The problem randomisation solves

Suppose you notice that people who take a certain supplement are healthier. Tempting conclusion: the supplement works. But the people who choose to take it are different from the start — they tend to exercise more, eat better, earn more, smoke less, and see doctors sooner. Any of those could explain the health difference. This is confounding, and it wrecks almost every simple comparison. (We give it a whole lesson next.)

Randomisation is the fix, and it's almost magical. Take one group of people and split them into two by pure chance — a coin flip. Now the two groups are, on average, identical in everything: age, diet, income, motivation, the factors you measured and the ones you never thought of. Give one group the supplement and the other a placebo, wait, and compare. Because the groups started the same, any difference at the end must have been caused by the one thing you changed: the pill.

Flow of a randomised controlled trial from participants to measured difference Volunteers mixed, real people RANDOM coin flip Group A → supplement blinded Group B → placebo blinded difference = the pill's effect
Randomise, blind, compare. The coin flip makes the two groups alike in everything except the pill, so the gap between them at the end is caused by the pill — not by who chose to take it. Everything else in trial design exists to protect this one logic.

Blinding and the placebo

Randomisation handles who-differs-from-whom. Two more pieces handle the mind:

  • Placebo control. People improve just from believing they're being treated — a real, measurable effect, especially for anything subjective like pain, mood, sleep or energy. So the comparison group gets an identical-looking dummy pill. The question is never "did people on the supplement improve?" but "did they improve more than the placebo group?"
  • Blinding. If participants know which pill they got, expectation contaminates the result; if the researchers know, they unconsciously nudge measurements. Double-blind — neither side knows until the end — closes both leaks. "Double-blind, placebo-controlled, randomised" is the phrase that means a trial took its own logic seriously.

Why a "clinical trial" can still be near-worthless

"Clinically tested" on a jar might point to a real RCT — or to something wearing the costume. The common failure modes:

  • Too small. 12 people can't reliably detect anything; noise swamps signal. Look for sample size in the dozens-to-hundreds at least.
  • No placebo / not blinded ("open-label"). Then you're measuring expectation as much as the supplement.
  • Too short, or the wrong outcome. Four weeks for something that matters over years, or a "surrogate" marker (a blood number) standing in for how you actually feel or function.
  • Funded and run by the seller, with the analysis chosen after seeing the data. Not automatically fraud, but a heavy thumb on the scale.
Worked example Two "clinical trials", same claim
CLAIM: “Clinically proven to boost energy” Study A (on the brand's site) 20 people · everyone took the product · no placebo · rated their own energy · 4 weeks · funded by the brand Study B (what you'd want) 240 people · randomly assigned product or identical placebo · double-blind · validated fatigue scale · 12 weeks · independently funded

Study A can't separate the supplement from the placebo effect, from regression to the mean, or from wanting to please the brand that's paying. It will almost always show "improvement" — that's the point of designing it that way.

Study B is the real question. If the product beats placebo here, that's worth something. When a label says "clinically proven", the entire value of the claim lives in which of these two it actually was — so that's the thing to ask.

You'll rarely read the trial yourself, and that's fine. Just hold the shape in mind — randomised, blinded, placebo-controlled, big enough, long enough, independently funded — and you can judge the strength of a "clinically proven" claim without ever opening the paper.

What to remember

Randomisation is what turns “linked to” into “caused by”.

  • The coin flip makes the two groups identical in everything except the pill — so the difference at the end is the pill's effect, not the kind of person who chose it.
  • Placebo + double-blind strip out expectation. The real question is always "better than placebo?", never "did they improve?"
  • A trial can wear the costume and skip the logic. Too small, open-label, too short, surrogate outcome, or seller-funded — each quietly guts a "clinically proven" claim.
Try it · ~1 min

Interrogate a “clinically proven” claim.

  1. Find a product that cites a study. See if you can learn how many people and whether there was a placebo.
  2. Check for the words “randomised” and “double-blind”. Their absence is the tell.
  3. Ask who paid and how long it ran. If you can't find any of this, treat the claim as marketing, not proof.

Next in this path

Lesson 04Observational studies — the confounding RCTs are built to defeat ResearchCreatine loading — what strong RCT evidence actually reads like

References

1
Schulz KF, Altman DG, Moher D; CONSORT Group. CONSORT 2010 Statement: updated guidelines for reporting parallel group randomised trials. BMJ. 2010;340:c332. doi:10.1136/bmj.c332
2
Hróbjartsson A, Gøtzsche PC. Placebo interventions for all clinical conditions. Cochrane Database Syst Rev. 2010;(1):CD003974. doi:10.1002/14651858.CD003974.pub3

Progress saves automatically in this browser · skip ahead anytime · no streak.