Published on 2026-07-25
Can a study of twelve people really prove anything?
A famous 1998 paper built a sweeping medical claim on just 12 children. The number itself should have been the first warning sign, long before anyone checked the facts.
In February 1998, a gastroenterologist named Andrew Wakefield published a paper in The Lancet describing what he called a new syndrome linking a childhood vaccine to autism. The paper went on to shape vaccination decisions in several countries for more than a decade. Here is the detail that gets lost in the story: the entire claim rested on 12 children.
Twelve. Not 12,000, not 1,200. Twelve case histories, no comparison group, no children who got the vaccine and stayed fine included for contrast. And yet the finding travelled around the world as if it were settled science. Before we even get to what turned out to be wrong with the study, there's a simpler question worth sitting with: could 12 people have shown a real pattern in the first place?
Why small numbers lie so easily
Flip a coin ten times and you can easily get seven heads. Nobody would conclude the coin is rigged: with so few flips, a lopsided run happens all the time by pure chance. Research works the same way. The Catalogue of Bias, a project curated by researchers at the University of Oxford, puts it plainly: the smaller a study, the less likely its findings are to hold up, because a handful of cases can produce a striking pattern for no reason other than chance.
That's not a flaw unique to any one field. It's arithmetic. With a dozen people, one unusual case can swing the whole result. With ten thousand, a single unusual case barely moves the needle. So when a headline leans on 'a study of 12 people' or 'researchers followed 15 volunteers', the honest first reaction isn't excitement, it's a mental note: this is a small enough group that coincidence alone could explain it.
There's a second problem hiding behind the first one, and it's arguably worse. Small studies don't just wobble randomly: they're also far easier to steer, whether the researcher means to or not. Choose which twelve people to include, decide which details to write down and which to leave out, and a study of a dozen can be nudged toward almost any conclusion you already believe. A study of ten thousand people, spread across different cities and different research teams, is much harder to bend that way. Size doesn't just protect against chance. It protects against thumbs quietly pressing on the scale.
What actually happened with those twelve
The Wakefield case turned out to be worse than a small sample stretched too far. A later investigation, reported by the public health outlet CIDRAP, found the children had been recruited through vaccine campaigners already preparing a lawsuit, not selected as a neutral group of patients. Medical records were altered: symptoms that families said began months after vaccination were reported as starting within days. Only one of the children clearly had the type of autism the paper described, though several were presented as textbook cases. The Lancet retracted the paper in 2010, twelve years after publishing it.
Here's the part worth keeping. The retraction wasn't only about fraud, though there was plenty of that. It was about a study built to look bigger than it was: hand-picked cases dressed up as an unbiased sample, run through a design too small to survive a single skewed case, let alone several. Twelve people can be an honest starting point for research. Twelve people chosen because they already fit the story is a different animal entirely.
A dozen cases can be the start of a real question. They become an answer only once thousands of other cases, checked independently, say the same thing.
Small isn't the crime, the label is
None of this means tiny studies are worthless. Early research on a new idea often starts with a handful of people, precisely because nobody wants to commit real resources before there's a hint worth chasing. A pilot study that says 'this deserves a bigger trial' is doing exactly its job, and plenty of important discoveries began exactly that way, as a small, honest hunch that later research confirmed at scale. The trouble starts when a small, hint-sized result gets marketed as a finished discovery, especially once a press release or a headline strips away the caveats the researchers themselves included.
So the next time a claim leans on a study, look for the number before you look at the conclusion. A dozen, a few dozen, even a few hundred: fine for a first look, thin for a verdict. What should follow a promising small study is more people, more places, a design where nobody handpicked the cases in advance. If that follow-up never shows up, or if it comes back with the opposite result, the original twelve were telling you less than the headline promised.
Sources
- Report says 1998 vaccine-autism study was fraud (CIDRAP, University of Minnesota) (cidrap.umn.edu)
- Wrong sample size bias (Catalogue of Bias) (catalogofbias.org)