.

What does it mean if a trial is described as double blind?

It means neither the participants nor the researchers interacting with them know who received the real treatment and who received the comparison. This stops expectations, from either side, from quietly shaping the results.

If a trial is well randomised, why do results still sometimes fail to repeat?

Randomisation protects against one type of error but leaves others in play, including chance findings, selective reporting of outcomes, and differences between the trial population and whoever tries to repeat the study. A clean design doesn't guarantee a single run of it reflects a stable, real effect.

Read next

• What a Meta-Analysis Actually Does to the Evidence

  A meta-analysis combines results from many studies into one estimate, but it only clarifies the truth if those studies were asking the same question well.

• Observational Study vs Experiment: Why Diet Advice Flip-Flops

  Most nutrition headlines come from observational studies that track habits, not experiments that test them, which is why the advice keeps changing.

• How P-Hacking Happens Without Anyone Lying

  P-hacking happens when researchers make ordinary, well-intentioned analysis choices that quietly raise the odds of a false positive, with no lie required.